Accounting Value Effects for Responsible Networking
•
1 like•283 views
The document discusses approaches to achieving a "responsible internet" through redistributing control and monitoring abilities to users. It identifies two issues with the existing "Responsible Internet" proposal: 1) A "responsibility gap" exists because low-level programmability for users does not address their ability to foresee outcomes or assess actions according to values. 2) The power relationships between users, network operators, and governance bodies are not partially programmable, requiring models of how the world functions and what is valuable, as well as mechanisms for consolidating norms from multiple sources into policies at different levels.
Report
Share
Accounting Value Effects for Responsible Networking
- 1. Accounting Value Effects for Responsible Networking Giovanni Sileno g.sileno@uva.nl Complex Cyber Infrastructure (CCI) Paola Grosso p.grosso@uva.nl Multiscale Networked Systems (MNS) University of Amsterdam ACM SIGCOMM 2021 Workshop on Technologies, Applications, and Uses of a Responsible Internet (TAURIN 2021), 23 August 2021
- 2. Why responsible computing? the more we delegate activities to machines,
- 3. Why responsible computing? the more we delegate activities to machines, the more we need machines to behave "responsibly"
- 4. Why responsible computing? the more we delegate activities to machines, the more we need machines to behave "responsibly" Today, focus on this issue is mostly on ML-based AI, however, any technology -- because it has effects on the world -- should fulfill the same requirement.
- 5. Why responsible computing? the more we delegate activities to machines, the more we need machines to behave "responsibly" Today, focus on this issue is mostly on ML-based AI, however, any technology -- because it has effects on the world -- should fulfill the same requirement. Including the Internet and networking!
- 6. Looking at application level, data-sharing has practical effects because having access to relevant information has value for agents!
- 7. Looking at application level, data-sharing has practical effects because having access to relevant information has value for agents! people, organizations, systems which act to achieve certain purposes
- 8. + - no exchange exchange enabled/allowed + - exchange disabled/disallowed Looking at application level, data-sharing has practical effects because having access to relevant information has value for agents!
- 9. + - no exchange exchange enabled/allowed + - exchange disabled/disallowed ability (power) dimension e.g. Internet with new forms of knowledge sharing, aggregation, making business, etc. ● technologies provide new abilities Looking at application level, data-sharing has practical effects because having access to relevant information has value for agents!
- 10. + - no exchange exchange enabled/allowed + - exchange disabled/disallowed ability (power) dimension permission dimension e.g. norms on privacy, data regulation (e.g. GDPR), competition laws, etc. ● societies introduce checks & balances Looking at application level, data-sharing has practical effects because having access to relevant information has value for agents!
- 11. + - no exchange exchange enabled/allowed + - exchange disabled/disallowed e.g. norms on privacy, data regulation (e.g. GDPR), competition laws, etc. ● societies introduce checks & balances How these checks and balances are reflected at infrastructural level? Looking at application level, data-sharing has practical effects because having access to relevant information has value for agents!
- 12. Data transmission as “logistic” task X Y How to transport data from node X to node Y? no path specification (only end-point nodes)
- 13. X Y ● Enabling transmission from X to Y requires the network to provide some form of routing services. partial path specification (with intermediary steps) full specification (all intermediary steps as “primitive” actions) X Y X Y no path specification (only end-point nodes) How to transport data from node X to node Y? Data transmission as “logistic” task
- 14. X Y ● Enabling transmission from X to Y requires the network to provide some form of routing services. partial path specification (with intermediary steps) full specification (all intermediary steps as “primitive” actions) X Y X Y no path specification (only end-point nodes) For inter-domain routing, network operators typically rely on BGP policies and community tags. How to transport data from node X to node Y? Data transmission as “logistic” task
- 15. Data transmission as “logistic” task X Y X Y Z abusive transmission (e.g. data-leak) X inefficient or expensive transmission (in terms of resources load/usage, environmental impact, etc.) Main issues possibly occurring at network level: faulty transmission
