Three part lecture series on the Philosophy and Ethics of AI
Sanjay Modgil
Reader in Artificial Intelligence, Department of Informatics, King’s College London
Part 1:Moral Machines and the Ethical impact of AI
Venerdì 6 giugno, ore 10:30, aula 111 Festa del Perdono
I review some of the challenges faced when designing and deploying moral agents as they evolve from narrow AI systems to Artificial General Intelligence (AGI) and beyond. I argue for an approach to design and use that combines the strengths and virtues of each of three main philosophical approaches to morality. I review some of the challenges for ensuring that AI systems are aligned with human values, and the extent to which uses of AI may exacerbate existing existential threats.
Part 2: Could, Would and Should Machines be Conscious?
Martedì 10 giugno, ore 10:30, aula 111 Festa del Perdono
If some entity is conscious, or ‘sentient’, it makes sense to wonder what it is like to be that entity. You might ask, what is it like to be a dog, or even, what is like to be a bat ? Could one in principle ask the same question of a current or future AI system? To help answer this question I will draw on our latest neuroscientific and philosophical theories of cognition and consciousness (in particular predictive processing theories of the Bayesian brain) , to suggest that consciousness might emerge as a byproduct of efforts to develop AGI. What then would be the ethical consequences of machines that are conscious ? Indeed, whether or not they are actually conscious, I will argue that there are overwhelming incentives to design AI systems that act, communicate and appear AsIf they are conscious, and that this in turn may exacerbate some of the ethical concerns raised in Part 1.
Part 3: Truth and the Dialogical Turn in Logic
Venerdì 13 giugno, ore 10:30, aula 111 Festa del Perdono
I briefly review some of the key philosophical approaches to truth, and the impact this has had on the development of logics for AI. Then, building on pragmatist approaches to philosophy, I propose that instead of seeing truth as picking out a relation between beliefs and the world, we conceptualise truth as a norm; a norm that describes how beliefs are shaped by perception and how beliefs should be shaped by our individual and collective reasoning practices. In view of this normative conception of truth, I then argue that we should revisit the relationship between truth and logic. In particular, logics should be formalised in terms of dialogical interactions with the world, ourselves and others. Finally, I review how this ‘dialogical turn’ in logic can help address some of the ethical challenges discussed in Part 1. In particular, aligning AI systems with human values through enculturation of these systems into human-AI collectives in which humans and AI reason together and learn from each other.