2

I haven't been thinking about this very long, but when I encounter an infinite regress, I recognize it as a problem and then look for ways around that (avoid, make it virtuous, I don't know). I may even be hardwired to do this, insofar as I have somewhat ineluctable potential toward belief in God, I trivially know that I cannot complete a truly infinite task, and so on.

What about LLM's and AI in general? How do they respond to infinte regresses, can they identify them, etc.?

In part, I am wondering whether self reflection, as a cognitive ability linked to human cognition/understanding in particular, must misrepresent itself, and whether programming that is impossible and why LLM do seem to lack the magic of wet brains. I don't know barely anything about this, but it seems that phenomenologists think there is a paradox at the heart of self reflection

reflection is the process whereby consciousness directs its intentional aim at itself, thereby taking itself as its own object. In this regard, reflection appears to be a higher-order act that reifies a state into a sort of mental item. Or to phrase it differently: reflection — as King Midas’ golden touch — turns all that which it reflects upon into an object, and thus subjectivity into a form of self-objectification. As a consequence, reflection paradoxically leads to an infinite regress.

https://cfs.ophen.org/2016/11/24/rocio-garces-the-paradox-of-self-reflection-and-alterity/

The question sounds silly, but there is no "loop" commmand in my brain I must obey, and I was just curious if that and infinite tasks had any relevance to today's AI.

Improve this question
7
  • hopefully not too conversational of me
    – andrós
    Commented Jul 18 at 2:49
  • 2
    In CS, recursive algorithms are common. I'm considering regression and recursion to be equivalent in this context. An infinite recursion causes a stack overflow so there are protection mechanisms to make sure a recursive algorithm doesn't suck up all the CPU resources.
    – Idiosyncratic Soul
    Commented Jul 18 at 3:28
  • 1
    ah interesting/fun last sentence, thanks @IdiosyncraticSoul
    – andrós
    Commented Jul 18 at 3:55
  • 1
    Isn't your question essentially a reformulation of the Halting problem? As long as AI and LLM run on turing machines, this seems applicable.
    – Philip Klöcking ♦
    Commented Jul 18 at 8:20
  • 1
    there is no "loop" commmand in my brain I must obey Really??? I will tell that my brain next time I lie awake in the night, with some crazy thought on autoplay.
    – AnoE
    Commented Jul 18 at 11:43

1 Answer 1

Reset to default
3

LLMs lack "the magic of wet brains" by virtue of being "just" highly advanced autocomplete. They are a language model, nothing more, nothing less.

That is no proof that it is impossible in general in an AI, it's just that we have to understand that LLMs are text completion, not some AGI.

Improve this answer
9
  • ha yeah ok let's stop thinking about LLM :D
    – andrós
    Commented Jul 18 at 6:33
  • 1
    @andrós Well your question at it's heart is actually two: 1) Is there reason to believe AI in general can't do self reflection? (I don't see why that would be). 2) How do current AIs/LLMs deal with infinite regresses (given that LLMs are text completion and can barely do logic at all because of that, the answer to that is probably "not great" but doesn't bring you closer to an answer to 1) ).
    – kutschkem
    Commented Jul 18 at 8:10
  • yes, well spotted, thanks, but they might be linked, at least in a conversational way. in general, it is dimwitted to assume everyone is
    – andrós
    Commented Jul 18 at 8:20
  • @kutschkem I am convinced at the heart of the question there is the halting problem, with the added layer of whether the ability to identify non-halting processes of one's own can count as (self-)consciousness, isn't it? In other words, given that LLMs/AIs gain the ability of identifying non-halting processes, wouldn't that mean that they transcended their turing machine based existence through their predictive (heuristic) abilities?
    – Philip Klöcking ♦
    Commented Jul 18 at 8:35
  • 1
    Note that there's a counter philosophy that humans don't really have intelligence and free will, it's just even more advanced auto-completion.
    – Barmar
    Commented Jul 18 at 14:31

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .