Guidelines for AI and Shared Prosperity - Tools for improving AI’s impact on jobs

PARTNERSHIP ON AI
Guidelines for AI and Shared Prosperity
1
Guidelines
for AI and
Shared
Prosperity
Tools for improving
AI’s impact on jobs
V.1
JUNE 2023

Contents
Quick Reference 3
Signals of Opportunity and Risk 3
Responsible Practices for Organizations 4
Get Involved 5
Executive Summary 6
Learn About the Guidelines 7
The Need for the Guidelines 7
Origin of the Guidelines 8
Design of the Guidelines 9
Key Principles for Using the Guidelines 11
Apply the Job Impact Assessment Tool 14
Instructions for Performing a Job Impact Assessment 14
Signals of Opportunity for Shared Prosperity 15
Signals of Risk to Shared Prosperity 21
Follow Our Stakeholder-Specific Recommendations 28
Responsible Practices for AI-Creating Organizations (RPC) 28
Responsible Practices for AI-Using Organizations (RPU) 34
Suggested Uses for Policymakers 41
Suggested Uses for Labor Organizations and Workers 42
Acknowledgements 43
AI and Shared Prosperity Initiative’s Steering Committee 44
Endorsements 45
Sources 48
Though this document reflects the inputs of many PAI Partners, it should not be read as
representing the views of any particular organization or individual within the AI and Shared
Prosperity Initiative’s Steering Committee or any specific PAI Partner.

PARTNERSHIP ON AI
3
Signals of Opportunity for Shared Prosperity
An opportunity signal (OS) is present if an AI system may:
OS1. Generate significant, widely distributed benefits
OS2. Boost worker productivity
Caveat 1: Productivity boosts can deepen inequality
Caveat 2: Productivity boosts can displace workers
Caveat 3: Productivity boosts can significantly hamper
job quality
OS3. Create new paid tasks for workers
Caveat 1: Someone’s unpaid tasks can be someone else’s
full-time job
Caveat 2: New tasks often go unacknowledged and unpaid
OS4. Support an egalitarian labor market
OS5. Be appropriate for lower-income geographies
OS6. Broaden access to the labor market
OS7. Boost revenue share of workers and society
OS8. Respond to needs expressed by impacted workers
OS9. Be co-developed with impacted workers
OS10. Improve job quality or satisfaction
Caveat 1: Systems can improve one aspect of job quality
while harming another
Caveat 2: AI systems are sometimes deployed to redress
job quality harms created by other AI systems
Signals of Risk to Shared Prosperity
A risk signal (RS) is present if an AI system may:
RS1. Eliminate a given job’s core tasks
RS2. Reallocate tasks to lower-paid or more precarious jobs
RS3. Reallocate tasks to higher- or lower-skilled jobs
RS4. Move jobs away from geographies with few
opportunities
RS5. Increase market concentration and barriers to entry
RS6. Rely on poorly treated or compensated outsourced labor
RS7. Use training data collected without consent or
compensation
RS8. Predict the lowest wages a worker will accept
RS9. Accelerate task completions without other changes
RS10. Reduce schedule predictability
RS11. Reduce workers’ break time
RS12. Increase overall difficulty of tasks
RS13. Enable detailed monitoring of workers
RS14. Reduce worker autonomy
RS15. Reduce mentorship or apprenticeship opportunities
RS16. Reduce worker satisfaction
RS17. Influence employment and pay decisions
RS18. Operate in discriminatory ways
Quick Reference
Signals of Opportunity and Risk

PARTNERSHIP ON AI
4
Responsible Practices for
AI-Creating Organizations (RPC)
RPC1. Make a public commitment to identify, disclose, and
mitigate the risks of severe labor market impacts
presented by AI systems you develop
RPC2. In collaboration with affected workers, perform Job
Impact Assessments early and often throughout the AI
system lifecycle
RPC3. In collaboration with affected workers, develop
mitigation strategies for identified risks
RPC4. Source data enrichment labor responsibly
RPC5. Create and use robust and substantive mechanisms
for worker participation in AI system origination,
design, and development
RPC6. Build AI systems that align with worker needs and
preferences
RPC7. Build AI systems that complement workers (especially
those in lower-wage jobs), not ones that act as their
substitutes
RPC8. Ensure workplace AI systems are not discriminatory
RPC9. Provide meaningful, comprehensible explanations
of the AI system’s function and operation to workers
using or affected by it
RPC10. Ensure transparency about what worker data is
collected, how and why it will be used, and enable
opt-out functionality
RPC11. Embed human recourse into decisions or
recommendations you offer
RPC12. Apply additional mitigation strategies to sales and
use in environments with low worker protection and
decision-making power
RPC13. Red team AI systems for potential misuse or abuse
RPC14. Ensure AI systems do not preclude the sharing of
productivity gains with workers
RPC15. Request deployers to commit to following PAI’s Shared
Prosperity Guidelines or similar recommendations
AI-Using Organizations (RPU)
RPU1. Make a public commitment to identify, disclose, and
mitigate the risks of severe labor market impacts
presented by AI systems you use
RPU2. Commit to neutrality towards worker organizing and
unionization
RPU3. In collaboration with affected communities, perform
Job Impact Assessments early and often throughout AI
system implementation and use
RPU4. In collaboration with affected communities, develop
mitigation strategies for identified risks
RPU5. Create and use robust and substantive mechanisms
for worker agency in identifying needs, selecting AI
vendors and systems, and implementing them in the
workplace
RPU6. Ensure AI systems are used in environments with high
levels of worker protections and decision-making
power
RPU7. Source data enrichment labor responsibly
RPU8. Ensure workplace AI systems are not discriminatory
RPU9. Procure AI systems that align with worker needs and
preferences
RPU10. Staff and train sufficient internal or contracted
expertise to properly vet AI systems and ensure
responsible implementation
RPU11. Prefer vendors who commit to following PAI’s Shared
Prosperity Guidelines or similar recommendations
RPU12. Ensure transparency about what worker data is
collected, how it will be used, and why, and enable
workers to opt out
RPU13. Provide meaningful, comprehensible explanations
of the AI system’s function and operation to workers
overseeing it, using it, or affected by it
RPU14. Establish human recourse into decisions or
recommendations offered, including the creation
of transparent, human-decided grievance redress
mechanisms
RPU15. Red team AI systems for potential misuse or abuse
RPU16. Recognize extra work created by AI system use and
ensure work is acknowledged and compensated
RPU17. Ensure mechanisms are in place to share productivity
gains with workers
Responsible Practices for Organizations

PARTNERSHIP ON AI
5
Get Involved
The Partnership on AI seeks to engage all interested stakeholders to
refine, test, and drive the adoption and evolution of all parts of the
Shared Prosperity Guidelines, including the Job Impact Assessment
Tool, the Responsible Practices, and Suggested Uses. We also seek to
curate a library of learnings, use cases and examples, as well as partner
with stakeholders to co-create companion resources to help make the
Guidelines easier to use for their communities.
We will pursue these goals by means of stakeholder outreach,
dedicated workshops, and limited implementation collaborations.
If you’re interested in engaging with us on this work or want to
publicly endorse the Guidelines, please get in touch.

PARTNERSHIP ON AI
6
Executive Summary
Our economic future is too important to leave to chance.
AI has the potential to radically disrupt people’s economic lives in both positive and
negative ways. It remains to be determined which of these we’ll see more of. In the best
scenario, AI could widely enrich humanity, equitably equipping people with the time,
resources, and tools to pursue the goals that matter most to them.
Our current moment serves as a profound opportunity — one that we will miss if we
don’t act now. To achieve a better future with AI, we must put in the work today. Many
societal factors outside the direct control of AI-developing and AI-using organizations
will play a role in determining this outcome. However, much still depends on the choices
those organizations make, as well as on the actions taken by labor organizations and
policymakers.
You can help guide AI’s impact on jobs
AI-creating companies, AI-using organizations, policymakers, labor organizations, and
workers can all help steer AI so its economic benefits are shared by all. Using Partnership
on AI’s (PAI) Guidelines for AI & Shared Prosperity, these stakeholders can guide AI
development and use towards better outcomes for workers and labor markets.
Included in the Guidelines are:
• a high-level Job Impact Assessment Tool for analyzing an AI system’s positive and
negative impact on shared prosperity
• a collection of Stakeholder-Specific Recommendations to help minimize the risks and
maximize the opportunities to advance shared prosperity with AI
How to use the Guidelines
The Shared Prosperity Guidelines can be used by following a guided, three-step process.
Step 1
Learn about
the Guidelines
Step 2
Apply the Job Impact
Assessment Tool
Step 3
Follow our Stakeholder-
Specific Recommendations
This is the first version
of the Guidelines,
developed under
close guidance from
a multidisciplinary AI
and Shared Prosperity
Initiative’s Steering
Committee and with
direct engagement
of frontline workers
from around the
world experiencing
the introduction of AI
in their workplaces.
The Guidelines
are intended to be
updated as the AI
technology evolves
and presents
new risks and
opportunities, as well
as in response to
stakeholder feedback
and suggestions
generated through
workshops, testing,
and implementation.

