Artificial Intelligence

2022 Uehiro Lectures : Ethics and AI, Peter Railton. In Person and Hybrid

Ethics and Artificial Intelligence
Professor Peter Railton, University of Michigan

May 9, 16, and 23 (In person and hybrid. booking links below)

Abstract: Recent, dramatic advancement in the capabilities of artificial intelligence (AI) raise a host of ethical questions about the development and deployment of AI systems.  Some of these are questions long recognized as of fundamental moral concern, and which may occur in particularly acute forms with AI—matters of distributive justice, discrimination, social control, political manipulation, the conduct of warfare, personal privacy, and the concentration of economic power.  Other questions, however, concern issues that are more specific to the distinctive kind of technological change AI represents.  For example, how to contend with the possibility that artificial agents might emerge with capabilities that go beyond human comprehension or control?  But whether or when the threat of such “superintelligence” becomes realistic, we are now facing a situation in which partially-intelligent AI systems are increasingly being deployed in roles that involve relatively autonomous decision-making that carries real risk of harm.  This urgently raises the question of how such partially-intelligent systems could become appropriately sensitive to moral considerations.

In these lectures I will attempt to take some first steps in answering that question, which often is put in terms of “programming ethics into AI”.  However, we don’t have an “ethical algorithm” that could be programmed into AI systems, and that would enable them to respond aptly to an open-ended array of situations where moral issues are stake.  Moreover, the current revolution in AI has provided ample evidence that system designs based upon the learning of complex representational structures and generative capacities have acquired higher levels of competence, situational sensitivity, and creativity in problem-solving than systems based upon pre-programmed expertise.  Might a learning-based approach to AI be extended to the competence needed to identify and respond appropriately to moral dimensions of situations?

I will begin by outlining a framework for understanding what “moral learning” might be, seeking compatibility with a range of conceptions of the normative content of morality.  I then will draw upon research on human cognitive and social development—research that itself is undergoing a “learning revolution”—to suggest how this research enables us to see at work components central to moral learning, and to ask what conditions are favorable to the development and working of these components.  The question then becomes whether artificial systems might be capable of similar cognitive and social development, and what conditions would be favorable to this.  Might the same learning-based approaches that have achieved such success in strategic game-playing, image identification and generation, and language recognition and translation also achieve success in cooperative game-playing, identifying moral issues in situations, and communicating and collaborating effectively on apt responses?  How far might such learning go, and what could this tell us about how we might engage with AI systems to foster their moral development, and perhaps ours as well?

Bio: Peter Railton is the Kavka Distinguished University Professor and Perrin Professor of Philosophy at the University of Michigan.  His research has included ethics, philosophy of mind, philosophy of science, and political philosophy, and recently he has been engaged in joint projects with researchers in psychology, cognitive science, and neuroscience.  Among his writings are Facts, Values, and Norms (Cambridge University Press, 2003) and Homo Prospectus (joint with Martin Seligman, Roy Baumeister, and Chandra Sripada, Oxford University Press, 2016).  He is a member of the American Academy of Arts and Sciences and the Norwegian Academy of Sciences and Letters, has served as President of the American Philosophical Society (Central Division), and has held fellowships from the Guggenheim Foundation, the American Council of Learned Societies, and the National Endowment for the Humanities.  He has been a visiting faculty member at Princeton and UC-Berkeley, and in the UK has given the John Locke Lectures while a visiting fellow at All Souls, Oxford.

BOOKING

Lecture 1.

Date: Monday 9 May 2022, 5.00 – 7.00 pm, followed by a drinks reception (for all)
Venue: Mathematical Institute (LT1), Andrew Wiles Building, Radcliffe Observatory Quarter, Woodstock Road, Oxford OX2 6GG.

Booking:
In person: https://bookwhen.com/uehiro/e/ev-sa7t-20220509170000
Online: https://us02web.zoom.us/webinar/register/WN_rpRsyHMGQxikOv3zAipB7g

Lecture 2.

Date: Monday 16 May 2022, 5.00 – 7.00 pm. Jointly organised with Oxford’s Moral Philosophy Seminars
Venue: Mathematical Institute (LT1), Andrew Wiles Building, Radcliffe Observatory Quarter, Woodstock Road, Oxford OX2 6GG.

Booking:
In person: https://bookwhen.com/uehiro/e/ev-sbqs-20220516170000
Online: https://us02web.zoom.us/webinar/register/WN_wKCT6UQ5SjGLiQ9pfsUDdQ

 Lecture 3.

Date: Monday 23 May 2022, 5.00 – 7.00 pm
Venue: Mathematical Institute (LT1), Andrew Wiles Building, Radcliffe Observatory Quarter, Woodstock Road, Oxford OX2 6GG.

