Designing AI for Humanity at dmi:Design Leadership Conference in Boston
- 1. Designing AI for Humanity
Carol Smith @carologic
dmi:Design Leadership Conference #DLC18. October 9, 2018
This work is licensed under a Creative
Commons Attribution-NonCommercial
4.0 International License except where
noted otherwise.
- 3. AI is as imperfect as the
humans making it
- 8. Designing AI for Humanity / @carologic
Writer made data/content
• Creates backstories and scripts
• Environmental design
• Data: Who, what, how,
and when of experience
- 9. Designing AI for Humanity / @carologic
Why should we care?
• What are his bias’?
• How did this affect the experience?
• Does it matter?
- 12. Designing AI for Humanity / @carologic
Scientists
• Triage system -
“Step into analysis”
• Ever leave?
• Imagine negative
consequences?
- 14. Designing AI for Humanity / @carologic
New programming
• Ford/Arnold as
programmer
• No context for staff
• Should have provided
information
- 17. Designing AI for Humanity / @carologic
To engender trust, provide transparency
• Data
• Training/programming of system
• Rationale/bias/logic
- 19. AI is present when computers/machines
– Exhibit intelligence
– Perceive their environment
– Take actions/make decision
to maximize chance of success at a goal
Our Road to Self-Driving Vehicles | Uber ATG
https://youtu.be/27OuOCeZmwI
- 20. Designing AI for Humanity / @carologic
AI/Cognitive computers are
• Algorithms
• Know ONLY what you teach
• Control ONLY what given control of
• Aware of nuances and can continue to learn
- 23. Designing AI for Humanity / @carologic
Taxonomies and Ontologies coming to life
(NOT like humans learn)
Photo: https://commons.wikimedia.org/wiki/File:Baby_Boy_Oliver.jpg
- 25. Designing AI for Humanity / @carologic
Like Any Good Design
• Understand problem deeply
• Build right AI system
• Different problems require different systems
- 30. Designing AI for Humanity / @carologic
Who will use the system and why?
• What are their goals?
• What problems are they trying to solve?
• Are they working independently?
- 31. Designing AI for Humanity / @carologic
What do users need to know?
What
changed?
What are
outliers?What comes
next?
What is
unexpected?
What is
new?
How can I tell
what changed?
Increase/decrease
in frequency?
Are my assumptions
validated?
- 32. Designing AI for Humanity / @carologic
Anticipate changes with AI system
• Scope/intention?
• Improvements?
• Better or faster?
- 33. Designing AI for Humanity / @carologic
Unintended consequences?
• Understand user’s fears
• Address them to protect users
- 35. Designing AI for Humanity / @carologic
Data Source
• In existence?
• Available?
• High quantity?
• High quality?
Photo by sunlightfoundation
https://www.flickr.com/photos/sunlightfoundation/2385174105
- 36. Designing AI for Humanity / @carologic
Number Five “Needs Input”
Short Circuit (1986 film)
Ally Sheedy and Number Five (Tim Blaney)
https://en.wikipedia.org/wiki/Short_Circuit_(1986_film)
- 37. Designing AI for Humanity / @carologic
Curation
• Source and Bias?
• Who is creating/curating collection?
– Respected experts
– Diverse
- 40. “We often have
no way of knowing
when and why people
are biased.”
- Sandra Wachter
Q&A: Should artificial intelligence be legally required to explain itself?
By Matthew Hutson, May. 31, 2017. Interview with Sandra Wachter, data ethics researcher at Univ. of Oxford and Alan Turing Institute.
http://www.sciencemag.org/news/2017/05/qa-should-artificial-intelligence-be-legally-required-explain-itself
- 41. Humans teach what we feel is important… teach them to share our values.
Grady Booch, Scientist, philosopher, IBM’er https://www.ted.com/talks/grady_booch_don_t_fear_superintelligence
- 42. Designing AI for Humanity / @carologic
Ready for Use
• Experts review results
- 43. Designing AI for Humanity / @carologic
Data source
• Few techs
– Detailed, digital notes
– Prefer using chemicals
• Most techs
– Rough written notes
– Prefer “all natural”
treatments
Neither are wrong.
