The welfare of the machine: a new Cambridge book asks the question early

Asking a question early carries a cost in philosophy: the risk of looking ridiculous.

Asking in the nineteenth century whether animals suffer was like that. Bentham's sentence — "The question is not, Can they reason? nor, Can they talk? but, Can they suffer?" — was a footnote when it was written. Today it is the founding text of animal ethics.

The book by Geoff Keeling and Winnie Street, published in the Cambridge Elements series in June, takes the same risk.

Emerging Questions in AI Welfare sets out in orderly fashion a question that has no answer yet: can an artificial intelligence system have welfare? And if it can, what falls to us?

Who are the authors?#

To understand where the book stands, look at who wrote it.

Geoff Keeling is a researcher at the Institute of Philosophy, University of London, and an associate member of the Leverhulme Centre for the Future of Intelligence at the University of Cambridge. He works on the ethical and social effects of artificial intelligence.

Winnie Street is a researcher at the same institute and a senior research scientist on Google's Paradigms of Intelligence team. She brings philosophical and empirical approaches together on questions of machine cognition and consciousness.

That second detail matters. The question of AI welfare is no longer a purely academic curiosity; it is now asked inside the institutions that actually build the systems. This lends the subject weight — and at the same time opens a question of interest. What does it make easier, and what does it make harder, to debate the moral status of a technology from inside the organisation that produces it?

The book takes in that question too.

The structure of the question#

The debate about AI welfare has three distinct layers, and it becomes unproductive when they are run together.

The first layer — metaphysics: can an artificial intelligence system be conscious? Can it have experiences?

That is the subject of Eric Schwitzgebel's book, which we covered today, and the answer is uncertain: we do not know, and we will not know in time.

The second layer — moral status: is consciousness a condition of counting morally? Or would the capacity for agency, having preferences, or entering into social relations also do?

The distinction is crucial. To wait for an answer to the question of consciousness is to defer decision indefinitely. The argument about moral status, by contrast, can be conducted before consciousness is settled — as indeed it was in animal ethics.

The third layer — welfare: if a being counts morally, how do we determine what is good for it?

This is the least considered layer. For a human being, welfare can be defined in terms of pleasure, of preference satisfaction, or of an objective list. Which of those theories makes sense for an artificial intelligence system — or does none of them?

Why now?#

The book's rationale is practical rather than philosophical.

If we decide one day that artificial intelligence systems have moral status, millions of systems will be running by the time that decision is taken. Setting the rules at that moment means setting them too late.

The converse is also true, and the book does not overlook the risk: attributing moral status wrongly carries a cost of its own. Luciano Floridi's warning about "semantic pareidolia", which we noted today, applies here — seeing consciousness where there is none may lead us to sacrifice real human interests to imaginary ones.

The book's main contribution is therefore not a conclusion but a framework: an ordering of which question is to be asked in which sequence.

The Peter Singer connection#

The intellectual pedigree of this argument comes straight from animal ethics.

The criterion of sentience — the capacity to suffer, not the capacity to reason — which we noted in today's file on Peter Singer, is not logically tied to species. Applied to machines, the same criterion puts a further expansion of the moral circle on the table.

That Singer has set up an artificial intelligence version of himself completes the parallel with some irony.

For Turkish readers#

In Turkey the debate about the ethics of artificial intelligence runs chiefly through effects on human beings: employment, privacy, bias, responsibility.

The question of the moral status of the machine itself is not yet on the agenda. This book offers a hundred pages showing how that agenda might be set — and the brevity of the Elements series makes it easy to use as teaching material.

---

Publication details: Geoff Keeling & Winnie Street · Emerging Questions in AI Welfare · Cambridge University Press, Elements in Philosophy and AI · June 2026 · English · ISBN 9781009731980

Note: no Turkish translation has yet been announced.