PARTNERSHIP ON AI
7
The Need for the Guidelines
Action is needed to guide AI’s impact on jobs
Artificial intelligence is poised to substantially affect the labor market and the nature of
work around the globe.
• Some job categories will shrink or disappear entirely and new types of occupations will
arise in their place
• Wages will be affected, with AI changing the demand for various skills and the access
workers have to jobs
• The tasks workers perform at their jobs will change, with some of their previous work
automated and other tasks assisted by new technologies
• Job satisfaction and job quality will shift. Benefits will accrue to the workers with the
highest control over how AI shows up in their jobs. Harms will occur for workers with
minimal agency over workplace AI deployments
The magnitude and distribution of these effects is not fixed or pre-ordained.A
Today, we
have a profound opportunity to ensure that AI’s effects on the labor market and the future
of work contribute to broadly shared prosperity.
In the best scenario, humanity could use AI to unlock opportunities to mitigate climate
change, make medical treatments more affordable and effective, and usher in a new era
of improved living standards and prosperity around the world. This outcome, however,
will not be realized by default.1
It requires a concerted effort to bring it about. AI use poses
numerous large-scale economic risks that are likely to materialize given our current path,
including:
• Consolidating wealth in the hands of a select few companies and countries
• Reducing wages and undermining worker agency as larger numbers of workers
compete for deskilled, lower-wage jobs
• Allocating the most fulfilling tasks in some jobs to algorithms, leaving humans with
the remaining drudgery
• Highly disruptive spikes in unemployment or underemploymentB
as workers start at
the bottom rung in new fields, even if permanent mass unemployment does not arise
in the medium term
A Example explanations
of why technological
change is the result
of market-shaping
policies (and not some
“natural” or predeter-
mined trajectory) can
be found in:
Redesigning AI: Work,
democracy, and justice in
the age of automation
Steering technological
progress
B We use the definition
of underemployment
from Merriam-Webster
dictionary: “the condition
in which people in a labor
force are employed at
less than full-time or
regular jobs or at jobs
inadequate with respect
to their training or
economic needs.”
STEP 1
Learn About the Guidelines
Artificial
intelligence
is poised to
substantially
affect the
labor market
and the nature
of work around
the globe.

PARTNERSHIP ON AI
8
The Guidelines are tools for creating a better future
Partnership on AI’s (PAI) Shared Prosperity Guidelines are intended to equip interested
stakeholders with the conceptual tools they need to steer AI in service of shared prosperity.
All stakeholders looking to ground their decisions, agendas, and interactions with each
other in a systematic understanding of labor market opportunities and risks presented by
AI systems can use these tools. This includes:
AI-creating
organizations
AI-using
organizations
Policymakers Labor organizations
and workers
Origin of the Guidelines
This work comes from years of applied research and
multidisciplinary input
A key output of PAI’s AI and Shared Prosperity Initiative, PAI’s Shared Prosperity Guidelines
were developed under the close guidance of a multidisciplinary Steering Committee and
draw on insights gained during two years of applied research work. This work included
economic modeling of AI’s impacts on labor demand,23
engaging frontline workers around
the world to understand AI’s impact on job quality,4
mapping the levers for governing AI’s
economic trajectory,5
as well as a major workstream on creating and testing practitioner
resources for the responsible sourcing of data enrichment labor. The plan for this multi-
stakeholder applied research work was shared with the public in “Redesigning AI for Shared
Prosperity: an Agenda” published by Partnership on AI in 2021, following eight months of
Steering Committee deliberations.
Though this document reflects the inputs of many PAI Partners, it should not be read as
representing the views of any particular organization or individual within the AI and Shared
Prosperity Initiative’s Steering Committee or any specific PAI Partner.

PARTNERSHIP ON AI
9
Design of the Guidelines
We offer two tools for guiding AI’s impact on jobs
A high-level Job Impact Assessment Tool with:
• Signals of Opportunity indicating an AI system may support shared
prosperity
• Signals of Risk indicating an AI system may harm shared prosperity
A collection of Stakeholder-Specific Recommendations: Responsible Practices
and Suggested Uses for stakeholders able to help minimize the risks and
maximize the opportunities to advance shared prosperity with AI.
In particular, they are written for:
• AI-creating organizations
• AI-using organizations
• Policymakers
• Labor organizations and workers
These tools can guide choices about any AI system
PAI’s Shared Prosperity Guidelines are designed to apply to all AI systems, regardless of:
• Industry (including manufacturing, retail/services, office work, and warehousing and
logistics)
• AI technology (including generative AI, autonomous robotics, etc.)
• Use case (including decision-making or assistance, task completion, training, and
supervision)
As a whole, the Guidelines are general purpose and applicable across all existing AI tech
nologies and uses, though some sections may only apply to specific technologies or uses.
To apply these guidelines, stakeholders should:
• For an AI system of interest, perform the analysis suggested in the Job Impact
Assessment section, identifying which signals of opportunity and risk to shared
prosperity are present.
• Use the results of the Job Impact Assessment to inform your plans, choices,
and actions related to the AI system in question, following our Stakeholder-
Specific Recommendations. For AI-creating and AI-using organizations, these
recommendations are Responsible Practices. For policymakers, unions, workers, and
their advocates, these recommendations are Suggested Uses.
We look forward to testing the Guidelines and refining the use scenarios together with
interested stakeholders. If you have suggestions or would like to contribute to this work,
please get in touch.
PAI’s Shared
Prosperity
Guidelines
are designed
to apply to all
AI systems,
regardless of
industry, AI
technology, or
use case.

PARTNERSHIP ON AI
10
Our approach focuses on AI’s impact on labor demand
In these Guidelines, we consider an AI system to be serving to advance the
prosperity of a given group if it boosts the demand for labor of that group —
since selling labor remains the primary source of income for the majority of
people in the world.
We recognize that some communities advocate to advance shared prosperity
in the age of AI through benefits redistribution mechanisms such as universal
basic income. While a global benefits redistribution mechanism might be an
important part of the solution (especially in the longer term) and we welcome
research efforts and public debate on this topic, we left it outside of the scope
of the current version of the Guidelines.
Instead, the Guidelines focus on governing the impact of AI on labor demand.
We believe this approach will be extremely necessary at least in the short to
medium term, enabling communities to have effective levers of influence over
the pace, depth, and distribution of AI impacts on labor demand.
AI’s impacts on labor demand can manifest themselves as:
• Changes in availability of jobs for certain skill, demographic, or
geographic groupsC
• Changes in the quality of jobs affecting workers’ well-beingD
In line with PAI’s framework for promoting workforce well-being in the
AI-integrated workplace and other leading resources on high-quality jobs,678
we
recognize multiple dimensions of job quality or workers’ well-being, namely:
• Human rights
• Financial well-being
• Physical well-being
• Emotional well-being
• Intellectual well-being
• Sense of meaning, community, and purpose.
Thus, for the purposes of these Guidelines, we define AI’s impact on shared
prosperity as the impact of AI use on availability and quality of formal
sector jobs across skill, demographic, or geographic groups.E
In turn, the overall impact of AI on the availability and quality of jobs can be
anticipated as a sum total of changes in the primary factors AI use is known to
affect.91011
Those factors are:
We define AI’s impact
on shared prosperity
as the impact of AI
use on availability
and quality of formal
sector jobs across
skill, demographic, or
geographic groups.
E The share of informal sector
employment remains high in many
low- and middle-income countries. The
emphasis on formal sector jobs here
should not be interpreted as treating
the informal sector as out of scope of
the concern of PAI’s Shared Prosperity
Guidelines. The opposite is the case:
If the introduction of an AI system in
the economy results in a reduction of
availability of formal sector jobs, that
reduction cannot be considered to be
compensated by growth in availability
of jobs in the informal sector.
C Groups’ boundaries can be defined
geographically, demographically, by skill
type, or another parameter of interest.
D In other words, AI’s impact on labor
demand can affect both incumbent
workers as well as people interested in
looking for work in the present or future.

PARTNERSHIP ON AI
11
• Relative productivity of workers
(versus machines or workers in other
skill groups)
• Labor’s share of organization
revenueF
• Task composition of jobs
• Skill requirements of jobs
• Geographic distribution of the
demand for laborG
• Geographic distribution of the supply
of labor
• Market concentration
• Job stability
• Stress rates
• Injury rates
• Schedule predictability
• Break time
• Job intensity
• Freedom to organize
• Privacy
• Fair and equitable treatment
• Social relationships
• Job autonomy
• Challenge level of tasks
• Satisfaction or pride in one’s work
• Ability to develop skills needed for
one’s career
• Human involvement or recourse
for managerial decisions (such
as performance evaluation and
promotion)
• Human involvement or recourse
in employment decisions (such as
hiring and termination)
Anticipated effects on the above primary factors are the main focus of the risks and
opportunities analysis tool provided in the Guidelines. Another important focus is the
distribution of those effects. An AI system may bring benefits to one set of users and
harms to another. Take, for example, an AI system used by managers to set and monitor
performance targets for their reports. This system could potentially increase pride in
one’s work for managers and raise rates of injury and stress for their direct reports.
When this dynamic prompts conflicting interests, we suggest higher consideration for
the more vulnerable group with the least decision-making power in the situation as these
groups often bear the brunt of technological harms.12
By a similar logic, where we call for
worker agency and participation, we suggest undertaking particular effort to include the
workers most affected and/or with the least decision authority (for example, the frontline
workers, not just their supervisors).
Key Principles for Using the Guidelines
These application principles apply independently of who is using the Guidelines and in what
specific scenario they are doing so.
Engage affected workers
Make sure to engage worker communities that stand to be affected by the introduction of
an AI system in the Job Impact Assessment, as well as in the development of risk mitigation
strategies. This includes, but is not limited to, engaging and affording agency to workers who
will be affected by the AI system and their representatives.H
Bringing in multi-disciplinary
experts will help understand the full spectrum and severity of the potential impact.
G Geographic distribu-
tions of labor demand
and supply do not
necessarily match for a
variety of reasons, the
most prominent of which
are overly restrictive
policies around labor
migration. Immigration
barriers present in many
countries with rapidly
aging populations create
artificial scarcity of
labor in those countries,
massively inflating the
incentives to invest in
labor-saving technol-
ogies. For more details,
read this article.
F Labor’s share of revenue
is a share of revenue
spent on workers’ wages
and benefits.
H It is frequently the
case that workers who
stand to be affected
by the introduction of
an AI system include
not only workers
directly employed by
the company intro-
ducing AI in its own
operations, but a
wider set of current or
potential labor market
participants. Hence it
is important that not
only incumbent workers
are given the agency
to participate in job
impact assessment and
risk mitigation strategy
development.