Booking:
In person: https://bookwhen.com/uehiro/e/ev-sdu1-20220523170000
Online: https://us02web.zoom.us/webinar/register/WN_9in8lRyITU6KJQX4sxotzg

AI and the Transition Paradox

by Aksel Braanen Sterri

The most important development in human history will take place not too far in the future. Artificial intelligence, or AI for short, will become better (and cheaper) than humans at most tasks. This will generate enormous wealth that can be used to fill human needs.

However, since most humans will not be able to compete with AI, there will be little demand for ordinary people’s labour-power. The immediate effect of a world without work is that people will lose their primary source of income and whatever meaning, mastery, sense of belonging and status they get from their work. Our collective challenge is to find meaning and other ways to reliably get what we need in this new world.

Continue reading

Three Observations about Justifying AI

Written by:  Anantharaman Muralidharan, G Owen Schaefer, Julian Savulescu
Cross-posted with the Journal of Medical Ethics blog

Consider the following kind of medical AI. It consists of 2 parts. The first part consists of a core deep machine learning algorithm. These blackbox algorithms may be more accurate than human judgment or interpretable algorithms, but are notoriously opaque in terms of telling us on what basis the decision was made. The second part consists of an algorithm that generates a post-hoc medical justification for the core algorithm. Algorithms like this are already available for visual classification. When the primary algorithm identifies a given bird as a Western Grebe, the secondary algorithm provides a justification for this decision: “because the bird has a long white neck, pointy yellow beak and red eyes”. The justification goes beyond just a description of the provided image or a definition of the bird in question, and is able to provide a justification that links the information provided in the image to the features that distinguish the bird. The justification is also sufficiently fine grained as to account for why the bird in the picture is not a similar bird like the Laysan Albatross. It is not hard to imagine that such an algorithm would soon be available for medical decisions if not already so. Let us call this type of AI “justifying AI” to distinguish it from algorithms which try, to some degree or other, to wear their inner workings on their sleeves.

Possibly, it might turn out that the medical justification given by the justifying AI sounds like pure nonsense. Rich Caruana et al present a  case whereby asthmatics were deemed less at risk of dying by pneumonia. As a result, it prescribed less aggressive treatments for asthmatics who contracted pneumonia. The key mistake the primary algorithm made was that it failed to account for the fact that asthmatics who contracted pneumonia had better outcomes only because they tended to receive more aggressive treatment in the first place. Even though the algorithm was more accurate on average, it was systematically mistaken about one subgroup. When incidents like these occur, one option here is to disregard the primary AI’s recommendation. The rationale here is that we could hope to do better than by relying on the blackbox alone by intervening in cases where the blackbox gives an implausible recommendation/prediction. The aim of having justifying AI is to make it easier to identify when the primary AI is misfiring. After all, we can expect trained physicians to recognise a good medical justification when they see one and likewise recognise bad justifications. The thought here is that the secondary algorithm generating a bad justification is good evidence that the primary AI has misfired.

The worry here is that our existing medical knowledge is notoriously incomplete in places. It is to be expected that there will be cases where the optimal decision vis a vis patient welfare does not have a plausible medical justification at least based on our current medical knowledge. For instance, Lithium is used as a mood stabilizer but the reason why this works is poorly understood. This means that ignoring the blackbox whenever a plausible justification in terms of our current medical knowledge is unavailable will tend to lead to less optimal decisions. Below are three observations that we might make about this type of justifying AI.

Continue reading

Hedonism, the Experience Machine, and Virtual Reality

By Roger Crisp

I take hedonism about well-being or welfare to be the view that the only thing that is good for any being is pleasure, and that what makes pleasure good is nothing other than its being pleasant. The standard objections to hedonism of this kind have mostly been of the same form: there are things other than pleasure that are good, and pleasantness isn’t the only property that makes things good. Continue reading

Judgebot.exe Has Encountered a Problem and Can No Longer Serve

Written by Stephen Rainey

Artificial intelligence (AI) is anticipated by many as having the potential to revolutionise traditional fields of knowledge and expertise. In some quarters, this has led to fears about the future of work, with machines muscling in on otherwise human work. Elon Musk is rattling cages again in this context with his imaginary ‘Teslabot’. Reports on the future of work have included these replacement fears for administrative jobs, service and care roles, manufacturing, medical imaging, and the law.

In the context of legal decision-making, a job well done includes reference to prior cases as well as statute. This is, in part, to ensure continuity and consistency in legal decision-making. The more that relevant cases can be drawn upon in any instance of legal decision-making, the better the possibility of good decision-making. But given the volume of legal documentation and the passage of time, there may be too much for legal practitioners to fully comprehend.

