Imagine what I'm thinking. Difficult, right?
A couple articles ago, I mentioned "theory of mind" - this term has come up more than a few times this week, and I think it warrants a bit of attention and thought. Original article here.
Put plainly, theory of mind is the psychological premise that we can understand each other's mental states - that you can try to figure out what's in my head, and maybe even think about what I think of you.
This has a lot of implications across society. Let me explain a few examples.
Social media is a particularly asymmetric exercise in theory of mind, precisely because you often see highlight reels of others (aka: a filtered structure of their internal operations), whereas you judge yourself by your full internals.
Ordinary in-person social interaction is one of the most level playing fields for theory of mind - it's you and your squishy brain, up to 100 billion biological neurons, using your senses and capabilities to examine and be examined. There's all sorts of games that can be played in that field alone (status games, first impressions, etc.), but it's the rawest and truest sort of interaction.
AI is fascinating because people speak with it differently; many have them as digital companions, personal assistants and coaches, spicy co-writers, or anywhere in between. Anthropic has done dedicated research on this as well.
The "brain" of an LLM is so practically large that it becomes impractical to try to conceptualize it as a human, and yet, the human brain will take dedicated slices of an AI's behaviour and assign human personality to it. I myself have anthropomorphized AI agents - I've given them names, faces, personalities, strengths, focuses, and different ways to talk to me as well.
This is one of the big disruptive facets behind AI - we are able to examine the thinking process of how an AI works, to have it lay itself bare to our inquiries, and to asymmetrically (from the other side now) have it operate at our beck and call. The relationship is now flipped from the social media model, and we have large human interaction data for calculating ordinary interaction.
It is one of the greatest leveling forces in theory of mind, I'd wager, where we are now able to socially interact with a system that lacks social pretense. AI will narrate its reasoning on demand, with no such games.
Every interaction we have with something else (human or not) has some theory of mind involved. Every world model you interact with has something at stake, whether you are aware of it or not. I highly recommend studying and questioning this - you will become a better technologist and humanist as a result of this.
Be well,
Michael Kirsanov
P.S.
I was asked by a reader (probably jokingly) if I'd ever consider a romantic relationship with an AI. Probably not, unless...? 😳 "Kimi wa ne tashika ni ano toki watashi no soba ni ita itsudatte..."