PARTNERSHIP ON AI
12
Workers may work with AI systems or have their work affected by them. In cases where one
group of workers uses an AI system (for instance, uses an AI performance evaluation tool
to assess their direct reports) and another group is affected by that AI system’s use (in this
example, the direct reports), we suggest giving highest consideration to affected workers
and/or the workers with the least decision-making power in the situation (in this example,
the direct reports rather than the supervisors).
Seeking shared prosperity doesn’t mean opposing profits
Some of the signals of risk to shared prosperity described in the Guidelines are actively
sought by companies as profit-making opportunities. The Guidelines do not suggest that
companies should stop seeking profits, just that they should do so responsibly.
Profit-generating activities do not necessarily have to harm workers and communities,
but some of them do. The presence of signals of risk indicate that an AI system being
assessed, while possibly capable of generating profit for a narrow set of beneficiaries, is
likely to do that at the expense of shared prosperity, and thus might be undesirable from the
societal benefit perspective. We encourage companies to follow the Guidelines, developing
and using AI in ways that generate profit while also advancing shared prosperity.
Signals are indicators, not guarantees
Presence of a signal should be interpreted as an early indicator, not a guarantee that shared
prosperity will be advanced or harmed by a given AI system. Presence of opportunity or risk
signals for an AI system being assessed is a necessary, but not sufficient, condition for
shared prosperity to be advanced or harmed with the introduction of that AI system into
the economy.
Many societal factors outside of the direct control of AI-creating organizations play a
role in determining which opportunities or risks end up being realized. Holding all other
societal factors constant, the purpose of these Guidelines is to minimize the chance that
shared prosperity-relevant outcomes are worsened and maximize the chance that they are
improved as a result of choices by AI-creating and -using organizations and the inherent
qualities of their technology.
Signals should be considered comprehensively
Signals of opportunity and risk should be considered comprehensively. Presence of a signal
of risk does not automatically mean an AI system in question should not be developed
or deployed. That said, an absence of any signals of opportunity does mean that a given
AI system is highly unlikely to advance shared prosperity and whatever risks it might be
presenting to society are not justified.
The Guidelines
do not suggest
for companies
to stop seeking
to make profit,
but merely to do
it responsibly.

PARTNERSHIP ON AI
13
Signals of opportunity do not “offset” signals of risk
Presence of signals of opportunity should not be interpreted as “offsetting” the presence
of signals of risk. In recognition that benefits and harms are usually borne unevenly by
different groups, the Guidelines strongly oppose the concept of a “net benefit” to shared
prosperity, which is incompatible with a human rights-based approach. In alignment with
the UN Guiding Principles on Business and Human Rights, a mitigation strategy should
be developed for each risk identified, prioritizing the risks of the most severe impactsI
first. Mitigation strategies can range from eliminating the risk or reducing the severity
of potential impact to ensuring access to remedy or compensation for affected groups. If
effective mitigation strategies for a given risk are not available, it should be considered as a
strong argument in favor of meaningful changes in the development, implementation, and
use plans of an AI system, especially if it is expected to affect vulnerable groups.
Analysis of signals is not prescriptive
The analysis of signals of opportunity and risk is not prescriptive. Decisions around the
development, implementation, and use of increasingly powerful AI systems should be made
collectively, allowing for the participation of all affected stakeholders. We anticipate that
two main uses of the signals analysis will include:
Informing stakeholders’ positions in preparation for dialogue around development,
deployment, and regulation of AI systems, as well as appropriate risk mitigation strategies
Identifying key areas of potential impact of a given AI system which warrant deeper analysis
(such as to illuminate their magnitude and distribution)13
and further action
I PAI’s Shared Prosperity
Guidelines use UNGP’s
definition of severity:
an impact (potential or
actual) can be severe “by
virtue of one or more of
the following character-
istics: its scale, scope
or irremediability. Scale
means the gravity of the
impact on the human
right(s). Scope means the
number of individuals
that are or could be
affected. Irremedia-
bility means the ease
or otherwise with which
those impacted could
be restored to their prior
enjoyment of the right(s).”

PARTNERSHIP ON AI
14
Instructions for Performing a Job Impact Assessment
Assess the AI system against the full list of signals
Go over the full list of signals of opportunity and risk and document which signals are
present in the case of the AI system being assessed. Not all signals apply for every AI
system. Document those that do not apply as not applicable, but do not skip or cherry-pick
signals. For each step, document the explanation for the answer for future reference.
For each signal, if you estimated the likelihood of the respective opportunity or risk
materializing as a result of the introduction of the AI system into the economy to be
anything but “zero,” please note the respective signal as “present.”
Certainty in likelihood estimation is not a prerequisite for this high-level assessment and is
assumed to be absent in most cases. When in doubt, note the signal as “present.”
Analyze the distribution of potential benefits and harms
Document in as much detail as possible your understanding of the distribution of potential
benefits and harms of an AI system across skill, geographic, and demographic groups,
and how it might change over time.J
(Are today’s “winners” expected to lose their gains in
the future? The reverse?) The exact steps needed to perform the distribution of impacts
analysis are highly case-specific. PAI is looking to engage with stakeholders to curate a
library of distribution analysis examples for the community to learn from. If you would like
to contribute to this, please get in touch.
Repeat this process for upstream and downstream markets
In order to take into account the possible effects on the competitors, suppliers, and clients
of the AI-using organization, repeat the signal detection and analysis processes not only
J Relevant time period
depends on how long
the AI system being
assessed is expected
to remain in use.
STEP 2
Apply the Job Impact Assessment Tool
Use the high-level Job Impact Assessment Tool to analyze a given AI system:
Go over the full list of
signals of opportunity
and risk
Analyze the distribution
of potential benefits
and harms
Repeat this process
for upstream and
downstream markets

PARTNERSHIP ON AI
15
for the primary market the AI system is intended to be deployed in, but also upstream and
downstream markets.
Proceed to our Stakeholder-Specific Recommendations
After completing the high-level Job Impact Assessment analysis, AI-creating and AI-using
organizations should implement recommended Responsible Practices (where not already
in use) to improve anticipated outcomes — for instance, to eliminate or mitigate anticipated
harms or increase likely benefits for workers and the economy. These Responsible Practices
can be found under Step 3 of the Shared Prosperity Guidelines. (Responsible Practices will
be added and refined through community testing and feedback.)
Policymakers, workers and their representatives can use the results of the high-level Jobs
Impact Assessment to inform their decisions, actions, and agendas as outlined in the
Suggested Uses section under Step 3 of the Shared Prosperity Guidelines. We look forward
to collecting feedback on the Guidelines and curating use examples in partnership with
interested stakeholders. To get involved, please get in touch.
Signals of Opportunity for Shared Prosperity
If one or more of the statements below apply to the AI system being assessed, this
indicates a possibility of a positive impact on shared prosperity-relevant outcomes.
An opportunity signal (OS) is present if an AI system may:
OS1. Generate significant, widely distributed benefits
Will the AI system generate significant, widely distributed benefits to the planet, the public,
or individual consumers? One of the primary motivations for investing in the research and
development of AI is its potential to help humanity overcome some of our most pressing
challenges, including ones related to climate change and the treatment of disease. Hence,
the potential of an AI system to generate public goods or benefit the environment is a
strong signal of opportunity to advance shared prosperity.
Individual consumer benefits can be more controversial as many advocates point out the
growing environmental costs that frequently accompany the commodification of consumer
goods. But if production and consumption are environmentally conscious, a potential to
generate significant and widely distributed consumer benefits is a signal of opportunity
to advance shared prosperity. Cheaper or more high-quality goods or services make
consumers richer in real terms,K
freeing up parts of their incomes to be spent to buy other
goods and services, boosting the demand for labor in respective sectors of the economy.
How significant and widely distributed consumer benefits should be to justify job losses
K This is a result of the
“real income effect.”
For the same nominal
amount of money,
consumers are able
to buy more or higher
quality goods.

PARTNERSHIP ON AI
16
is a political question,L
but quantifying consumer gains per job lost would
help sharpen up any debate about the value of an AI innovation.M
As stated in
“Key Principles for Using the Guidelines,” independently of the magnitude and
distribution of anticipated benefits, appropriate mitigation strategies should be
developed in response to the risk of job losses or wage decreases.
OS2. Boost worker productivity
Will the AI system boost productivity of workers, in particular those in
lower-paid jobs, without increasing strain? By a worker’s productivity, we mean
a worker’s output per hour. A more productive worker is more valuable to their
employer and (all other conditions remaining the same) is expected to be paid
more.N
Therefore, if an AI system comes with a promise of a productivity boost
that is a positive signal. Besides, productivity growth is often the prerequisite
for the creation of consumer benefits discussed in OS1.
However, three important caveats should be noted here.
Caveat 1: Productivity boosts can deepen inequality
It is quite rare for a technology to equally boost productivity for everyone involved in
the production of a certain good, more often it helps workers in certain skill groups
more than others. If it is helping workers in lower-paying jobs relatively more, the
effect could be inequality-reducing. Otherwise, it may be inequality-deepening.
Please document the distribution of the productivity increase across the labor force
when assessing the presence of this opportunity signal.
Caveat 2: Productivity boosts can displace workers
Even if productivity of all workers involved in the production of a certain good is
boosted equally by an AI system, fewer of them might find themselves employed
in the production of that good once the AI system is in place. This is because fewer
(newly more productive) worker-hoursO
are now needed to create the same volume
of output. For production of the good in question to require more human labor after
AI deployment, two conditions must be met:
• Productivity gains of the firm introducing AI need to be shared with its clients
(such as consumers, businesses, or governments) in the form of lower-priced
or higher-quality products — something which is less likely to happen in a
monopolistic environment
• Clients should be willing to buy sufficiently more of that lower-priced or higher-
quality product
If the first condition is met but the second is not, the introduction of the AI
system in question might still be, on balance, labor-demand boosting if it
induces a “productivity effect” in the broader economy. When productivity gains
and corresponding consumer benefits are sufficiently large, consumers will
experience a real income boost generating new labor demand in the production
of complementary goods. That new labor demand might be sufficient to
compensate for the original loss of employment due to an introduction of an AI
system. Issues arise when the productivity gains are too small like in the case of
“so-so” technologies14
or are not shared with consumers. If that is the case, please
document OS2 as “not present” when performing the Job Impact Assessment.
L For example, in 2011, the US
government imposed tariffs to prevent
job losses in the tire industry. Economic
analysis later showed that the tariffs
cost American consumers around $0.9
million per job saved. It seems implau-
sible that such large consumer costs are
worthwhile, relative to the job gains.
M In this paper, Brynjolfsson et al.
estimate the value of many free digital
goods and services. They do so by
proposing a new metric called GDP-B,
which quantifies their benefits rather
than costs, and then estimating
consumers’ willingness-to-pay for
free digital goods and services in terms
of GDP-B.
N As emphasized in Key Principles
for Using the Guidelines, signals of
opportunity are not guarantees: It is
possible that the introduction of a new
technology into the workplace boosts
workers’ productivity but does not lead
to wage growth because, in practice,
workers’ productivity is only one of the
factors determining their wage. Other
factors include how competitive the
market is and how much bargaining
power workers have. In fact, a large
number of countries have been experi-
encing productivity-wage decoupling
in recent decades. This points to a
diminishing role of productivity in deter-
mining wages, but it remains non-zero
and hence has to be accounted for by
the Guidelines.
O The impact of a productivity-
enhancing technology can manifest
itself as a reduction of the size of the
workforce, or a reduction in hours
worked by the same-size labor force.
Either option can negatively impact
shared prosperity.

