AI machines as moral agents, Conclusion and references (part 13)

Now it’s time for the final part — the conclusion of the thesis.

For a mission statement, see Part 1 — for an index, see the Overview.

6. Summary

I have argued that an artificial agent can be a morally accountable agent. This claim relies on Floridi and Sanders’ definition of moral accountability which entails that the agent is the active cause of the act but does not take into account whether the agent was aware of, understood or should be blamed for the act; was morally responsible. Accountability is a thinner form of moral classification than responsibility. And while a large part of moral philosophizing belongs to the field of responsibly, there is still enough room left for important and interesting debates concerning accountability. As the complexity and quantity of artifacts most likely will increase in the near future, so will the social interactions between humans and artifacts. This is a reason for why this field could become more important in the future. I also see a possibility that discussions concerning accountability could open up other discussions outside the area of artificial intelligence. Floridi and Sanders’ search-and-rescue dogs is an example of this.

The argument also relies on Barandiaran et al’s definition of minimal agency with the conditions of individuality, asymmetry and normativity. It is a wider definition than intentionality agency and a definition that does not presuppose consciousness. I have tried to show that minimal agency is more psychologically and linguistically intuitive and serves as a better way to define agency than intentionality agency. This can be viewed as a mere pragmatical response, as saying that “since the original task of defining agency is so hard, let us instead pick another way to define it that is not strictly correct, but much more convenient to use”. For a philosophical argument, that would be very bad indeed. My argument is not that minimal agency “will do”, but rather that intentionality agency relies on vague and uncorroborated features and that minimal agency is actually more in line with empirical evidence, more in line with what is actually meant by ‘agency’. Minimal agency is the minimal amount of what is meant, the common denominator for a basic agency, and it tries to not say anything more than what is needed for that purpose. Intentionality agency has its merits, but it invites unresolved philosophical problems. Dennett’s instrumental view of beliefs and desires could explain all the mental states that is needed for agency without requiring consciousness. Defining agency is, as should be clear by now, a complicated and hard task, and Barandiaran et al’s minimal agency is merely the beginning of that process.

Lastly the argument relies on Floridi and Sanders’ criterion adaptability (which I named the ‘learning criterion’ but applied it differently). The agent needs to meet the learning criterion to be held accountable. For an agent to meet the criterion it needs to be able to change the rules that govern how the agent interacts with the world in a relevant way. A minimal agent that passes the learning criterion is then equivalent with Floridi and Sanders’ accountable moral agent except that the former definition does not allow arbitrary behavior and does not rely on the concept of levels of abstraction. What I have not addressed is the importance of context for the learning criterion. If the agent does have the ability to learn, then it also depends much on why a particular learning ability is relevant in a particular moral context. I have not addressed what it takes for this ability to be relevant in the right way.

I have addressed concerns by Himma and Johnson that artifacts cannot be free in the relevant sense for moral agency by using my example of an indirectly programmed artifact. If the behavior of the artifact is significantly shaped by its input from the environment, the designer’s intentionality does not have the control over the artifact’s behavior that Johnson needs, and the artifact does not have to be fully determined as Himma suggests. Indirect programming is analogous to how children are being raised in society and are thereby formed by their experience in the world, and since children can become moral agents this suggests that artifacts can too.

Comments:

There are a few things here that I would like to explore more. What are the boundaries of an AI — is a autonomous vehicle an AI, the whole system (including the software) or only the software? Can you break down consciousness into clearer, more definable parts and can those parts be achieved individually and collectively in AI agents? Can we develop a ‘anthropology’ for AI artifacts that explains their type of thinking and their type of mind? Exactly what does it take to say that an artifact is a moral agent? Can AI artifacts have their own set of morality?

I will try to expand on some of these topics in coming texts. Please comment or contact me if you have questions or related ideas.

References

Barandiaran, X., Di Paolo, E. & Rohde, M. 2009 ‘Defining Agency. individuality, normativity, asymmetry and spatio-temporality in action (v. 1.0.)’, Journal of Adaptive Behavior

Clarke, Randolph and Capes, Justin, “Incompatibilist (Nondeterministic) Theories of Free Will”, The Stanford Encyclopedia of Philosophy (Spring 2017 Edition), Edward N. Zalta (ed.),

URL = <https://plato.stanford.edu/archives/spr2017/entries/incompatibilism-theories/>.

Cole, David, 2019, ‘The Chinese Room Argument’, The Stanford Encyclopedia of Philosophy (Spring 2019 Edition), Edward N. Zalta (ed.),
URL = <https://plato.stanford.edu/archives/spr2019/entries/chinese-room/>.

Davenport, David 2014, ‘Moral Mechanisms’, Philosophy and Technology, Vol. 27, pp. 97–111,
DOI 10.1007/s13347–013–0147–2

Dennett, Daniel C. 1987, The Intentional Stance, MIT Press

Dennett, D., 2009. Intentional systems theory. The Oxford handbook of philosophy of mind, pp.339–350.

Eshleman, Andrew 2016, ‘Moral Responsibility’, The Stanford Encyclopedia of Philosophy (Winter 2016 Edition), Edward N. Zalta (ed.), URL = <https://plato.stanford.edu/archives/win2016/entries/moral-responsibility/>

Etzioni, A., Etzioni, O. 2017, ‘Incorporating Ethics into Artificial Intelligence’, Ethics, Vol. 21, pp. 403–418 DOI 10.1007/s10892–017–9252–2

Fiala, B., Arico, A., Nichols, S. 2012, ‘On the psychological origins of dualism: Dual-process cognition and the explanatory gap’,
DOI:10.1093/acprof:oso/9780199794393.003.0005

Floridi, L., Sanders, J.W. 2004, ‘On the Morality of Artificial Agents’,
Minds and Machine, Vol. 14, pp. 349–379

Himma, Kenneth E. 2009, ‘Artificial agency, consciousness, and the criteria for moral agency: what properties must an artificial agent have to be a moral agent?’, Ethics and Information Technology, Vol. 11, pp. 19–29, DOI 10.1007/s10676–008–9167–5

Johnson, Deborah G. 2006, ‘Computer systems: Moral entities but not moral agents’, Ethics, and Information Technology, Vol. 8, pp. 195–204 DOI 10.1007/s10676–006–9111–5

Jacob, Pierre 2019, ‘Intentionality’ The Stanford Encyclopedia of Philosophy (Spring 2019 Edition), Edward N. Zalta (ed.),
URL = <https://plato.stanford.edu/archives/spr2019/entries/intentionality/>.

Neely, Erica L. 2014, ‘Machines and the Moral Community’,
Philosophy and Technology, Vol. 27, pp. 97–111, DOI 10.1007/s13347–013–0114-y

Schlosser, Markus 2015, ‘Agency’ The Stanford Encyclopedia of Philosophy (Fall 2015 Edition), Edward N. Zalta (ed.),
URL = <https://plato.stanford.edu/archives/fall2015/entries/agency/>.

Torrance, Steve 2014, ‘Artificial Consciousness and Artificial Ethics: Between Realism and Social Relationism’, Philosophy and Technology, Vol. 27, pp. 9–29,
DOI 10.1007/s13347–013–0114-y

Søraker, Johnny H. 2014, ‘Continuities and Discontinuities Between Humans, Intelligent Machines, and Other Entities’, Philosophy and Technology, Vol. 27, pp. 31–46

Turing, Alan M. 1950, ‘Computing Machinery and Intelligence’, Mind, Vol. 49, pp. 433–460