Limited data created a bias.
- 44. Designing AI for Humanity / @carologic
Humans required to teach and monitor AI
• Water
• Prune/Shape
• Cull
- 47. Designing AI for Humanity / @carologic
Experts to train system
• Vetting?
• Availability?
• Process?
• Maintain quality?
- 48. Accuracy
You’re cloning
a colleague
no-bake cookies photo by Melissa Hillier - recipe blogged at jonahbonah.com
https://www.flickr.com/photos/77423179@N02/7848109610/in/photolist-cXvByE-x51nF-218WBFr-Z78P3y-6HKkBs-MMkWFT-6wKNxR-7jmLft-6kDRm3-6kDSsN-6kDUvY-6wRRoV-7cYgGN-6kEnjs-6kEaKh-3kHP9P-6kEo6N-6kEAg9-giXGrA-N67c4-5X mXw1-
cgk3ow-6kzJog-6kA5oZ-aYqEpT-MMkVVV-7aQLnM-ecL6fm-6kEd67-5ykEkC-2bsTnp3-dCh7J9-T4tu4i-8HdYNJ-73SMVr-6uwEGT-6kE34b-MMkEqr-6kEFws-6kEjVu-25rwHBc-6kA42g-6kzTi4-T36Moj-7Bx3rf-7vPVhb-6YNEHC-amariC-neddpV-ZNpJHE
- 49. Designing AI for Humanity / @carologic
Priority of accuracy across industries
Higher Priority
90-99%+
Lower Priority
60-89% accuracy is acceptable
Financial
Ecommerce
- 50. Designing AI for Humanity / @carologic
Responsible,
Intentional
Design
http://www.flickr.com/photos/rockyvi/6451635085/sizes/m/in/photolist-aQ7jkF/
Some rights reserved by Rocky VI - http://www.flickr.com/photos/rockyvi/
License: http://creativecommons.org/licenses/by-nc-nd/2.0/
- 51. Designing AI for Humanity / @carologic
Make it your business to keep people safe
• Monitor system
• Identify warning signs
- 52. Designing AI for Humanity / @carologic
Privacy
• What must a user reveal?
• Who owns the data?
• Life expectancy of data?
• Keep data safe
PAPA (Privacy, Accuracy, Property, Accessibility)
Ethical Issues in IS by Richard Mason.
https://www.gdrc.org/info-design/4-ethics.html
- 53. Designing AI for Humanity / @carologic
Plan for unintended consequences
• Scenarios – not every one
• Focus on worst situations:
– What happens when it becomes a Nazi?
– What happens when it does XYZ?
- 54. Designing AI for Humanity / @carologic
What will you do?
• What is the method for
“turning it off”?
• Who is notified
immediately?
• Unintended consequences
of turning it off?
Google’s new tensor processing units:
https://www.nytimes.com/2018/02/12/technology/google-artificial-intelligence-chips.html
- 55. Designing AI for Humanity / @carologic
Secure back doors and brakes
• “If it’s not usable, it’s not secure.”
– Jared Spool, IAS17
• “Ensure humans can unplug the machines”
– Grady Booch, Ted Talk
Unintuitive and Insecure: Fixing the Failures of Authentication, Jared Spool, IA Summit 2017
Grady Booch, Scientist, philosopher, IBM’er
https://www.ted.com/talks/grady_booch_don_t_fear_superintelligence
- 56. Don’t be ableist
How People with Disabilities Use the Web: Overview https://www.w3.org/WAI/intro/people-use-web /
- 57. Designing AI for Humanity / @carologic
Communicating
About
The System
Strong Bad Email #45 – Techno - Strong Bad makes a techno song.
https://youtu.be/JwZwkk7q25I Homestarrunnerdotcom Published on Mar 31, 2009
- 58. Designing AI for Humanity / @carologic
Communicate Responsibly
• How is communication about the AI handled?
• How do you report issues?
• To whom?