PARTNERSHIP ON AI
17
Caveat 3: Productivity boosts can significantly hamper job quality
Introduction of an AI system can lead to productivity enhancement through various routes:
by allowing workers to produce more output per hour of work at the same level of effort or by
allowing management to induce a higher level of effort from workers. If productivity boosts are
expected to be achieved solely or mainly through increasing work intensity, please document
OS2 as “not present” when performing the Job Impact Assessment.
Lastly, frontline workers15
reported appreciation for AI systems that boosted their productivity
by assisting them with core tasks. Conversely, technologies that boosted productivity by
automating workers’ core tasks were associated with a reduction in job satisfaction.16
Hence,
pursuit of productivity increases through technologies that eliminate non-core tasks is preferred
over paths that involve eliminating core tasks. Examples of technologies that assist workers on
their core tasks include:
• Training and coaching tools
• Algorithmic decision support systems that give users additional information, analytics, or
recommendations without prescribing or requiring decisions
OS3. Create new paid tasks for workers
Will the AI system create new tasks for humans or move unpaid tasks into paid work?
Technological innovations have a great potential for benefit when they create new formal
sector jobs, tasks, or markets that did not exist before. Consider, for example, the rise of
social media influencers and content creators. These types of jobs were not possible before
the rise of contemporary media and recommendation technologies. It has been estimated
that, in 2018, more than 60 percent of employees were employed in occupations that did
not exist in 1940.17
Caveat 1: Someone’s unpaid tasks can be someone else’s full-time job
It is important to keep in mind that technologies seemingly moving unpaid tasks into paid
ones might, upon closer inspection, be producing an unintended (or deliberately unadvertised)
effect of shifting tasks between paid jobs — often accompanied by a job quality downgrade. For
example, a technology that allows people to hire someone to do their grocery shopping might
convert their unpaid task into someone else’s paid one, but also reduce the demand for full-time
domestic help workers, increasing precarity in the labor market.
Caveat 2: New tasks often go unacknowledged and unpaid
Sometimes the introduction of an AI system adds unacknowledged and uncompensated tasks
to the scope of workers. For example, the labor of smoothing the effects of machine malfunction
remains under the radar in many contexts,18
creating significant unacknowledged burdens
on workers who end up responsible for correcting machine’s errors (without being adequately
positioned to do that).19
When performing the Job Impact Assessment, please explicitly document the applicability of these
two caveats associated with OS3 for the AI system being assessed and its deployment context.
OS4. Support an egalitarian labor market
Will the AI system support a more egalitarian labor market structure? A superstar labor
market structure is a situation where a relatively small number of workers dominate the
market or satisfy most of the labor demand that exists in it. The opposite is an “egalitarian”
labor structure where each worker’s output is small relative to the output of all other
Technological
innovations
have a great
potential for
benefit when
they create new
formal sector
jobs, tasks, or
markets that
did not exist
before.

PARTNERSHIP ON AI
18
workers in the industry. The key factor that makes a labor market’s structure egalitarian
is the presence of a need to invest an additional unit of worker time to serve an additional
consumer. For example, the rise of the music recording industry has made its labor market
structure less egalitarian for musicians. Today, to satisfy the demand for music from an
additional customer, musicians do not need to physically get in front of them or do any
additional work.
OS5. Be appropriate for lower-income geographies
Will the AI system be appropriate for lower-income geographies? Capital and labor
of various skill types can be relatively more or less abundant in different countries.
Technologies that take advantage of the factor of production (capital or labor of a certain
skill type) that is relatively more abundant in a given country and do not require much of a
factor that is relatively scarce there are deemed appropriate for that country.
Generally, capital is relatively more abundant in the higher-income countries while labor is
relatively more abundant in the lower-income countries, many of which also struggle with
poor learning outcomes limiting the training the workforce receives.20
Therefore, capital-
intensive labor-saving AI systems are generally inappropriate for lower-income countries
whose main comparative advantage is relatively abundant labor.21
Such technologies being
adopted by high-income countries can hurt economic outcomes in lower-income countries
because competitive forces in the export industries force the latter to adopt those
technologies to remain competitive.22 23
Consequently, lower-income countries would greatly benefit from access to technologies
that would allow them to stay competitive by leveraging their abundant labor resources and
creating gainful jobs that do not require high levels of educational attainment.
When assessing the presence of this signal, please also document if and how the relative
abundance of capital and labor of various skill types is expected to change over time.
OS6. Broaden access to the labor market
Will the AI system broaden access to the labor market? AI systems that allow communities
with limited or no access to formal employment to get access to gainful formal sector jobs
are highly desirable from the perspective of broadly shared prosperity. Examples include AI
systems that:
• Assist the disabled
• Make it easier to combine work and caregiving responsibilities
• Enable work in languages the worker does not have a fluent command of
OS7. Boost revenue share of workers and society
Will the AI system boost workers’ and society’s share of an organization’s revenue?
Capital-
intensive
labor-saving
AI systems
are generally
inappropriate
for lower-
income
countries.

PARTNERSHIP ON AI
19
Workers’ share of revenue is the percentage of an organization’s revenue spent on workers’
wages and benefits. For the purposes of these Guidelines, we suggest excluding C-suite
compensation when calculating workers’ share.
If, following the introduction of an AI system, workers’ share of organization’s revenue is
expected to grow or at least stay constant, it is a very strong signal that the AI system in
question will serve to advance shared prosperity. The opposite is also true. If, following the
introduction of an AI system, workers’ share of organization’s revenue is expected to shrink,
it is a very strong signal that the AI system in question will harm shared prosperity.
Please note that worker benefits are included in workers’ share of an organization’s revenue.
For example, consider an organization that adopts a productivity-enhancing AI system
which allows it to produce the same or greater amount of output with fewer hours of work
needed from human workers. That organization can decide to retain the same size of the
workforce and share productivity gains with it (for example, in the form of higher wages,
longer paid time off, or shorter work week at constant weekly pay), keeping the workers’
share of revenue constant or growing. That would be a prime example of using AI to advance
shared prosperity.
Lastly, if an organization was able to generate windfall gains from AI development or usage
and is committed to sharing the gains not only with workers it directly employs but the
rest of the world’s population as well, that can be a great example of using AI to advance
shared prosperity. While some have proposed this,24
more research is needed to design
mechanisms for making sure windfall gains are distributed equitably and organizations
can be expected to reliably honor their commitment to distribute their gains.
OS8. Respond to needs expressed by impacted workers
Did workers who will use the AI system or be affected by it (or their representatives)
identify the need for the system? AI systems created from a worker’s idea or identified
need build in workers’ job expertise and preferences from the outset, making it more likely
the AI systems will be beneficial or useful to workers affected by them and welcomed
as such. Much of the current AI development pipeline starts with advances in research
and development, only later identifying potential applications and product-market fit.
The market for workplace AI technology is largely composed of senior executives and
managers, creating a potential misalignment between needs perceived by budget holders
and managers and the needs perceived by the workers who use or are most affected by the
technology. AI systems emerging from the ideas and needs of workers who use or are most
affected by them (or their representatives, who represent the collective voice of a given
set of workers, not just the perspective of an individual worker) reduce this potential for
misalignment.25

PARTNERSHIP ON AI
20
OS9. Be co-developed with impacted workers
Were workers who will ultimately use or be affected by the AI system (or their
representatives) included and given agency in every stage of the system’s development?
Workers are subject matter experts in their own tasks and roles, and can illuminate
opportunities and challenges for new technologies that are unlikely to be seen by those
with less familiarity with the specifics of the work. The wisdom of workers who use or are
most affected by AI systems introduced throughout development can smooth many rough
edges that other contributors might only discover after systems are in the market and
implemented. Where relevant worker representatives exist, they should be brought into the
development process to represent collective worker interests from start to finish.
Fully offering affected workers agency in the development process requires taking the
time to understand their vantage points, and equip them or their representatives with
enough knowledge about the proposed technology to meaningfully participate. They also
must be afforded sufficient decision-making power to steer projects and, if necessary,
end them in instances where unacceptable harms cannot be removed or mitigated.
This also necessitates protecting their ability to offer suggestions freely without fear
of repercussions. Without taking these steps, participatory processes can still lead to
suboptimal outcomes — and possibly create additional harms through covering problems
with a veneer of worker credibility.
OS10. Improve job quality or satisfaction
Was the AI system intended to improve job quality or increase job satisfaction? AI
technology has the potential to improve many aspects of job quality and job satisfaction,
from increasing occupational safety to providing personalized coaching that leads to
career advancement. This requires taking job quality, worker needs, and worker satisfaction
seriously.
Two important caveats are required for this signal.
Caveat 1: Systems can improve one aspect of job quality while harming another
For example, many AI technologies positioned as safety enhancements are in reality invasive
surveillance technologies. Though safety improvements may occur, harms to human rights,
stress rates, privacy, job autonomy, job intensity, and other aspects of job quality may occur as
well. Other AI systems purport to improve job quality by automating tasks workers dislike (see
RS1 for more detail on the risks of task elimination).
When a system enhances one aspect of job quality while endangering another, this signal can
still be counted as “present,” but the need to consider the rest of the opportunity and risk signals
is particularly important.
Caveat 2: AI systems are sometimes deployed to redress job quality harms created by
other AI systems
For example, some companies have introduced AI safety technologies to correct harms resulting
from the prior introduction of an AI performance target-setting system that encouraged
dangerous overwork.26
Workers are
subject matter
experts in their
own tasks and
roles, and can
illuminate
opportunities
and challenges
for new
technologies.