- 16. X Y X Y Z X Main issues possibly occurring at network level: Intuitively, a ‘responsible’ networking should reduce these issues. faulty transmission abusive transmission (e.g. data-leak) inefficient or expensive transmission (in terms of resources load/usage, environmental impact, etc.) Data transmission as “logistic” task
- 17. X Y X Y Z X Main issues possibly occurring at network level: Intuitively, a ‘responsible’ networking should reduce these issues. but... who defines what is faulty, abusive, expensive? who monitors? who prevents (predicts) or reacts to failures? faulty transmission abusive transmission (e.g. data-leak) inefficient or expensive transmission (in terms of resources load/usage, environmental impact, etc.) Data transmission as “logistic” task
- 18. X Y X Y Z X Main issues possibly occurring at network level: Intuitively, a ‘responsible’ networking should reduce these issues. but... who defines what is faulty, abusive, expensive? who monitors? who prevents (predicts) or reacts to failures? faulty transmission abusive transmission (e.g. data-leak) inefficient or expensive transmission (in terms of resources load/usage, environmental impact, etc.) “responsibility” is a matter of social coordination policy Data transmission as “logistic” task
- 19. Internet social structure Three main roles can be recognized around Internet’s activities: ● users (applications, software agents, etc.) ● network operators ● governance bodies users governance bodies network operators data logistic tasks requirements
- 20. Responsible Internet social structure users governance bodies network operators data logistic tasks requirements POL NIP NCP Cristian Hesselman, Paola Grosso, Ralph Holz, Fernando Kuipers, Janet Hui Xue, Mattijs Jonker, Joeri de Ruiter, Anna Sperotto, Roland van Rijswijk-Deij, Giovane C.M. Moura, Aiko Pras, and Cees de Laat. A Responsible Internet to Increase Trust in the Digital World. Journal of Network and Systems Management 28, 4 (2020), 882–922. The Responsible Internet proposal (Hesselman et al., 2020) essentially envisions to redistribute control and monitoring abilities to users, supported by regulations issued by relevant societal stakeholders.
- 21. Responsible Internet social structure users governance bodies network operators data logistic tasks requirements POL NIP NCP Cristian Hesselman, Paola Grosso, Ralph Holz, Fernando Kuipers, Janet Hui Xue, Mattijs Jonker, Joeri de Ruiter, Anna Sperotto, Roland van Rijswijk-Deij, Giovane C.M. Moura, Aiko Pras, and Cees de Laat. A Responsible Internet to Increase Trust in the Digital World. Journal of Network and Systems Management 28, 4 (2020), 882–922. open-source programmable networks (e.g. for routing, via BGP policies) large-scale measuring techniques The Responsible Internet proposal (Hesselman et al., 2020) essentially envisions to redistribute control and monitoring abilities to users, supported by regulations issued by relevant societal stakeholders.
- 22. Responsible Internet social structure users governance bodies network operators data logistic tasks requirements POL NIP NCP Cristian Hesselman, Paola Grosso, Ralph Holz, Fernando Kuipers, Janet Hui Xue, Mattijs Jonker, Joeri de Ruiter, Anna Sperotto, Roland van Rijswijk-Deij, Giovane C.M. Moura, Aiko Pras, and Cees de Laat. A Responsible Internet to Increase Trust in the Digital World. Journal of Network and Systems Management 28, 4 (2020), 882–922. open-source programmable networks (e.g. for routing, via BGP policies) informed policy-making The Responsible Internet proposal (Hesselman et al., 2020) essentially envisions to redistribute control and monitoring abilities to users, supported by regulations issued by relevant societal stakeholders. large-scale measuring techniques
- 23. Responsible Internet social structure users governance bodies network operators data logistic tasks requirements POL NIP NCP Cristian Hesselman, Paola Grosso, Ralph Holz, Fernando Kuipers, Janet Hui Xue, Mattijs Jonker, Joeri de Ruiter, Anna Sperotto, Roland van Rijswijk-Deij, Giovane C.M. Moura, Aiko Pras, and Cees de Laat. A Responsible Internet to Increase Trust in the Digital World. Journal of Network and Systems Management 28, 4 (2020), 882–922. open-source programmable networks (e.g. for routing, via BGP policies) informed policy-making IS THIS COMPLETE? The Responsible Internet proposal (Hesselman et al., 2020) essentially envisions to redistribute control and monitoring abilities to users, supported by regulations issued by relevant societal stakeholders. large-scale measuring techniques
- 24. Our paper raises two critiques ● RESPONSIBILITY GAP: Low-level programmability (e.g. for routing, via BGP policies) is not sufficient to capture and behaviourally operationalize the value structure of users. ● REGULATIVE CONTINGENCY: Power-relationships between roles should not be hard-coded (that is, should be partially programmable).