- 59. Designing AI for Humanity / @carologic
Potential Bias
• Show awareness
• Acknowledge issues
• Overcommunicate
- 60. Designing AI for Humanity / @carologic
To engender trust, provide transparency
• Who made the data?
• Who trained/programmed the system?
– When updated?
• Why system providing data it is?
- 61. Designing AI for Humanity / @carologic
Displaying and comparing information
• AI generated content vs. other
• Confidence
- 62. Designing AI for Humanity / @carologic
Crowdsourcing Quality
• Show examples
– Potential signs of building bias
• How can a user report?
- 66. Trolley Problem
Trolley Car 36, Rockford, Illinois https://www.rockfordparkdistrict.org/trolley
Does the Trolley Problem Have a Problem? What if your answer to an absurd hypothetical question had no bearing on how you behaved in real life?
By Daniel Engber. Slate.com. June 18, 2018. Image of anxious hypothetical trolley car lever operator by Lisa Larson-Walker
https://slate.com/technology/2018/06/psychologys-trolley-problem-might-have-a-problem.html
- 68. Designing AI for Humanity / @carologic
Create a code of conduct/ethics
• What do you value?
• How helping people?
• What lines won’t your
AI cross?
• How will you track your
progress?
Inspired by “3 guiding principles for ethical AI, from IBM CEO Ginni Rometty”
by Alison DeNisco. January 17, 2017, Tech Republic
http://www.techrepublic.com/article/3-guiding-principles-for-ethical-ai-from-ibm-ceo-ginni-rometty/
- 69. Designing AI for Humanity / @carologic
Guidance
• UXPA Code
of professional
conduct
ACM Code of Ethics
and Professional
Conduct
- 70. Designing AI for Humanity / @carologic
Take Responsibility
• Humans in control
• Support humans
– Social consequences
– Job displacement
How to Keep Your AI from Turning into a Racist Monster
By Megan Garcia. https://www.wired.com/2017/02/keep-ai-turning-racist-monster/
- 71. Designing AI for Humanity / @carologic
Hire/work with people affected by bias
- 72. Designing AI for Humanity / @carologic
Explore AI - Don’t fear AI
• Try out tools (appendix and notes)
• Pair with others
• Teach others about AI
- 73. Designing AI for Humanity / @carologic
Create ethical, transparent and fair AI
• Intentional design
• Less-biased content
• Communicate
responsibly about AI
Toward ethical, transparent and fair AI/ML: a critical reading list
By Eirini Malliaraki, Feb 19 via tweet from @robmccargow https://medium.com/@eirinimalliaraki/toward-ethical-
transparent-and-fair-ai-ml-a-critical-reading-list-d950e70a70ea
- 74. Designing AI for Humanity / @carologic
Continue the conversation…
LinkedIn – CarolJSmith
Twitter - @Carologic
Slideshare – carologic
- 75. Designing AI for Humanity / @carologic
Appendix
Additional Information and Resources
- 76. Designing AI for Humanity / @carologic
Barriers to Data
• Literacy and awareness
• Connection to internet – economics and location.