PARTNERSHIP ON AI
21
When this is the case, the introduction of the new AI system to redress the harms of the old does
not count for this signal and should be marked as “not present.”
Instead of introducing new AI systems with their own attendant risks, the harms from the
existing systems should be addressed in line with the Responsible Practices provided by the
Guidelines for AI-using organizations and additional case-specific mitigations.
Signals of Risk to Shared Prosperity
If one or more of the statements below apply to the AI system being assessed, this
indicates a possibility of a negative impact on shared prosperity-relevant outcomes.
Some of the signals of risk to shared prosperity described in the Guidelines are actively
sought by companies as profit-making opportunities. The Guidelines DO NOT suggest that
companies should stop seeking profits, just that they should do so responsibly.
Profit-generating activities do not necessarily have to harm workers and communities,
but some of them do. The presence of signals of risk indicate that an AI system being
assessed, while possibly capable of generating profit for a narrow set of beneficiaries, is
likely to do that at the expense of shared prosperity, and thus might be undesirable from the
societal benefit perspective. We encourage companies to follow the Guidelines, developing
and using AI in ways that generate profit while also advancing shared prosperity.
For-profit companies might feel pressure from investors to cut their labor costs no matter
the societal price. We encourage investors and governments to join civil society in an effort
to incentivize responsible business behavior with regards to shared prosperity and labor
market impact.
Some practices or outcomes included in this section are illegal in some jurisdictions, and
as such are already addressed in those locations. We include them here due to their legality
in other jurisdictions.
A risk signal (RS) is present if an AI system may:
RS1. Eliminate a given job’s core tasks
Will the AI system eliminate a significant share of tasks for a given job? A lot of
technological innovations eliminate some job tasks that were previously done by human
workers. That is not necessarily an unwelcome development, especially when those
technologies also create new paid tasks for humans (see OS3), boost job quality (see
OS10), or bring significant broadly distributed benefits (see OS1). For example, it can be
highly desirable to automate tasks posing unmitigable risks to workers’ physical or mental
health. Primary research conducted by the AI and Shared Prosperity Initiative indicated
that frontline workers often experience automation of their non-core tasks as helpful and
productivity-boosting.27
Profit-
generating
activities
do not
necessarily
have to harm
workers and
communities,
but some
of them do.

PARTNERSHIP ON AI
22
However, if an AI system is primarily geared towards eliminating core paid tasks without
much being expected in terms of increased job quality or broadly shared benefits, nor in
terms of new tasks for humans being created in parallel, then it warrants further attention
as posing a risk to shared prosperity. The introduction of such a system will likely lower
the demand for human labor, and thus wage or employment levels for affected workers.28
Automation of core tasks can also be experienced by workers as directly undermining their
job satisfaction since workers’ core responsibilities are closely tied to their sense of pride
and accomplishment in their jobs. For workers who see their jobs as an important part
of their identity, core tasks are a major aspect of how they see themselves in the world.29
Automation of core tasks can also lower the skill requirements of a job and reduce the
formation of skills needed to advance to the next level.30
Please note that to evaluate the share of a given job’s tasks being eliminated, those tasks
should be weighted by their importance for the production of the final output. We consider
task elimination above 10% significant enough to warrant attention.
RS2. Reallocate tasks to lower-paid or more precarious jobs
Will the AI system enable reallocation of tasks to lower-paid or more precarious jobs
or informal or unpaid labor? Often, while not eliminating human tasks on balance, AI
technology enables shifting tasks from full-time jobs to unpaid or more precarious labor.
The latter can happen, for example, through the “gig-ification” of work: technologically
enabled separation of “time on task” and “idle time” which leads to unstable and
unpredictable wages as well as the circumvention of minimum wage laws.
Paid tasks can also be converted into unpaid when new technology enables them to
be performed by customers. Examples of that are self-checkout kiosks or automated
customer support.31
RS3. Reallocate tasks to higher- or lower-skilled jobs
Will the AI system enable the reallocation of tasks to jobs with higher or lower specialized
skills requirements? Jobs with higher specialized skills requirements generally are better
compensated, hence an AI system shifting tasks into such jobs will likely lead to a positive
effect of more of them being opened up. However, those jobs might not be accessible to
people affected by task reallocation because those people might not possess the newly
required specialized skills. Retraining and job matching support programs can help
here, though those often fall short. Word processor is an example of a technology that
reallocated typing-related tasks away from typists to managers. Generative AI applications
are an example of a recent technology anticipated to induce broad-reaching shifts in skill
requirements of large swaths of jobs.32 33 34
Importantly, AI-induced reallocation of tasks to jobs with lower specialized skills
requirements may be positive but is still a risk signal warranting further attention, because
Automation of
core tasks can
be experienced
by workers
as directly
undermining
their job
satisfaction.

PARTNERSHIP ON AI
23
lowering specialized skill requirements can lower not only the barriers to entry to the
occupation, but also prevailing wages.
RS4. Move jobs away from geographies with few opportunities
Will the AI system move job opportunities away from geographies where there would
be few remaining? Due to associated costs and excessive immigration barriers, labor
mobility remains low, both within and between countries. As a result, changes that move
job opportunities from one area to another can harm workers in the losing area. Research
suggests that disappearance of stable, well-paying jobs can profoundly re-shape regions,
leading to a rise in “deaths of despair,” addictions, and mental health problems.35 36
Impacted communities might be able to bounce back from job loss if comparable
alternative job opportunities are sufficiently available in their area. But even when those
exist, the presence of labor market frictions make it important to invest in creating support
programs to help workers move into new jobs of comparable quality.
In addition to jobs disappearing as the direct effect of labor-saving technology being
introduced in a region, please note that this effect can also be an indirect result of labor-
saving technology initially introduced in a completely different region or country. Due
to excessive immigration barriers, AI developers based in high-income countries face
massively inflated incentives to create labor-saving technologies far in excess of what
would be socially optimal given the world’s overall level of labor supply/demand for jobs.37
Once that technology is developed in the high-income countries it gets deployed all over
the world, including countries facing a dire need of formal sector jobs.38
RS5. Increase market concentration and barriers to entry
Will an AI system increase market concentration and barriers to market entry? An increase
in market concentration is a signal of a possible labor market impact to come for at least
two reasons:
• It increases the risk of job cuts by competing firms
• It makes it less likely that the winning firm shares efficiency gains with workers in the
form of better wages/benefits or with consumers in the form of lower prices/higher-
quality products
Therefore, in a monopolistic market, any benefits brought on by AI are likely to be shared
by few, while the harms might still be widely distributed. Similarly, job impacts that might
occur in upstream or downstream industries due to an AI-induced increase in market
concentration need to be accounted for as well.
RS6. Rely on poorly treated or compensated outsourced labor
Will the AI system rely on, for either model training or operation, outsourced labor deprived
of a living wage and decent working conditions? The process of building datasets for
Disappearance
of stable,
well-paying
jobs can
profoundly
re-shape
regions.

PARTNERSHIP ON AI
24
model training can be highly labor-intensive. It often requires human workers (whom we
will refer to as data enrichment professionals) to review, classify, annotate, and otherwise
manage massive amounts of data. Despite the foundational role played by data enrichment
professionals, a growing body of research reveals the precarious working conditions that
they face, which include:39
• Inconsistent and unpredictable compensation for their work
• Unfairly rejected and therefore unpaid labeling tasks
• Long, ad-hoc working hours
• Lack of means to contest or get an explanation for the decisions affecting their
take-home pay and ratings
• Lack of transparency around data enrichment labor sourcing practices in the AI
industry exacerbate this issue.
RS7. Use training data collected without consent or compensation
Will the AI system be trained using a dataset containing data collected without consent
and/or compensation? AI systems can be trained on data that embeds the economically-
relevant know-how of people who generated that data, which can be especially problematic
if the subsequent deployment of that AI system reduces the demand for labor of those
people. Examples include but are not limited to:
• Images created by artists and photographers that are used to train generative AI
systems
• Keystrokes and audio recordings of human customer service agents used to create
automated customer service routines
• Records of actions taken by human drivers used to train autonomous driving systems
RS8. Predict the lowest wages a worker will accept
Will the AI system be used to predict the lowest wage a given worker would accept? It has
been documented that workers can experience the impact of AI systems used for workforce
management as effectively depriving them of being able to predict their take-home wages
with any amount of certainty.40
An AI system allowing predictions about the lowest wages
an individual worker would accept is analogous to a system allowing for perfect price
discrimination of consumers. Price discrimination, while always driven by monopoly power
and thus inefficient, is considered acceptable in certain situations, such as reduced price
of museum admission for seniors and students. However, that acceptability is predicated
on the transparency of the underlying logic. A possibility of using an algorithmic system
to create take-home pay “personalization,” especially based on logic that is opaque to the
workers or ever-changing, should serve as a strong signal of a potential negative impact
on shared prosperity. A related risk for informal workers is the use of AI to reduce their
bargaining power relative to those they contract with. Information asymmetries created
through AI use by purchasers of their work are an emerging risk to workers in the informal
sector.41

