Computational Accountability M. Baldoni, C. Baroglio , K. M. May, R. - - PowerPoint PPT Presentation

Computational Accountability

• M. Baldoni, C. Baroglio, K. M. May, R. Micalizio and S.

Tedeschi1

1Universit`

a degli Studi di Torino

Genova, Nov. 28th, 2016

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 1 / 20

Ubiquitous AI

AI methodologies and techniques: are raising greater and greater attention are practically used in sensitive fields: search engines, electronic markets, smart homes, military, data analysis, self-driving cars, ... The research community is discussing the ethical implications Two main perspectives have emerged ...

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 2 / 20

Ubiquitous AI

AI methodologies and techniques: are raising greater and greater attention are practically used in sensitive fields: search engines, electronic markets, smart homes, military, data analysis, self-driving cars, ... The research community is discussing the ethical implications Two main perspectives have emerged ...

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 2 / 20

Main lines: ethical concerns about

Adoption of AI techniques and methodologies

impact of AI on work and

• ther aspects of our lives:

threat to or saviour for humanity? safeguards necessary within AI research (applications in the military, commerce, ...)

Design and application of artificially intelligent systems

Machine ethics: moral behavior of artificially intelligent beings Roboethics: moral behavior

• f humans as they design,

construct, use artifically intelligent beings

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 3 / 20

Testimonies

AAAI 2016 WS on AI, Ethics and Society; IJCAI 2016 WS on Ethics in Artificial Intelligence; ECAI 2016 WS on Ethics in the Design of Intelligent Agents; Guide to the ethical design and application of robots and robotic systems (BS 8611:2016 standard); ETHICAA project: “Machines and software acting on behalf of humans get more autonomy and are increasingly less under the control of human operators or users [...] the question of an ethical regulation and control of autonomous agents is raised.”

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 4 / 20

A third perspective: Computational Accountability

AI as a tool

that supports the realization of ethical business through the traceability, evaluation, and communication of values and good conduct “We need an inclusive, robust yet flexible accountability framework”, Ban Ki-moon, Secretary-General of the United Nations, 2014 What is accountability? Why is it a complex problem to solve? Why AI?

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 5 / 20

A third perspective: Computational Accountability

AI as a tool

that supports the realization of ethical business through the traceability, evaluation, and communication of values and good conduct “We need an inclusive, robust yet flexible accountability framework”, Ban Ki-moon, Secretary-General of the United Nations, 2014 What is accountability? Why is it a complex problem to solve? Why AI?

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 5 / 20

A third perspective: Computational Accountability

AI as a tool

that supports the realization of ethical business through the traceability, evaluation, and communication of values and good conduct “We need an inclusive, robust yet flexible accountability framework”, Ban Ki-moon, Secretary-General of the United Nations, 2014 What is accountability? Why is it a complex problem to solve? Why AI?

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 5 / 20

What is Accountability?

Ontological views, e.g.: accountability is the intersection between responsibility and answerability

Societies develop models of responsibility

Definitions vary in approach and scope and different communities do not share a same understanding

In English

Responsible: who does Accountable: who signs off (the account giver)

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 6 / 20

What is Accountability?

Ontological views, e.g.: accountability is the intersection between responsibility and answerability

Societies develop models of responsibility

Definitions vary in approach and scope and different communities do not share a same understanding

In English

Responsible: who does Accountable: who signs off (the account giver)

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 6 / 20

A characterization of accountability

To trace, evaluate, and communicate the value of accountability

we need a more “operational” characterization, that sets accountability inside processes, associates it to individuals and roles, etc. Accountability is a composition of processes Accountability does not hinder autonomy Accountability implies agency Accountability implies causation Accountability implies significance A system of accountability must be sound and complete

[Burgemeestre and Hulstijn, 2015, Yazdanpanah and Dastani, 2016]

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 7 / 20

A characterization of accountability

To trace, evaluate, and communicate the value of accountability

we need a more “operational” characterization, that sets accountability inside processes, associates it to individuals and roles, etc. Accountability is a composition of processes Accountability does not hinder autonomy Accountability implies agency Accountability implies causation Accountability implies significance A system of accountability must be sound and complete

[Burgemeestre and Hulstijn, 2015, Yazdanpanah and Dastani, 2016]

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 7 / 20

A characterization of accountability

Accountability is a composition of processes: identification, judgement, sanction Accountability does not hinder autonomy: an accountable agent has complete freedom but later may requested to account for Accountability implies agency: agents are accountable, tools are not Accountability implies causation: (in)action of the accountable party causes the situation to verify Accountability implies significance: judgement (guilty/not guilty) must consider the degree to which the action of the accountable party was significant in creating the unwanted situation A system of accountability must be sound and complete: only faulty agents will be blamed, all faulty agents will be identified

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 8 / 20

Accountability a complex problem? Need AI?

Ted and Bill are painters, called by Jim for painting a room The walls are white, Jim would like to have them painted of the same color Ted makes a better offer and gets the job Bill, in the night, paints the walls black Ted will not be able to satisfy his commitment with Jim! He needs to use twice as much paint as expected at a much higher cost

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 9 / 20

Accountability a complex problem? Need AI?

Ted is unable to fulfill the contract: is he accountable? Conditions changed in a hardly predictable way Contextual conditions that hardly change over time are presumed implicitly stipulated even when they are not formalized Is Jim accountable? The door may have been left unlocked Can the real responsible be considered in deciding who is accountable?