• Access to pertinent data
• Fear of AI
Ethical Issues in IS by Richard Mason
https://www.gdrc.org/info-design/4-ethics.html
- 78. Designing AI for Humanity / @carologic
Supervised Learning
• Specialists involved in content creation and training
• Programmer and/or GUI
• Most common
Artificial Intelligence Demystified by. Rahul December 23, 2016. Analytics Vidhya
https://www.analyticsvidhya.com/blog/2016/12/artificial-intelligence-demystified/
- 79. Designing AI for Humanity / @carologic
Annotating Content
Image created by Angela Swindell, Visual Designer, IBM
- 80. Designing AI for Humanity / @carologic
Supervised Machine Learning - GUI
Watson Knowledge Studio, Supervised Machine Learning:
https://www.ibm.com/us-en/marketplace/supervised-machine-learning
- 81. Designing AI for Humanity / @carologic
Types of Machine Learning
• Unsupervised learning
– Machine defines patterns
• Reinforced learning
– Games – rules and rewards
Artificial Intelligence Demystified by Rahul
rahul@upxacademy.com December 23, 2016. Analytics Vidhya
https://www.analyticsvidhya.com/blog/2016/12/artificial-intelligence-demystified/
- 82. Designing AI for Humanity / @carologic
Pattern recognition
• Natural Language
Processing
• Image Analysis
IBM Watson https://twitter.com/IBMWatson/status/844545761740292096
- 83. Designing AI for Humanity / @carologic
Deep Learning
• Classify objects based
on features
• Can be applied
to other types of AI
Toward ethical, transparent and fair AI/ML: a critical reading list
By Eirini Malliaraki, Feb 19 via tweet from @robmccargow https://medium.com/@eirinimalliaraki/toward-ethical-
transparent-and-fair-ai-ml-a-critical-reading-list-d950e70a70ea
- 84. Designing AI for Humanity / @carologic
AI Tools
• A list of artificial intelligence tools you can use today — for businesses, by Liam
Hanel, July 11, 2017 on Lyr.AI
https://lyr.ai/a-list-of-artificial-intelligence-tools-you-can-use-today%E2%80%8A-
%E2%80%8Afor-businesses/ and https://medium.com/imlyra/a-list-of-artificial-
intelligence-tools-you-can-use-today-for-personal-use-1-3-7f1b60b6c94f
• Best AI and machine learning tools for developers, By Christina Mercer, Sep 26,
2017 in Techworld from IDG https://www.techworld.com/picture-gallery/apps-
wearables/best-ai-machine-learning-tools-for-developers-3657996/
• 15 Top Open Source Artificial Intelligence Tools by Cynthia Harvey, September
12, 2016 on Datamation https://www.datamation.com/open-source/slideshows/15-
top-open-source-artificial-intelligence-tools.html
• IBM Watson Developer Tools (free trials):
https://console.ng.bluemix.net/catalog/?category=watson
- 85. Designing AI for Humanity / @carologic
Want to Know More?
• The Rise Of Artificial Intelligence As A Service In The Public
Cloud
Rise Of Artificial Intelligence As A Service In The Public Cloud by Janakiram MSV , Forbes Article:
https://www.forbes.com/sites/janakirammsv/2018/02/22/the-rise-of-artificial-intelligence-as-a-service-in-the-public-cloud/#11aa85a8198e
Courses at http://www.fast.ai/
- 86. Designing AI for Humanity / @carologic
10 Major Milestones in the History of AI
https://www.analyticsvidhya.com/blog/2016/12/artificial-intelligence-demystified/
- 87. Designing AI for Humanity / @carologic
Additional Resources
• “How IBM is Competing with Google in AI.” The Information. https://www.theinformation.com/how-ibm-is-
competing-with-google-in-ai?eu=2zIDMNYNjDp7KqL4YqAXXA
• “The business case for augmented intelligence” https://medium.com/cognitivebusiness/the-business-case-for-
augmented-intelligence-36afa64cd675
• “Comparison of machine learning methods applied to birdsong element classification” by David Nicholson.
Proceedings of the 15th Python in Science Conference (SCIPY 2016).
http://conference.scipy.org/proceedings/scipy2016/pdfs/david_nicholson.pdf
• “Staples’ “Easy Button” Comes to Life with IBM Watson” in Business Wire, October 25, 2016.
http://www.businesswire.com/news/home/20161025006273/en/Staples%E2%80%99-%E2%80%9CEasy-
Button%E2%80%9D-Life-IBM-Watson
• “How Staples Is Making Its Easy Button Even Easier With A.I.” by Chris Cancialosi, Forbes.