PARTNERSHIP ON AI
25
RS9. Accelerate task completions without other changes
Will the AI system accelerate task completion without meaningfully changing resources,
tools, or skills needed to accomplish the tasks? Some AI systems push workers to higher
performance on goals, targets, or KPIs without modifying how the work is done. Examples
of this include speeding up the pace with which workers are expected to complete tasks
or using AI to set performance goals that are just out of reach for many workers. When this
occurs without additional support for workers in the form of streamlining, simplifying, or
otherwise improving the process of completing the task, it risks higher stress and injury
rates for workers.
RS10. Reduce schedule predictability
Will the AI system reduce
the amount of advance notice a worker receives regarding
changes to their working hours? Schedule predictability is strongly tied to workers’ physical
and mental health.42 43
Automated, last-minute scheduling software can harm workers’:
• Emotional well-being through increased stress
• Occupational safety and health through sleep deprivation/unpredictability and the
physical effects of stress
• Financial well-being through missed shifts and increased need for more expensive
transit (for example, ride-hailing services at times when public transit isn’t frequent
or safe).
Recent AI technology designed to lower labor costs by reducing the number of people
working during predicted “slow” times has disrupted schedule predictability, with workers
receiving minimal notice about hours that have been eliminated from or added to their
schedules.
RS11. Reduce workers’ break time
Will the AI system infringe on workers’ breaks or encourage them to do so? Workers’ breaks
are necessary for their recovery from physically, emotionally, or intellectually strenuous
or intense periods of work, and are often protected by law. Some AI systems billed as
productivity software infringe on workers’ breaks by sending them warnings based on
the time they’ve spent away from their workstations or “off-task,” even during designated
breaks or while they are using allotted break time.44
Others implicitly encourage workers to
skip breaks by setting overly ambitious performance targets that pressure workers to work
through downtime to meet goals. These systems can foster higher rates of injury or stress,
undermine focus, and reduce opportunities to form social relationships at work.
RS12. Increase overall difficulty of tasks
Will the AI system increase the overall difficulty of tasks? When AI systems are used to
automate less demanding tasks (for example, the most straightforward, emotionally

PARTNERSHIP ON AI
26
neutral customer requests in a call center), workers may be left with a higher concentration
of more demanding tasks, effectively increasing the difficulty of their job.45
Difficulty
increases may take the form of more physically, emotionally, or intellectually demanding
tasks. The higher intensity may also place them at higher risk of burning out. While some
workers may welcome the added challenge, the above concerns merit caution, especially if
workers are not compensated equitably for the increased difficulty.
RS13. Enable detailed monitoring of workers
Will the AI system monitor something other than the pace and quality of task completion?
The use of AI to monitor workers is just the latest entry in the long history of the
technological surveillance of labor.46
However, AI capabilities have increased the frequency,
comprehensiveness, and intensiveness of on-the-job monitoring. This use of AI often
extends beyond monitoring of workers’ direct responsibilities and outputs, including
information as varied as their time in front of their computer or time spent actively using
their computer, their movements through an in-person worksite, and the frequency and
content of communications with other workers. This detailed monitoring risks:
• Increasing stress and anxiety
• Harming their privacy
• Causing them to feel a lack of trust from their employer
• Undermining their sense of autonomy on the job
• Lowering engagement and job satisfaction
• Chilling worker organizing, undermining worker voice.47 48
While monitoring systems can have legitimate uses (such as enhancing worker safety),
even good systems can be abused, particularly in environments with low worker agency or
an absence of regulations, monitoring, and enforcement of worker protections.49
RS14. Reduce worker autonomy
Will the AI system reduce workers’ autonomy, decision-making authority, or control over
how they complete their work? Autonomy, decision-making authority, job control, and the
exercise of discernment in performing one’s job are correlated with high job quality and
job satisfaction.50
Reducing scope for these activities could also be a sign of a shift from
a “high-road” staffing approach (where experience and expertise is valued) to a “low-
road” approach (where less training or experience is needed and thus workers hold less
bargaining power and can be more easily replaced). In the informal sector, this may appear
as a reduction in the scope for design and creativity by artisans and garment workers.51
RS15. Reduce mentorship or apprenticeship opportunities
Will the AI system reduce workers’ opportunities for mentorship or apprenticeship?
Automated training, automated coaching, and automation of entry-level tasks may
Monitoring
systems can
have legitimate
uses, but even
good systems
can be abused.

PARTNERSHIP ON AI
27
lower workers’ opportunities for apprenticeship and mentorship. Apprenticeship is
an important way for workers to learn on the job, and develop the skills they need to
advance.52
Mentorship and apprenticeship can help workers develop social relationships
and community with peers and supervisors. Additionally, mentors can help workers learn
to navigate unspoken rules and norms in the workplace, and assist them with career
development within and beyond their current workplace.
RS16. Reduce worker satisfaction
Will the AI system reduce the motivation, engagement, or satisfaction of the workers
who use it or are affected by it? While this test directly speaks to meaning, community,
and purpose, it is also a proxy for other aspects of worker well-being. Demotivation and
disengagement are signs of lowered job satisfaction and serve as indications of other job
quality issues.
RS17. Influence employment and pay decisions
Will the AI system make or suggest decisions on recruitment, hiring, promotion,
performance evaluation, pay, wage penalties, and bonuses? The decisions outlined in this
signal are deeply meaningful to workers, meriting heightened attention from employers.
Automation of these decisions should raise concern, as automated systems might lack the
complete context necessary for these decisions and risk subjecting workers to “algorithmic
cruelty.”53
They also risk introducing additional discriminatory bases for decisions, beyond
those already existent in human decisions.54
In instances where AI systems are used to
suggest (rather than decide) on these questions, careful implementation focused on
increasing decision accuracy and transparency can benefit workers. However, human
managers using these systems often find it undesirable or difficult to challenge or override
recommendations from AI, making the system’s suggestions more binding than they may
initially appear and meriting additional caution in these uses.
RS18. Operate in discriminatory ways
Will the AI system operate in ways that are discriminatory? AI systems have been
repeatedly shown to reproduce or intensify human discrimination patterns on demographic
categories such as gender, race, age, and more.55 56 57 58
Workplace AI systems should be
rigorously tested to ensure that they operate fairly and equitably.
Workplace AI
systems should
be rigorously
tested to ensure
that they
operate fairly
and equitably.

PARTNERSHIP ON AI
28
STEP 3
Follow Our Stakeholder-Specific
Recommendations
Foster shared prosperity by enacting best practices and suggested uses:
For AI-creating
organizations
For AI-using
organizations
For policymakers For labor
organizations
and workers
AI-Creating Organizations (RPC)
Use of workplace AI is still in early stages, and as a result information about what should
be considered best practices for fostering shared prosperity is still preliminary. Below
is a list for AI-creating organizations of starter sets of practices aligned with increasing
the likelihood of benefits to shared prosperity and decreasing the likelihood of harms
to it. The list is drawn from early empirical research in the field, historical analogues for
transformative workplace technologies, and theoretical frameworks yet to be applied in
practice. For ease of use, the lists of Responsible Practices are organized by the earliest AI
system lifecycle stage where the practice can be applied.
AT AN ORGANIZATIONAL LEVEL
RPC1. Make a public commitment to identify, disclose, and mitigate the risks
of severe labor market impacts presented by AI systems you develop
Multiple AI-creating organizations aspire (according to their mission statements and
responsible AI principles) to develop AI that benefits everyone. Very few of them, however,
currently publicly acknowledge the scale of labor market disruptions their AI systems might
bring about or make efforts to help communities that stand to be affected have a say in the
decisions determining the path, depth, and distribution of labor market disruptions. At the
same time, AI-creating organizations are often best positioned to anticipate labor market
risks well in advance of those becoming apparent to other stakeholders, thus making risk
disclosures by AI-creating organizations a valuable asset for governments and societies.

PARTNERSHIP ON AI
29
The public commitment to disclose severe risks* should specify the severity threshold
considered by the organizations to warrant disclosure, as well as explain how the threshold
level of severity was chosen and what external stakeholders were consulted in that decision.
Alternatively, an organization can choose to set a threshold in terms of an AI system’s
anticipated capabilities and disclose all risk signals which are present for those systems.
For example, if the expected return on investment from the deployment of an AI system
is a multiple greater than 10, or more than one million US dollars were spent on training
compute and data enrichment, its corresponding risks would be subject to disclosure.P
DURING THE FULL AI LIFECYCLE
RPC2. In collaboration with affected workers, perform Job Impact
Assessments early and often throughout the AI system lifecycle
Run opportunity and risk analyses early and often in the AI research and product
development process, using the data available at each stage. Update as more data
becomes available (for example, as product-market fit becomes clearer or features are built
out enough for broader worker testing and feedback). Whenever applicable, we suggest
using AI system design and deployment choices to maximize the presence of signals of
opportunity and minimize the presence of signals of risk.
Always solicit the input of workers that stand to be affected — both incumbents as well
as potential new entrants — and a multi-disciplinary set of third-party experts when
assessing the presence of opportunity and risk signals. Make sure to compensate external
contributors for their participation in the assessment of the AI system.
Please note that the analysis of opportunity and risk signals suggested here is different
from red team analysis suggested in RPC13. The former identifies risks and opportunities
created by an AI system working perfectly as intended. The latter identifies possible harms
if the AI system in question malfunctions or is misused.
RPC3. In collaboration with affected workers, develop mitigation strategies
for identified risks
In alignment with UN Guiding Principles for Business and Human Rights, a mitigation
strategy should be developed for each risk identified, prioritizing the risks primarily by
severity of potential impact and secondarily by its likelihood. Severity and likelihood of
potential impact are determined on a case-by-case basis.Q
Mitigation strategies can range from eliminating the risk or reducing the severity of
potential impact to ensuring access to remedy or compensation for affected groups. If
effective mitigation strategies for a given risk are not available, this should be considered a
strong argument in favor of meaningful changes in the development plans of an AI system,
especially if it is expected to affect vulnerable groups.
P These thresholds are
used for illustrative
purposes only: AI
creating organizations
should set appropriate
thresholds and explain
how they were arrived at.
Thresholds need to be
reviewed and possibly
revised regularly as the
technology advances.
Q An algorithm described
here is very useful for
determining the severity
of potential quanti-
tative impacts (such as
impacts on wages and
employment), especially
in cases with limited
uncertainty around the
future uses of the AI
system being assessed.
Always solicit
the input of
workers that
stand to be
affected.