- 25. Requirements for responsibility has the ability to control its own behaviour has the ability to foresee the associated outcomes has the ability to assess actions according to a certain preference/value structure An agent has (agentive) responsibility if it: Sileno, G., Boer, A., Gordon, G., Rieder, B., Like Circles in the Water: Responsibility as a System-Level Function. Proceedings of 3rd XAILA workshop: Explainable and Responsible AI and Law, in conjunction with JURIX 2020 (2020)
- 26. Requirements for responsibility has the ability to control its own behaviour has the ability to foresee the associated outcomes has the ability to assess actions according to a certain preference/value structure An agent has (agentive) responsibility if it: Sileno, G., Boer, A., Gordon, G., Rieder, B., Like Circles in the Water: Responsibility as a System-Level Function. Proceedings of 3rd XAILA workshop: Explainable and Responsible AI and Law, in conjunction with JURIX 2020 (2020) necessary e.g. to identify wrong behaviour
- 27. has the ability to control its own behaviour has the ability to foresee the associated outcomes has the ability to assess actions according to a certain preference/value structure An agent has (agentive) responsibility if it: necessary e.g. to inhibit wrong behaviour Sileno, G., Boer, A., Gordon, G., Rieder, B., Like Circles in the Water: Responsibility as a System-Level Function. Proceedings of 3rd XAILA workshop: Explainable and Responsible AI and Law, in conjunction with JURIX 2020 (2020) Requirements for responsibility necessary e.g. to identify wrong behaviour
- 28. Responsibility gap has the ability to control its own behaviour has the ability to foresee the associated outcomes has the ability to assess actions according to a certain preference/value structure In the Responsible Internet proposal, users gain controllability by low-level programmability (via the NCP).
- 29. Responsibility gap has the ability to control its own behaviour has the ability to foresee the associated outcomes has the ability to assess actions according to a certain preference/value structure In the Responsible Internet proposal, users gain controllability by low-level programmability (via the NCP). ...but nothing is said about the two other components.
- 30. Responsibility gap has the ability to control its own behaviour has the ability to foresee the associated outcomes has the ability to assess actions according to a certain preference/value structure In the Responsible Internet proposal, users gain controllability by low-level programmability (via the NCP). ...but nothing is said about the two other components. HOW CAN WE REPAIR THIS?
- 31. has the ability to control its own behaviour has the ability to foresee the associated outcomes has the ability to assess actions according to a certain preference/value structure [1] We need a model of how the world functions. EXPECTATIONS artefact Reducing the responsibility gap
- 32. has the ability to control its own behaviour has the ability to foresee the associated outcomes has the ability to assess actions according to a certain preference/value structure [1] We need a model of how the world functions. EXPECTATIONS artefact [2] We need a model of what is valuable in the world. high-level POLICY artefact Reducing the responsibility gap
- 33. Reducing the responsibility gap has the ability to control its own behaviour has the ability to foresee the associated outcomes has the ability to assess actions according to a certain preference/value structure [1] We need a model of how the world functions. EXPECTATIONS artefact [2] We need a model of what is valuable in the world. high-level POLICY artefact norm as in normal norm as in normative
- 34. Reducing the responsibility gap has the ability to control its own behaviour has the ability to foresee the associated outcomes has the ability to assess actions according to a certain preference/value structure [1] We need a model of how the world functions. EXPECTATIONS artefact(s) [2] We need a model of what is valuable in the world. high-level POLICY artefact(s) norm as in normal norm as in normative norm pluralism [3] We need to take into account several external sources of norms
- 35. Reducing the responsibility gap has the ability to control its own behaviour has the ability to foresee the associated outcomes has the ability to assess actions according to a certain preference/value structure [1] We need a model of how the world functions. EXPECTATIONS artefact(s) [2] We need a model of what is valuable in the world. high-level POLICY artefact(s) norm as in normal norm as in normative [3] We need to take into account several external sources of norms need for “normware” technology norm pluralism
- 36. high-level policy expectations value effects synthesis situated policy low-level policy From higher-level to lower-level policies user’s policy on policies ex policy artefact import import external expectations artefact external expectations artefact external expectations artefact ex policy artefact external policy artefact CONSOLIDATION NCP OPERATIONALIZATION The paper elaborates on the functions required to pass from high-level specifications to low-level policies (e.g. BGP policies and community tags for routing)... user’s policy on expectations
- 37. Regulative contingency Users, network operators, and the various governance bodies have all legitimate interests to play a role in policy-making. Prototypical conflictual design choice: anonymity vs accountability.
