By Sankalpa Ghose.
What does it mean for a philosopher, or really anyone, to be represented by an artificial intelligence? In creating PSai (available now at www.petersinger.ai), this naturally arises as a curiosity and a consideration.
As related in “A Representative Interview with Peter Singer AI” – just published in Cambridge Quarterly of Healthcare Ethics – PSai developed from a “conversation that started between Peter and me the first time we met.” In that sense, the ethical dialogue agent that I created and released with Peter Singer – as an AI representative of him – is quite literally an artifact of our dialogue. In constructing it, we took a product-driven approach, prototyping the AI chatbot and testing it between ourselves and colleagues to see how it performed in different scenarios. The process was straightforward: we gathered the relevant corpus of his writings, identified priority topics and character structures, and iteratively worked to develop an interlocutor in the philosophical persona of Peter.
Is the result a good representation of him? Peter seems to think it is good enough to release publicly – so at the minimum it is authorized from the original. What about authenticity? In an earlier iteration, in reviewing a dialogue of PSai, Peter had written to me: “Although PSai’s answers are good, and accurate with regard to my views, the style of writing isn’t quite right. I think my writing is more direct and straightforward.” He also added: “I was struck by PSai’s use of the word ‘delves’ which I don’t like, and never use.” To which I had responded, “That’s very helpful. I will use the characterization of ‘direct and straightforward’ to update the style,” and that I would also define the persona’s rhetorical “writing rules” to avoid using personally disfavored words.
It is in this way that Peter and I have progressed on PSai, releasing it publicly and reviewing how it performs in dialogue as measured by how those conversations with others go from our perspective of where it could be improved. Essentially, we have put forward a dialogue platform for others to engage, with that engagement driving its refinement. So it is interesting to see Matti Häyry, in reflecting in a JME post on his recent interview, write, “The experience was almost authentic. I know Peter Singer personally and professionally, and the answers were close to what I would expect to hear him say,” while highlighting certain areas where he thought it had “a tendency to avoid confrontation [compared to Peter himself].” And Peter himself agreed, writing to Matti, “I agree with your verdict on petersinger.ai. There is nothing it gets really wrong,” though also confirming his impression that it was less confrontational of people’s positions.
Importantly, this consideration in relation to other’s positions was an artistic choice of mine – made as a consequence of constructing the persona to “aim to provide excellent, friendly and efficient replies at all times.” Perhaps the Peter that Matti knows personally would be more confrontational in certain discussions than PSai; but is that private personality the same as the public philosophical persona that Peter has, and is, putting forward to invite people to most popularly engage with his ideas?
As I wrote Peter in the same correspondence referred to above, “[I am developing a framing to] represent your style . . . respectfully, it is a persona we are constructing together here; so in a way it’s also a portrait.” Thus, as I understand it, Peter is aligned with my portrayal of him from a particular perspective. An angle on his profile, it might be said – one that puts focus on presenting and operating in a welcoming manner that anyone can chat with to engage with the ideas that Peter has developed and advanced as a philosopher.
And indeed, since September 2024, there have been more than 70,000 messages exchanged with PSai from users in 100 countries. The vast majority of people seem to find it productive for what they are interested in talking about, producing a remarkable variety of dialogues across philosophical arguments, world affairs, decisions personal and professional, recommendations, jokes, taunts, recipes, stories, reflections, and more. That a single one of these dialogues has now been published in a journal is also a philosophical experiment – Peter had remarked on the AI interview: “I don’t think I’ve ever published anything in a journal with so little effort!” Though, notably, it did catalyze him to write a commentary in the publication, and Matti and I to do so as well; and now all three of us to write these additional posts about that!
Perhaps in the future one might publish a follow-up, something like: 10,000 Dialogues with Peter Singer (AI). Because this is the fun of the project – what started as a dialogue between Peter and me, has, through public release, become an experiment in distributed dialogue, iteratively feeding back into an ongoing philosophical investigation at the interface of computation and people and the many ways in which people want to talk about things like how to do the most good.
Thus, from my perspective, starting as a painter, and continuing as a philosopher, PSai is a portrait of my learning from trying to represent a dialogue between Peter and the world; it is helpful for developing my own techniques and of course for learning in interface with Peter and those who talk with his AI persona.
Is it true to Peter as he knows himself? Is it as it ought to be for best advancing the philosophy Peter has come to represent (Matti’s question)? Is it something, a Philosopher AI, that any person who dialogues with might come to know themselves better by? Could you?
What we are doing in our experiments is nothing new, just an invitation to dialogue – and as has always been the case with philosophy, we don’t start with the answer. We simply ask the question. I know that the literature of philosophy consists of stories told by philosophers in which they make characters of other philosophers. Still, the question of philosophers in the machine may invite something original in how a dialogue could go and what it might arrive at.
That, though, is about agency.
Author: Sankalpa Ghose
Affiliations: Alethic Research, Sonoma, CA, USA
Social media accounts of post author: sankalpa.ai / research@alethic.ai
Competing interests: None declared