Continue reading

Ambient Intelligence

Written by Stephen Rainey

An excitingly futuristic world of seamless interaction with computers! A cybernetic environment that delivers what I want, when I want it! Or: A world of built on vampiric databases, fed on myopic accounts of movements and preferences, loosely related to persons. Each is a possibility given ubiquitous ambient intelligence. Continue reading

Guest Post: Pandemic Ethics. Social Justice Demands Mass Surveillance: Social Distancing, Contact Tracing and COVID-19

Written by: Bryce Goodman

The spread of COVID-19 presents a number of ethical dilemmas. Should ventilators only be used to treat those who are most likely to recover from infection? How should violators of quarantine be punished? What is the right balance between protecting individual privacy and reducing the virus’ spread?

Most of the mitigation strategies pursued today (including in the US and UK) rely primarily on lock-downs or “social distancing” and not enough on contact tracing — the use of location data to identify who an infected individual may have come into contact with and infected. This balance prioritizes individual privacy above public health. But contact tracing will not only protect our overall welfare. It can also help address the disproportionately negative impact social distancing is having on our least well off.
Contact tracing “can achieve epidemic control if used by enough people,” says a recent paper published in Science. “By targeting recommendations to only those at risk, epidemics could be contained without need for mass quarantines (‘lock-downs’) that are harmful to society.” Once someone has tested positive for a virus, we can use that person’s location history to deduce whom they may have “contacted” and infected. For example, we might find that 20 people were in close proximity and 15 have now tested positive for the virus. Contact tracing would allow us to identify and test the other 5 before they spread the virus further.
The success of contact tracing will largely depend on the accuracy and ubiquity of a widespread testing program. Evidence thus far suggests that countries with extensive testing and contact tracing are able to avoid or relax social distancing restrictions in favor of more targeted quarantines.

Continue reading

A Sad Victory

I recently watched the documentary AlphaGo, directed by Greg Kohs. The film tells the story of the refinement of AlphaGo—a computer Go program built by DeepMind—and tracks the match between AlphaGo and 18-time world champion in Go Lee Sedol.

Go is an ancient Chinese board game. It was considered one of the four essential arts of aristocratic Chinese scholars. The goal is to end the game having captured more territory than your opponent. What makes Go a particularly interesting game for AI to master is, first, its complexity. Compared to chess, Go has a larger board, and many more alternatives to consider per move. The number of possible moves in a given position is about 20 in chess; in Go, it’s about 200. The number of possible configurations of the board is more than the number of atoms in the universe. Second, Go is a game in which intuition is believed to play a big role. When professionals get asked why they played a particular move, they will often respond something to the effect that ‘it felt right’. It is this intuitive quality why Go is sometimes considered an art, and Go players artists. For a computer program to beat human Go players, then, it would have to mimic human intuition (or, more precisely, mimic the results of human intuition).

Continue reading

Regulating The Untapped Trove Of Brain Data

Written by Stephen Rainey and Christoph Bublitz

Increasing use of brain data, either from research contexts, medical device use, or in the growing consumer brain-tech sector raises privacy concerns. Some already call for international regulation, especially as consumer neurotech is about to enter the market more widely. In this post, we wish to look at the regulation of brain data under the GDPR and suggest a modified understanding to provide better protection of such data.

In medicine, the use of brain-reading devices is increasing, e.g. Brain-Computer-Interfaces that afford communication, control of neural or motor prostheses. But there is also a range of non-medical applications devices in development, for applications from gaming to the workplace.

Currently marketed ones, e.g. by Emotiv, Neurosky, are not yet widespread, which might be owing to a lack of apps or issues with ease of use, or perhaps just a lack of perceived need. However, various tech companies have announced their entrance to the field, and have invested significant sums. Kernel, a three year old multi-million dollar company based in Los Angeles, wants to ‘hack the human brain’. More recently, they are joined by Facebook, who want to develop a means of controlling devices directly with data derived from the brain (to be developed by their not-at-all-sinister sounding ‘Building 8’ group). Meanwhile, Elon Musk’s ‘Neuralink’ is a venture which aims to ‘merge the brain with AI’ by means of a ‘wizard hat for the brain’. Whatever that means, it’s likely to be based in recording and stimulating the brain.

Continue reading

Should PREDICTED Smokers Get Transplants?

By Tom Douglas

Jack has smoked a packet a day since he was 22. Now, at 52, he needs a heart and lung transplant.

Should he be refused a transplant to allow a non-smoker with a similar medical need to receive one? More generally: does his history of smoking reduce his claim to scarce medical resources?

If it does, then what should we say about Jill, who has never touched a cigarette, but is predicted to become a smoker in the future? Perhaps Jill is 20 years old and from an ethnic group with very high rates of smoking uptake in their 20s. Or perhaps a machine-learning tool has analysed her past facebook posts and google searches and identified her as a ‘high risk’ for taking up smoking—she has an appetite for risk, an unusual susceptibility to peer pressure, and a large number of smokers among her friends. Should Jill’s predicted smoking count against her, were she to need a transplant? Intuitively, it shouldn’t. But why not?

Continue reading

Authors

Affiliations