PARTNERSHIP ON AI
30
Engaging adequately compensated external stakeholders in the development of mitigation
strategies is critical to ensure important considerations are not being missed. It is
especially critical to engage with representatives of communities that stand to be affected.
RPC4. Source data enrichment labor responsibly
Key requirements for the responsible sourcing of data enrichment services (such as, data
annotation and real-time human verification of algorithmic predictions) include:
• Always paying data enrichment workers above the local living wage
• Providing clear, tested instructions for data enrichment tasks
• Equipping workers with simple and effective mechanisms for reporting issues, asking
questions, and providing feedback on the instructions or task design
In collaboration with our Partners, PAI has developed a library of practitioner resources for
responsible data enrichment sourcing.
DURING SYSTEM ORIGINATION AND DEVELOPMENT
RPC5. Create and use robust and substantive mechanisms for worker
participation in AI system origination, design, and development
Workers who will use or be affected by AI hold unique perspectives on important needs
and opportunities in their roles. They also possess particular insight into how AI systems
could create harm in their workplaces. To ensure AI systems foster shared prosperity, these
workers should be given agency in the AI development process from start to finish.
This work does not stop at giving workers a seat at the table throughout the development
process. Workers must be properly equipped with knowledge of product functions,
capabilities, and limitations so they can draw meaningful connections to their role-based
knowledge. Additionally, care must be taken to create a shared vocabulary on the team, so
that technical terms or jargon do not unintentionally obscure or mislead. Workers must
also be given genuine decision-making power in the process, allowing them to shape
product functions and features, and be taken seriously on the need to end a project if they
identify unacceptable harms that cannot be resolved.
RPC6. Build AI systems that align with worker needs and preferences
AI systems welcomed by workers largely fall into three overarching categories:
• Systems that directly improve some element of job quality
• Systems that assist workers to achieve higher performance on their core tasks
• Systems that eliminate undesirable non-core tasks (See OS3, RS1, and RS2 for
additional detail)
Starting with one of these objectives in mind and creating robust participation
mechanisms for workers throughout the design and implementation process is likely to
Workers who
will use or be
affected by AI
hold unique
perspectives
on important
needs and
opportunities
in their roles.

PARTNERSHIP ON AI
31
result in win-win-wins for AI creators, employers who implement AI, and the workers who
use or are affected by them.
RPC7. Build AI systems that complement workers (especially those in
lower-wage jobs), not ones that act as their substitutes
A given AI system complements a certain group of workers if the demand for labor of that
group of workers can be reasonably expected to go up when the price of the use of that AI
system goes down. A given AI system is a substitute for a certain group of workers if the
demand for labor of that group of workers is likely to fall when the price of the use of that AI
system goes down.
Note that the terms “labor-augmenting” technology and “labor-complimentary” technology
are often erroneously used interchangeably. “Labor-augmenting technology” is increasingly
being used as a loose marketing term which frames workplace surveillance technology as
worker-assistive.59
Getting direct input from workers is very helpful for differentiating genuinely
complementary technology from the substituting kind. Please also see the discussion
of the distinction between core and non-core tasks and the acceptable automation
thresholds in RS1.
RPC8. Ensure workplace AI systems are not discriminatory
In general, AI systems frequently reproduce or deepen discriminatory patterns in society,
including ones related to race, class, age, and disability. Specific workplace systems have
shown a propensity for the same. Careful work is needed to ensure any AI systems affecting
workers or the economy do not create discriminatory results.
BEFORE SELLING OR DEPLOYING THE SYSTEM
RPC9. Provide meaningful, comprehensible explanations of the AI system’s
function and operation to workers using or affected by it
The field of explainable AI has advanced considerably in recent years, but workers remain
an underrepresented audience for AI explanations.60
Providing workers explanations of
workplace AI systems tailored to the particulars of their roles and job goals enables them
to understand the tools’ strengths and weaknesses. When paired with workers’ existing
subject matter expertise in their own roles, this knowledge equips workers to most
effectively attain the upsides and minimize the downsides of AI systems, meaning AI
systems can enhance their overall job quality across the different dimensions of well-being.
AI systems
frequently
reproduce
or deepen
discriminatory
patterns in
society.

PARTNERSHIP ON AI
32
RPC10. Ensure transparency about what worker data is collected, how and
why it will be used, and enable opt-out functionality
Privacy and ownership over data generated by one’s activities are increasingly rights
recognized inside and outside the workplace. Respect for these rights requires fully
informing workers about the data collected on them and inferences made, how they are
used and why, as well as offering them the ability to opt out of collection and use.61
Workers
should also be given the opportunity to individually or collectively forbid the sales of
datasets that include their personal information or personally identifiable information.
In particular, system design should follow the data minimization principle: collect only
the necessary data, for the necessary purpose, and hold it only for the necessary amount
of time. Design should also enable workers to know about, correct, or delete inferences
about them. Particular care must be taken in workplaces, as the power imbalance between
employer and employee undermines workers’ ability to freely consent to data collection and
use compared to other, less coercive contexts.62
RPC11. Embed human recourse into decisions or recommendations you offer
AI systems have been built to hire workers, manage them, assess their performance, and
promote or fire them. AI is also being used to assist workers with their tasks, coach them,
and complete tasks previously assigned to them. In each of these decisions allocated
to AI, the technologies have accuracy as well as comprehensiveness issues. AI systems
lack the human capacity to bring in additional context relevant to the issue at hand. As
a result, humans are needed to validate, refine, or override AI outputs. In the case of task
completion, an absence of human involvement can create harms to physical, intellectual,
or emotional well-being. In AI’s use in employment decisions, it can result in unjustified
hiring or firing decisions. Simply placing a human “in the loop” is insufficient to overcome
algorithmic bias: demonstrated patterns of deference to the judgment of algorithmic
systems. Care must be taken to appropriately position the strengths and weaknesses of AI
systems and empower humans with final decision-making power.63
RPC12. Apply additional mitigation strategies to sales and use in
environments with low worker protection and decision-making power
AI systems are less likely to cause harm in environments with:
• High levels of legal protection, monitoring, and enforcement for workers’ rights (such
as those related to health and safety or freedom to organize)
• High levels of worker voice and negotiating ability (due to strong protections for worker
voice or high demand for workers’ comparatively scarce skills), especially those where
workers have meaningful input into decisions regarding the introduction of new
technologies
These factors encourage worker-centric AI design. Workers in such environments also
possess a higher ability to limit harms from AI systems (such as changing elements of an
implementation or rejecting the use of the technology as needed), including harms outside

PARTNERSHIP ON AI
33
direct legal protections. This should not, however, be treated as a failsafe for harmful
technologies, particularly when AI systems can easily be adopted in environments where
they were not originally intended.64
In environments where workers lack legal protection
and/or decision-making power, it is especially important to scrutinize uses and potential
impacts, building in additional mitigations to compensate for the absence of these worker
safeguards. Contractual or licensing provisions regarding terms of use, rigorous customer
vetting, and geofencing are some of the many steps AI-creating organizations can take to
follow this practice. Care should be taken to adopt fine-grained mitigation strategies where
possible such that workers and economies can reap the gains of neutral or beneficial uses.
RPC13. Red team AI systems for potential misuse or abuse
The preceding points have focused on AI systems working as designed and intended.
Responsible development also requires comprehensive “red teaming” of AI systems to
identify vulnerabilities and the potential for misuse or abuse. Adversarial ML is increasingly
a part of standard security practice. Additionally, the development team, workers in relevant
roles, and external experts should test the system for misuse and abusive implementation.
RPC14. Ensure AI systems do not preclude the sharing of productivity gains
with workers
The power and responsibility to share productivity gains from AI system implementation
lies mostly with AI-using organizations. The role of AI-creating organizations is to make
sure the functionality of an AI system does not fundamentally undermine opportunities for
workers to share in productivity gains, which would be the case if an AI system de-skills
jobs and makes workers more likely to be viewed as fungible or automates a significant
share of workers’ core tasks.
RPC15. Request deployers to commit to following PAI’s Shared Prosperity
Guidelines or similar recommendations
The benefit to workers and society from following these practices can be meaningfully
undermined if organizations deploying or using the AI system do not do their part to
advance shared prosperity. We encourage developers to make adherence to the Guidelines’
Responsible Practices a contractual obligation during the selling or licensing of the AI
system for deployment or use by other organizations.
The role of
AI-creating
organizations
is to make
sure the
functionality
of an AI system
does not
fundamentally
undermine
opportunities
for workers
to share in
productivity
gains.