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 10 / 20

Beyond Ted, Bill, and Jim

For firms, business ethics and compliance programs are becoming critical and the realization of accountability systems is crucial

Modern enterprises [Bridgeland and Zahavi, 2008] are:

complex and distributed because they involve offices, activities, actors, resources, often heterogeneous, geographically distributed, and dependant on one another; aleatory because they are affected by unpredictable events like new laws, market trends, incidents A further aspect is dynamicity: people are hired/fired, promoted, offices are restructured, moved, stakeholders change, ..., who is accountable?

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 11 / 20

Beyond Ted, Bill, and Jim

For firms, business ethics and compliance programs are becoming critical and the realization of accountability systems is crucial However, methodologies and tools for designing business processes and workflow engines do not yet support the realization of accountability systems

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 12 / 20

AI and accountability

How to represent and manipulate responsibilities? Something goes wrong, who is accountable? How to distribute responsibilities to workers? How to coordinate workers with different responsibilities? Knowledge representation (delegation, release, ...) Reasoning about the past, looking for causes (diagnosis) Reasoning about the future: construction of individual or joint plans

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 13 / 20

AI and accountability

How to represent and manipulate responsibilities? Something goes wrong, who is accountable? How to distribute responsibilities to workers? How to coordinate workers with different responsibilities? Knowledge representation (delegation, release, ...) Reasoning about the past, looking for causes (diagnosis) Reasoning about the future: construction of individual or joint plans

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 13 / 20

Pathway to accountability

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 14 / 20

Main approaches to the development of business models

Business processes

describe how a set of interrelated activities lead to a measurable result in response to an external event connect and coordinate people, offices,

• rganizations, and

software in a compound flow of execution.

Artifact-centric approach [Bhattacharya et al., 2007]

data-centric: concrete, identifiable, self-describing chunks of information they include a lifecycle model, that contains the key states through which data evolve (used at design time to distribute tasks)

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 15 / 20

Limits

Accountability mainly concerns interactions

but in business processes: interactions among actors

• nly indirectly represented

as of input/output dependencies between activities this hinders reasoning about accountability

Artifacts only tell who is assigned this or that task

business artifacts disregard the design of the processes that operate on them but accountability is affected by the execution flow, by the context, and by the role of the involved parties so artifacts hinder the realization of accountability systems Need to join forces ...[Baldoni et al., 2016]

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 16 / 20

Our approach

Basic Notion: Social Commitments [Singh, 1999]

C(x, y, s, u): the debtor agent x is bound to the creditor agent y to bring about u when s holds commitments have a normative value commitments have a life cycle they populate the social state of the interaction commitments/social state can be realized as business artifacts

Agents embody the activity-centric view

They can further be coordinated by means of protocols

Commitment embody the data-centric view

Business artifacts constitute the environment in which agents are situated

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 17 / 20

One possible picture

Responsibility: concerns the future [Yazdanpanah and Dastani, 2016] Accountability: concerns the past [Burgemeestre and Hulstijn, 2015]

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 18 / 20

An Example from a Logistic Problem

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 19 / 20

Conclusion

Computational Accountability is a new field AI as a necessary tool to support traceability, communication and evaluation of values (ethical business) Possibility (and need) to integrate plenty of methodologies and approaches

Questions?

• Acknowledgements. This work was partially supported by the Accountable Trustworthy Organizations and Systems (AThOS)

project, funded by Universit` a degli Studi di Torino and Compagnia di San Paolo (CSP 2014). BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 20 / 20

Conclusion

Computational Accountability is a new field AI as a necessary tool to support traceability, communication and evaluation of values (ethical business) Possibility (and need) to integrate plenty of methodologies and approaches

Questions?

• Acknowledgements. This work was partially supported by the Accountable Trustworthy Organizations and Systems (AThOS)

project, funded by Universit` a degli Studi di Torino and Compagnia di San Paolo (CSP 2014). BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 20 / 20

Baldoni, M., Baroglio, C., Calvanese, D., Micalizio, R., and Montali,

• M. (2016).

Towards data- and norm-aware multiagent systems. In Post-Proc. of the 4th International Workshop on Engineering Multi-Agent Systems, EMAS 2016, Revised Selected and Invited Papers, LNAI. Springer. Bhattacharya, K., Caswell, N. S., Kumaran, S., Nigam, A., and Wu,

• F. Y. (2007).

Artifact-centered operational modeling: Lessons from customer engagements. IBM Systems Journal, 46(4):703–721. Bridgeland, D. M. and Zahavi, R. (2008). Business Modeling: A Practical Guide to Realizing Business Value. Morgan Kaufmann Publishers Inc., San Francisco, CA, USA. Burgemeestre, B. and Hulstijn, J. (2015). Handbook of Ethics, Values, and Technological Design: Sources, theory, values and application domains, chapter Designing for

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 20 / 20

Accountability and Transparency: A value-based argumentation approach. Springer. Singh, M. P. (1999). An ontology for commitments in multiagent systems.

• Artif. Intell. Law, 7(1):97–113.

Yazdanpanah, V. and Dastani, M. (2016). Distant group responsibility in multi-agent systems. In Baldoni, M., Chopra, A. K., Son, T. C., Hirayama, K., and Torroni, P., editors, PRIMA 2016: Principles and Practice of Multi-Agent Systems - 19th Int. Conf, volume 9862 of Lecture Notes in Computer Science, pages 261–278. Springer.

BBMMT (UniTO) URANIA 2016 Genova, Nov. 28th, 2016 20 / 20