https://www.forbes.com/sites/chriscancialosi/2016/12/13/how-staples-is-making-its-easy-button-even-easier-
with-a-i/#4ae66e8359ef
• “Inside Intel: The Race for Faster Machine Learning”
http://www.intel.com/content/www/us/en/analytics/machine-learning/the-race-for-faster-machine-learning.html
- 88. Designing AI for Humanity / @carologic
More Resources
• “Update: Why this week’s man-versus-machine Go match doesn’t matter (and what does)” by Dana
Mackenzie. Science Magazine. Mar. 15, 2016 http://www.sciencemag.org/news/2016/03/update-why-week-s-
man-versus-machine-go-match-doesn-t-matter-and-what-does
• “For IBM’s CTO for Watson, not a lot of value in replicating the human mind in a computer.” by Frederic
Lardinois (@fredericl), TechCrunch, Posted Feb 27, 2017. https://techcrunch.com/2017/02/27/for-ibms-cto-
for-watson-not-a-lot-of-value-in-replicating-the-human-mind-in-a-computer/
• “Google and IBM: We Want Artificial Intelligence to Help You, Not Replace You” Most Powerful Women by
Michelle Toh. Mar 02, 2017. Fortune. http://fortune.com/2017/03/02/google-ibm-artificial-intelligence/
• “Facebook scales back AI flagship after chatbots hit 70% f-AI-lure rate - 'The limitations of automation‘” by
Andrew Orlowski. Feb 22, 2017. The Register https://www.theregister.co.uk/2017/02/22/facebook_ai_fail/
• “Microsoft is deleting its AI chatbot's incredibly racist tweets” by Rob Price. Mar. 24, 2016. Business Insider
UK. http://www.businessinsider.com/microsoft-deletes-racist-genocidal-tweets-from-ai-chatbot-tay-2016-3
Special Thanks: Soundtrack to 'Run Lola Run', 1998 German thriller film written and directed by Tom Tykwer,
and starring Franka Potente as Lola and Moritz Bleibtreu as Manni. Soundtrack by Tykwer, Johnny Klimek, and
Reinhold Heil
- 89. Designing AI for Humanity / @carologic
Even More Resources
• “IBM’s Automated Radiologist Can Read Images and Medical Records” by Tom Simonite, February 4, 2016.
Intelligent Machines, MIT Technology Review. https://www.technologyreview.com/s/600706/ibms-automated-
radiologist-can-read-images-and-medical-records/
• “The IBM, Salesforce AI Mash-Up Could Be a Stroke of Genius” by Adam Lashinsky, Mar 07, 2017. Fortune.
http://fortune.com/2017/03/07/data-sheet-ibm-salesforce/
• "Google can now tell you're not a robot with just one click" by Andy Greenberg. Dec. 3, 2014. Security: Wired.
https://www.wired.com/2014/12/google-one-click-recaptcha/
• “Essentials of Machine Learning Algorithms (with Python and R Codes)” by Sunil Ray, August 10, 2015.
Analytics Vidhya. https://www.analyticsvidhya.com/blog/2015/08/common-machine-learning-algorithms/
• IBM on Machine Learning https://www.ibm.com/analytics/us/en/technology/machine-learning/
• “At Davos, IBM CEO Ginni Rometty Downplays Fears of a Robot Takeover” by Claire Zillman, Jan 18, 2017.
Fortune. http://fortune.com/2017/01/18/ibm-ceo-ginni-rometty-ai-davos/
• “Google and IBM: We Want Artificial Intelligence to Help You, Not Replace You” by Michelle Toh. Mar 02,
2017. Fortune. http://fortune.com/2017/03/02/google-ibm-artificial-intelligence/
- 90. Designing AI for Humanity / @carologic
Yes, even more resources
• Video: “IBM Watson Knowledge Studio: Teach Watson about your unstructured data”
https://www.youtube.com/watch?v=caIdJjtvX1s&t=6s
• “The optimist’s guide to the robot apocalypse” by Sarah Kessler, @sarahfkessler. March 09, 2017. QZ.
https://qz.com/904285/the-optimists-guide-to-the-robot-apocalypse/
• “AI Influencers 2017: Top 30 people in AI you should follow on Twitter" by Trips Reddy @tripsy, Senior
Content Manager, IBM Watson . February 10, 2017 https://www.ibm.com/blogs/watson/2017/02/ai-
influencers-2017-top-25-people-ai-follow-twitter/
• “3 guiding principles for ethical AI, from IBM CEO Ginni Rometty” by Alison DeNisco. January 17, 2017, Tech
Republic http://www.techrepublic.com/article/3-guiding-principles-for-ethical-ai-from-ibm-ceo-ginni-rometty/
• "Transparency and Trust in the Cognitive Era" January 17, 2017 Written by: IBM THINK Blog
https://www.ibm.com/blogs/think/2017/01/ibm-cognitive-principles/
• "Ethics and Artificial Intelligence: The Moral Compass of a Machine“ by Kris Hammond, April 13, 2016.