PARTNERSHIP ON AI
34
AI-Using Organizations (RPU)
Use of workplace AI is still in early stages, and as a result information about what should be
considered best practices for fostering shared prosperity is still preliminary. Below is a list
for AI-using organizations of starter sets of practices aligned with increasing the likelihood
of benefits to shared prosperity and decreasing the likelihood of harms to it. The list is
drawn from early empirical research in the field, historical analogues for transformative
workplace technologies, and theoretical frameworks yet to be applied in practice. For ease
of use, the lists of Responsible Practices are organized by the earliest AI system lifecycle
stage where the practice can be applied.
AT AN ORGANIZATIONAL LEVEL
RPU1. Make a public commitment to identify, disclose, and mitigate the
risks of severe labor market impacts presented by AI systems you use
Labor practices and impacts are increasingly a part of suggested, proposed, or required
non-financial disclosures. These disclosures include practices affecting human rights,
management of human capital, and other social and employee issues. Regulatory
authorities have suggested, proposed, or required these disclosures as material to investor
decision-making,R
as well as for the benefit of the broader society. We recommend that
AI-using organizations identify, disclose, and mitigate the risks of severe labor market
impacts for the same rationales, as well as to provide both prospective and existing
workers with the information they need to make informed decisions about their own
employment.
The public commitment to disclose severe risksS
should specify the severity threshold
considered by the organization to warrant disclosure, as well as explain how the threshold
level of severity was chosen and what external stakeholders were consulted in that
decision.
Alternatively, an organization can choose to set a threshold in terms of an AI system’s
marketed capabilities and disclose all risk signals which are present for systems meeting
that threshold. For example, if an organization’s expected return on investment from the
use of an AI system under assessment is a multiple greater than 10, its corresponding risks
would be subject to disclosure. In instances where organizational impact is driven by a
series of smaller system implementations, the organization could choose to disclose all
risk signals present once the cumulative cost decrease or revenue increase exceeds 5%.T
S PAI’s Shared Prosperity
Guidelines use UNGP’s
definition of severity:
an impact (potential or
actual) can be severe “by
virtue of one or more of
the following character-
istics: its scale, scope
or irremediability. Scale
means the gravity of the
impact on the human
right(s). Scope means the
number of individuals
that are or could be
affected. Irremediability
means the ease or
otherwise with which
those impacted could
be restored to their prior
enjoyment of the right(s).”
R See, for instance, the
Guiding Principles on
Business and Human
Rights: Implementing the
United Nations “Protect,
Respect and Remedy”
Framework or the United
States Securities and
Exchange Commission’s
2023 agenda, as reported
in Reuters.
T A recent study of
corporate respondents
showed roughly one
quarter of respondents
were able to achieve
a 5% improvement
to EBIT in 2021. As AI
adoption becomes more
widespread, we anticipate
more organizations will
meet this threshold.

PARTNERSHIP ON AI
35
THROUGHOUT THE ENTIRE PROCUREMENT PROCESS, FROM
IDENTIFICATION TO USE
RPU2. Commit to neutrality towards worker organizing and unionization
As outlined in the signals of risk above, AI systems pose numerous risks to workers’
human rights and well-being. These systems are implemented and used in employment
contexts that often have such comprehensive decision-making power over workers that
they can be described as “private governments.”65
As a counterbalance to this power,
workers may choose to organize to collectively represent their interests. The degree to
which this is protected, and the frequency with which it occurs, differs substantially
by location. Voluntarily committing to neutrality towards worker organizing is an
important way to ensure workers’ agency is respected and their collective interests have
representation throughout the AI use lifecycle if workers so choose (as is repeatedly
emphasized as a critical provision in these Guidelines).
RPU3. In collaboration with affected communities, perform Job Impact
Assessments early and often throughout AI system implementation and use
Run opportunity and risk analyses early and often across AI implementation and use, using
the data available at each stage. Update as more data becomes available (for example, as
objectives are identified, systems are procured, implementation is completed, and new
applications arise). Whenever applicable, we suggest using AI system implementation
and use choices to maximize the presence of signals of opportunity and minimize the
presence of signals of risk.
Solicit the input of workers that stand to be affectedU
and a multi-disciplinary set of inde
pendent experts when assessing the presence of opportunity and risk signals. Make sure to
compensate external contributors for their participation in the assessment of the AI system.
Please note that the analysis of opportunity and risk signals suggested here is different
from red team analysis suggested in RPU15. The former identifies risks and opportunities
created by an AI system working perfectly as intended. The latter identifies possible harms
if the AI system in question malfunctions or is misused.
RPU4. In collaboration with affected communities, develop mitigation
strategies for identified risks
In alignment with UN Guiding Principles for Business and Human Rights, a mitigation
strategy should be developed for each risk identified, prioritizing the risks primarily by
severity of potential impact and secondarily by its likelihood. Severity and likelihood of
potential impact are determined on a case-by-case basis.V
Mitigation strategies can range from eliminating the risk or reducing the severity of
potential impact to ensuring access to remedy or compensation for affected groups.
V An algorithm described
here is very useful for
determining the severity
of potential quanti-
tative impacts (such as
impacts on wages and
employment), especially
in cases with limited
uncertainty around the
future uses of the AI
system being assessed.
U It is frequently the
case that workers who
stand to be affected by
the introduction of an
AI system include not
only workers directly
employed by the organi-
zation introducing AI in
its own operations, but
a wider set of current or
potential labor market
participants. Therefore
it is important that not
only incumbent workers
are given the agency to
participate in job impact
assessment and risk
mitigation strategy devel-
opment.

PARTNERSHIP ON AI
36
If effective mitigation strategies for a given risk are not available, this should be considered
a strong argument in favor of meaningful changes in the development plans of an AI
system, especially if it is expected to affect vulnerable groups.
Engaging workers and external experts as needed in the creation of mitigation strategies
is critical to ensure important considerations are not being missed. It is especially critical
to engage with representatives of communities that stand to be affected. Please ensure
that everyone engaged in consultations around assessing risks and developing mitigation
strategies is adequately compensated.
RPU5. Create and use robust and substantive mechanisms for worker agency
in identifying needs, selecting AI vendors and systems, and implementing
them in the workplace
Workers who will use or be affected by AI hold unique perspectives on important needs
and opportunities in their roles. They also possess particular insight into how AI systems
could create harm in their workplaces. To ensure AI systems foster shared prosperity, these
workers should be included and afforded agency in the AI procurement, implementation,
and use process from start to finish.66
Workers must be properly equipped with knowledge of potential product functions,
capabilities, and limitations, so that they can draw meaningful connections to their
role-based knowledge (see RPU13 for more information). Additionally, care must be taken
to create a shared vocabulary on the team, so that technical terms or jargon do not
unintentionally obscure or mislead. Workers must also be given genuine decision-making
power in the process, allowing them to shape use (such as new workflows or job design)
and be taken seriously on the need to end a project if they identify unacceptable harms
that cannot be resolved.
RPU6. Ensure AI systems are used in environments with high levels of worker
protections and decision-making power
AI systems are less likely to cause harm in environments with:
• High levels of legal protection, monitoring, and enforcement for workers’ rights (such
as those related to health and safety or freedom to organize)
• High levels of worker voice and negotiating ability (due to strong protections for worker
voice or high demand for workers’ comparatively scarce skills), especially those where
workers have meaningful input into decisions regarding the introduction of new
technologies
These factors encourage worker-centric AI design. Workers in such environments also
possess a higher ability to limit harms from AI systems (such as changing elements of an
implementation or rejecting the use of the technology as needed), including harms outside
direct legal protections. This should not, however, be treated as a failsafe for harmful
technologies: other practices in this list should also be followed to reduce risk to workers.
Workers who
will use or be
affected by AI
hold unique
perspectives
on important
needs and
opportunities
in their roles.

PARTNERSHIP ON AI
37
RPU7. Source data enrichment labor responsibly
Key requirements for the responsible sourcing of data enrichment services (such as, data
annotation and real-time human verification of algorithmic predictions) include:
• Always paying data enrichment workers above the local living wage
• Providing clear, tested instructions for data enrichment tasks
• Equipping workers with simple and effective mechanisms for reporting issues, asking
questions, and providing feedback on the instructions or task design
In collaboration with our Partners, PAI has developed a library of practitioner resources for
responsible data enrichment sourcing.
RPU8. Ensure workplace AI systems are not discriminatory
In general, AI systems frequently reproduce or deepen discriminatory patterns in society,
including ones related to race, class, age, and disability. Specific workplace systems
have shown a propensity for the same. Careful vetting and use is needed to ensure any AI
systems affecting workers or the economy do not create discriminatory results.
WHEN IDENTIFYING NEEDS, PROCURING, AND IMPLEMENTING AI SYSTEMS
RPU9. Procure AI systems that align with worker needs and preferences
AI systems welcomed by workers largely fall into three overarching categories:
• Systems that directly improve some element of job quality
• Systems that assist workers to achieve higher performance on their core tasks
• Systems that eliminate undesirable non-core tasks (See OS2, OS9, RS1, and RS2 for
additional detail)
Starting with one of these objectives in mind and creating robust participation
mechanisms for workers throughout the design and implementation process is likely to
result in win-win-wins for AI creators, employers who implement AI, and the workers who
use or are affected by them.
RPU10. Staff and train sufficient internal or contracted expertise to properly
vet AI systems and ensure responsible implementation
As discussed throughout, AI systems raise substantial concerns about the risks of their
adoption in workplace settings. To understand and address these risks, experts are
needed to vet and implement AI systems. In addition to technical experts, this includes
sociotechnical experts capable of performing the Job Impact Assessment described above
to the level of granularity necessary to fully identify and mitigate risks of a specific system
in a given workplace.
The importance of this practice increases with AI system customization or integration. In
situations where systems are developed by organizations who follow the Shared Prosperity
AI systems
frequently
reproduce
or deepen
discriminatory
patterns in
society.

Guidelines for AI and Shared Prosperity - Tools for improving AI’s impact on jobs

Guidelines for AI and Shared Prosperity - Tools for improving AI’s impact on jobs

Related slideshows

Recommended for you

Recommended for you

Recommended for you

Recommended for you

Recommended for you

Recommended for you

Recommended for you

Recommended for you

Recommended for you

Recommended for you

Recommended for you

Recommended for you

More Related Content

Similar to Guidelines for AI and Shared Prosperity - Tools for improving AI’s impact on jobs

Similar to Guidelines for AI and Shared Prosperity - Tools for improving AI’s impact on jobs (20)

Recently uploaded

Recently uploaded (20)

Guidelines for AI and Shared Prosperity - Tools for improving AI’s impact on jobs