> "Based on our research and replication studies, we conclude that contemporary LLMs demonstrate an enhanced yet limited degree of Theory of Mind abilities."
Looking at their data and their experiments, I'd actually come to the opposite conclusion of the title. It's true that current LLMs are probably not quite at human level performance for these tasks, they're not that far off either and clearly we see as models increase in size and sophistication their performance on these tasks are improving.
So it seems like maybe a better title would be "LLMs don't have as advanced a theory of mind as a human does... for now..."
Questions about whether an LLM truly has a "theory of mind" or has "human level consciousness" or not are kind of beside the point. It can ingest a corpus of human interactions and produce outputs that take into account unstated human emotions and thoughts to optimize whatever it's optimizing. That's scary because of what it can and will do, even if it's just a giant bag of tensor products.
More than just theory of mind. I guess this gets into the alignment problem.
It seems to me that LLMs do not keep on thinking, "I'd better get this right, or I'm going to lose credibility." They have a theory of mind, but it stops there at simply having one. It's not like they're thinking about the 2nd and 3rd order implications.
To put this into perspective: Imagine interacting with another person, who doesn't value Truth at all. Or perhaps remember an occasion when such an interaction happened. In general people don't like these interactions, and they react with distrust and even hostility towards such people.
One intriguing possibility is that LLMs may have stumbled upon an as yet undiscovered structure/"world model" underpinning the very concept of intelligence itself.
Should such a structure exist (who knows really, it may), then what we are seeing may well be displays of genuine intelligence and reasoning ability.
Can LLMs ever experience consciousness and qualia though? Now that is a question we may never know the answer to.
All this is so fascinating and I wonder how much farther LLMs can take us.
i dont think LLMs in their current state are anything like the human mind. They need the ability to have multiple thoughts ongoing, background thoughts, planning... right now LLMs are a little like snap responses answering questions, the type you give without thinking, like intuition. Which can very easily fall outside the bounds of an acceptable answer
The article doesn't say that LLMs aren't useful - the "hype" they mean is overestimating their capabilities. An LLM may be able to pass a "theory of mind" test, or it may fail spectacularly, depending on how you prompt it. And that's because, despite all of its training data, it's not capable of actually reasoning. That may change in the future, but we're not there yet, and (AFAIK) nobody can tell how long it will take to get there.
One of the surprising results in research lately was the theory of mind paper the other week that found around half of humans failed the transparent boxes version of the theory of mind questions - something previously assumed to be uniquely a LLM failure case.
I suspect over the next few years we're going to see more and more behaviors in LLMs that turn out to be predictive of human features.
At the same time, if LLMs are based on all or enough human writings then don't they necessarily contain a theory of mind? A rather general, smoothed out and still neurotic one probably. But still just like an LLM can't be expected to have a specific knowledge of hydraulics, it also has read more hydraulics than even experts might be expected to. That's the entire issue about it, right? This issue of "is most of our mind basically just mixing and matching stuff we have seen, read, heard?" Do humans have some magical theory of mind that somehow stands ASIDE from all the "normal" learned stuff?
Of course, yes, we do know one thing that's missing in LLMs, which is "loop and helpers" like you describe. Which I'm sure many people are currently hacking at - one way being for the LLM to talk to itself.
But as for "a theory of mind", if enough writings served as input, then LLMs do have plenty of that.
Another question is whether LLMs are raised to behave like humans (which might be where they most NEED some theory of mind). Of course not. The ones we know most about are only question answerers. The theory of mind they might have (that is not negated by the lack of loop and internal deliberation) may be overwhelmed by the pre- and post-processing: "no sex, no murder plots, talk to the human like they are 5, bla bla bla". And yet you can ask things like "Tell it like you are speaking to 5 year olds who want to have a fun time". Some theory of mind makes it through.
If LLMs are beating humans in theory of mind tasks, your theory of mind is incorrect.
I’d not heard of Jo Cameron but one outlier, for me, wouldn’t hyper-negate the whole idea out of existence. A huge amount of what I suffer with is not physical pain.
I think it's interesting that human minds generally (though not always!) improve when exposed to the output of other human minds. It seems to be the opposite for current LLMs.
It's more to point out how far the LLMs we have today are from anything that ought to be considered thoughts. They are far more mechanical than anything else
"LLMs can mimic the language patterns necessary to express 'Theory of Mind' concepts" != "Theory of Mind May Have Spontaneously Emerged"
Let's imaging I have an API. This API tells me how much money I have in my bank account. One day, someone hacks the API to always return "One Gajillion Dollars." Does that mean that "One Gajillion Dollars" spontaneously emerged from my bank account?
ToM tests are meant to measure a hidden state that is mediated by (and only accessible through) language. Merely repeating the appropriate words is insufficient to conclude ToM exists. In fact, we know ToM doesn't exist because there's no hidden state.
The authors know this, and write "theory of mind-like ability" in the abstract, rather than just "theory of mind."
This is a cool new task it ChatGPT learned to complete! I love that they did this! But this is more "we beat the current record BLEU record" and less "this chatbot is kinda sentient"
> We don’t know if they reason; we don’t know if they have their own internal goals that they’ve learned or what they might be.
I keep seeing researches say this and I don't really understand.
LLMs, as they are now, are "frozen in time." Their internal state is defined by what is directly in front of them at the moment, but otherwise does not change once trained. Hence I dont think they can have these higher thinking abilities.
That will change when they start training themselves as they go, but we are not there yet.
"Exactly this. I too find this to be the best intuition for LLMs right now: they're not comparable to an entire combined human mind - they're comparable to subconscious, or inner voice"
reply