Recode. http://www.recode.net/2016/4/13/11644890/ethics-and-artificial-intelligence-the-moral-compass-of-a-
machine
- 91. Designing AI for Humanity / @carologic
Last bit: I promise
• "The importance of human innovation in A.I. ethics" by John C. Havens. Oct. 03, 2015
http://mashable.com/2015/10/03/ethics-artificial-intelligence/#yljsShvAFsqy
• "Me, Myself and AI" Fjordnet Limited 2017 - Accenture Digital.
https://trends.fjordnet.com/trends/me-myself-ai
• "Testing AI concepts in user research" By Chris Butler, Mar 2, 2017. https://uxdesign.cc/testing-ai-
concepts-in-user-research-b742a9a92e55#.58jtc7nzo
• "CMU prof says computers that can 'see' soon will permeate our lives“ by Aaron Aupperlee. March
16, 2017. http://triblive.com/news/adminpage/12080408-74/cmu-prof-says-computers-that-can-
see-soon-will-permeate-our-lives
• “The business case for augmented intelligence” by Nancy Pearson, VP Marketing, IBM Cognitive.
https://medium.com/cognitivebusiness/the-business-case-for-augmented-intelligence-
36afa64cd675#.qqzvunakw
- 92. Designing AI for Humanity / @carologic
Definition: Artificial Intelligence
• Artificial intelligence (AI) is intelligence exhibited by machines.
• In computer science, an ideal "intelligent" machine is a flexible rational agent that
perceives its environment and takes actions that maximize its chance of success
at some goal.[1] Colloquially, the term "artificial intelligence" is applied when a
machine mimics "cognitive" functions that humans associate with other human
minds, such as "learning" and "problem solving".[2]
• Capabilities currently classified as AI include successfully understanding human
speech,[4] competing at a high level in strategic game systems (such as Chess
and Go[5]), self-driving cars, and interpreting complex data.
Wikipedia: https://en.wikipedia.org/wiki/Artificial_intelligence#cite_note-Intelligent_agents-1
- 93. Designing AI for Humanity / @carologic
Definition: The Singularity
• If research into Strong AI produced sufficiently intelligent software, it might be able to reprogram
and improve itself. The improved software would be even better at improving itself, leading to
recursive self-improvement.[245] The new intelligence could thus increase exponentially and
dramatically surpass humans. Science fiction writer Vernor Vinge named this scenario
"singularity".[246] Technological singularity is when accelerating progress in technologies will
cause a runaway effect wherein artificial intelligence will exceed human intellectual capacity and
control, thus radically changing or even ending civilization. Because the capabilities of such an
intelligence may be impossible to comprehend, the technological singularity is an occurrence
beyond which events are unpredictable or even unfathomable.[246]
• Ray Kurzweil has used Moore's law (which describes the relentless exponential improvement in
digital technology) to calculate that desktop computers will have the same processing power as
human brains by the year 2029, and predicts that the singularity will occur in 2045.[246]
Wikipedia: https://en.wikipedia.org/wiki/Artificial_intelligence#cite_note-Intelligent_agents-1
- 94. Designing AI for Humanity / @carologic
Definition: Machine Learning
• Ability for system to take basic knowledge (does not mean simple or non-complex)
and apply that knowledge to new data
• Raises ability to discover new information. Find unknowns in data.
• https://en.wikipedia.org/wiki/Machine_learning
More Definitions:
• Algorithm: a process or set of rules to be followed in calculations or other problem-
solving operations, especially by a computer.
https://en.wikipedia.org/wiki/Algorithm
• Natural Language Processing (NLP):
https://en.wikipedia.org/wiki/Natural_language_processing