- 38. Regulative contingency Users, network operators, and the various governance bodies have all legitimate interests to play a role in policy-making. Prototypical conflictual design choice: anonymity vs accountability. Governmental, public agencies are users of the infrastructure, and play a role in the infrastructure governance bodies. The Responsible Internet proposal says that POL ● should be informed by NIP and ● should drive the NCP. users governance bodies network operators data logistic tasks requirements POL NIP NCP
- 39. Regulative contingency Users, network operators, and the various governance bodies have all legitimate interests to play a role in policy-making. Prototypical conflictual design choice: anonymity vs accountability. Governmental, public agencies are users of the infrastructure, and play a role in the infrastructure governance bodies. The Responsible Internet proposal says that POL ● should be informed by NIP and ● should drive the NCP. users governance bodies network operators data logistic tasks requirements POL NIP NCP but how? to what extent regulators can intervene on users’ activity on the infrastructure?
- 40. Regulative contingency Users, network operators, and the various governance bodies have all legitimate interests to play a role in policy-making. Prototypical conflictual design choice: anonymity vs accountability. Govermental, public agencies are users of the infrastructure, and play a role in the infrastructure governance bodies. The Responsible Internet proposal says that POL ● should be informed by NIP and ● should drive the NCP. users governance bodies network operators data logistic tasks requirements POL NIP NCP but how? to what extent regulators can intervene on users’ activity on the infrastructure? There is no definitive, global solution: checks & balances vary on a local basis. ● Power-relationships between roles should not be hard-coded, but programmable.
- 41. Regulative contingency Users, network operators, and the various governance bodies have all legitimate interests to play a role in policy-making. Prototypical conflictual design choice: anonymity vs accountability. Govermental, public agencies are users of the infrastructure, and play a role in the infrastructure governance bodies. The Responsible Internet proposal says that POL ● should be informed by NIP and ● should drive the NCP. users governance bodies network operators data logistic tasks requirements POL NIP NCP but how? to what extent regulators can intervene on users’ activity on the infrastructure? There is no definitive, global solution: checks & balances vary on a local basis. ● Power-relationships between roles should not be hard-coded, but programmable. → an additional use case for a policy-based technology
- 42. Conclusions (1a) “give me eyes, and I’ll know where I’ll go” Computation cannot be “responsible” if the computational agent has no means to evaluate the effect of its actions, and then to prevent wrong outcomes.
- 43. Conclusions (1b) “give me eyes, and I’ll know where I’ll go” Computation cannot be “responsible” if the computational agent has no means to evaluate the effect of its actions, and then to prevent wrong outcomes. “pipes are dumb, water drinkers are not” Networks are supposed to operate blindly with respect to the content they transport, by making decisions on packets and unaware of the value of the whole transactions. But this information is (to some extent) available at the users’ endpoints! In full control, users should be able to provide some artefact specifying their preference/value structure and their expectations. Network operators should operate, still blindly, just according to these directives.
- 44. Conclusions (2a) “do not hard-code what is soft-coded” It is premature, if not wrong, to aim to a definitive solution concerning power-relationships (e.g. full-control for users and full-blindness for network operators). Too many local contextual factors intervene to set which are the “right” checks and balances. We need programmability also at this level. But what to program?
- 45. Conclusions (2b) “do not hard-code what is soft-coded” It is premature, if not wrong, to aim to a definitive solution concerning power-relationships (e.g. full-control for users and full-blindness for network operators). Too many local contextual factors intervene to set which are the “right” checks and balances. We need programmability also at this level. But what to program? “what works, it may work” For a global network like the Internet, possible starting points would be normative constructs and frameworks developed in non-computational contexts, as in international law, or most plausibly in international private law, already operative across very diverse jurisdictions.
- 46. Accounting Value Effects for Responsible Networking Giovanni Sileno g.sileno@uva.nl Complex Cyber Infrastructure (CCI) Paola Grosso p.grosso@uva.nl Multiscale Networked Systems (MNS) University of Amsterdam ACM SIGCOMM 2021 Workshop on Technologies, Applications, and Uses of a Responsible Internet (TAURIN 2021), 23 August 2021