Hacker News Reader

Comment by ggm | Hacker News Reader

Yet, across both sides of morphogenesis and reproduction, some remarkably complex behaviour persists: food preferences, predatory and predator evading behaviours, mating, signalling for mating, epigenetic or not, so something cellular and mechanistically related to "memory" in a linguistic sense persists. How is this so?

The problem of english: we call it "memory" but it encompasses more than just overt neuronal knowledge acquisition, it includes the gut brain relationship and intracellular activity.

Maybe all the things I list are not in fact "memories" but then..

View on HN

What you are contemplating here is a question of English-language semantics and usage, not biology. It is commonplace for words previously used in one restricted sense to be adopted to talk about other situations that are merely analogous in some sense. This is useful and probably inevitable, but it does not mean we can take these disparate uses to imply a sameness, where the expanded usage is merely the consequence of a similarity (and sometimes the similarity is quite abstract.)

In this case, neuronal information storage and retrieval, which is the biological basis of memory as the word is ordinarily used, is quite different and separable from the genetic storage and use of information.

Quite right. I argue this point frequently in connection with AI. Undoubtedly artificial, no intelligence!

Thank you for another example of where arguments from language semantics and usage fails to address a complex issue!

This example happens to contain all there is to Searle's "Chinese Room" argument.

You just lost, “The Game”.

And I was on a real heater too. Why does this keep happening.

It's been a while since I lost the game.

Same with the recent linguistic discussion of whether AI is lying, hallucinating, or confabulating (personally, I think confabulation has the closest linguistic meaning to what AI does). All of these are really terms that apply to agents with agency and don’t apply to LLMs, but we don’t currently have terms for them, yet.

Of these, lying is most clearly (definitely, I would say) a term that applies to entities with agency, but I don't see hallucinating being so. Confabulation seems to me to fall somewhere between: insofar as it is not simply a synonym for lying, that seems to lie in it being a pathology, and one which tends to compromise the subject's ability to act with agency (a person fully aware that their confabulation is false is, if they hope their listeners will believe them, lying, and has apparently chosen to do so.)

Personally, I feel that the last two are useful analogies while the first is not, but they are all of limited applicability because I doubt an LLM's use of language follows from, or is indicative of, an understanding that there is an external world.

Ooh good word choice. Lying and hallucinate have felt inadequate, I agree with you that confabulation best captures what is going on.

Plus I think its kind of archaic now, so it has less baggage to contend with/be translated between folk

AI has a degree of intelligence.

Preclude even that and you preclude some humans from the same.

Show me your definition of intelligence and I will show you how people might reasonably disagree with it.

(The simplest refutation would be that demonstrating intelligence requires first demonstrating motivation on the part of the subject. The current way we interact with "AI" is much more akin to prodding an organism with a stick, seeing it recoil in response to the stimulus, and then claiming that it therefore has intelligence. No, you haven't demonstrated intelligence, you have at most proven it has an autonomous sensory response -- anything beyond that is confusing the complexity of the response with complexity of the mechanism producing it).

Motivation has nothing to do with intelligence

What? Can you give an example of a being that does (or could) be considered intelligent without the ability to engage in self directed actuons?

We don't know intelligent beings that aren't mammals, either, but that doesn't imply that one is a prerequisite for the other. One could reasonably argue that self-directed agencies are an outcome of the evolutionary process, and predate evolution of intelligence, so the latter necessarily comes with those strings attached in nature. But we're past the "in nature" part by now.

I know that in my mental model of what intelligence is, a completely passive entity that responds to my questions in a certain way would qualify.

One definition I have heard is the ability to achieve a goal in more than one way.

Biological beings so indeed code information at multiple levels including genetic, epigenetic, at the cellular level in the form of interneuronal connections, etc. Whether you can that memory or not is irrelevant eyes, as it's still information that the biological being has encoded at some level to achieve some function

DNA persists, so of course behaviors in it also persist.

But can learned behavior persist in DNA? Could a gene activated by environmental triggers be later still be activeated after metamorphosis?

Maybe via epigenetic changes? From my limited knowledge of the topic, I can imagine that a species might evolve such that, if the larva is in an unusually hostile environment (too cold, too dry...) then a gene controlling the imago's behavior gets methylated.

I doubt. I've found the "proof" the butterfly remembers very dubious. A lot of modern studies are very low quality. Has anyone reproduced it? No. I doubt they will.

Reply on news.ycombinator.com