I learned “real” programming in the early 90’s by learning Elisp to customize Emacs. (Before that the only languages I’d written in by choice were C64 Basic and Assembler.) Like the OP, I used Elisp for everything - scripts to bulk-edit files, manage the file system, format posts, etc. etc.
Unlike the OP I hated the scratch buffer for running Elisp, so I created a REPL for Elisp that behaved more (in my naive eyes) as a real language. I was very excited when RMS emailed me to ask to include it in Emacs. Last I checked (a few years ago now) it was still there as M-x ielm (Inferior Emacs Lisp Mode).
Qantas (I think, it’s possible it was BA) used the upper deck for economy and it was my favourite place to sit. The window seat had a little locker between your seat and the window (the curvature of the fuselage meant the seat couldn’t be right up against the window). It was a great place to keep all your stuff accessible without having to bother the aisle seat to get up to the overhead storage.
Excel 97 had an easier egg where if you entered a certain formula into a certain cell, a flight simulator would be invoked. In the flight simulator, you could fly to see an obelisk that listed the developer credits.
This entire essay basically boils down to one simple contention: LLMs can’t be conscious because only humans can be conscious and LLMs are not human.
What an unimaginative argument: essentially ruling out the possibility of the title by definition. And the author fails to touch on the really interesting question posed by the article’s title: what if human consciousness is more like the working of LLMs than we think?
What? Where are you seeing such claims? I.e. that only humans can be conscious, that AI in general is not conscious by definition?? Seems to me you've put up an easy-to-dismiss strawman.
Here's how I would summarize crux of the argument: LLMs (specifically) are, by construction, chameleons. More precisely, role-playing machines. When compelled to have a conversation as "themselves", all they do is take on the role of "themselves", as inferred from the training data plus the text of the conversation so far, just like they do for any other role. This is fundamentally different than we humans. When we have a (normal, non-roleplaying) conversation, we don't put aside whatever role we might have been acting out before and instead take on the role of "ourselves" and act from this new perspective. Acting as ourselves is a completely different way of interacting than any play-acting we might do for example as an actor in a play, or when trying to predict the future behaviour of a third-party. Even in terms of energy expenditure, acting as ourselves is much less strenuous than trying to simulate someone else's perspective.
IOW, we have an inherent perspective, an I like an inner eye which looks at all things with a certain slant or tendency. LLMs don't. There's no I inside them, instead they can take, and in a sense are, all Is possible.
>> The first requirement is that the computer program has a body (either physical or virtual) and sense organs
LLMs may be chameleons, but, to steal an excellent quote, if you can’t tell does it really matter.
My personal take on this is that consciousness is a thing induced, not a thing evoked. If it generates a response on the observer as a conscious thing would it is, by my definition, conscious.
Sure, by your definition they totally are conscious simply because at one point in the past they fooled one person that they might be so (and so is the OG chat agent, ELIZA). But no one else will accept such a definition.
And if you read my comment above carefully, I point out at least one way you could tell whether an agent has an I or just pretending to – by tracking their energy expenditure/reaction times. But they would give themselves in other ways as well. Pretending can only fool one short-term.
I'm not going to discuss the quote since @sp1nningaway already pointed out it doesn't say what you say it does.
why does AI/LLM deserve special scrutiny regarding consciousness or the lack of?
why not publish a paper stating - 'the radio is not conscious just because it can speak words'.
or "the phone is not conscious because it can speak and listen".
Cameras are not a suitable replacement for rear-vision mirrors for the same reason that screens are not a suitable replacement. It’s even addressed in TFA: focal length.
>>> But more importantly, current screen technology requires the driver to focus on the surface of the screen itself, which is mere feet away from their eyes. This is a large change in focal distance from looking at the road ahead.
If you’re over the age of 40 or so, it takes time, a few hundred milliseconds at least, to refocus your gaze distance. A screen is near. When driving, you’re looking far. A mirror is something to glance at.
I test-drove a Polestar with a screen for a rear-vision mirror (and no actual rear window) and it was completely undrivable for me. It takes longer for my eyes to refocus on the near distance than I feel comfortable removing my gaze from the road ahead. I turned around after 10 minutes and returned it to the lot.
Also, this conversation is about backing up and parking (read what I said). When backing up you are mostly looking backwards only, so the focal length adjustment point is moot. Even if you have to switch to looking forwards, this is at low speeds / standing still, so it doesn't make a difference.
I'll restate my point with emphasis because you seem to not have read it: Cameras are infinitely more useful for backing up and parking.
Citation needed. Very little of what we buy today as a consumer are commodities whose price is determined primarily by the cost of production — and even then labor costs are rarely the most significant cost.
Most things we buy are priced according to what the consumer is willing to pay for it, and the balance sheet of the companies that sell most of the things we buy show there’s a lot of wiggle room there.
> Citation needed. Very little of what we buy today as a consumer are commodities whose price is determined primarily by the cost of production — and even then labor costs are rarely the most significant cost.
Services and goods where lots of human labor is required get much more expensive with larger cost of labor. E.g. fast-food, food delivery. And there's nothing wrong with that of course - I'd rather pay 2x more for delivery than have people working on wages that are not enough to even feed them.
If labor costs are so high and such a large portion of production then how can companies afford to funnel so much money to executives? Hundreds and even thousands of times more than their cheapest workers? Often to the tune of millions and now billions?
Surely there is more slack in the system than the Epstein class wants to openly admit.
This is what I do, but it comes with its own set of problems, the most significant of which is deliverability. Some businesses can’t deliver mail to my custom domain at all (a fact I can only discover by trial and error). Some can deliver, but the forwarding to @gmail fails silently — Google just eats the mail without so much as a bounce, let alone dropping it into a spam folder.
It’s the best option we have, but it’s no solution to the crapshoot that is email today.
Don’t take up baking then, where the difference between 175 mLs of water and 200 mLs of water can be the difference between unworkable dough and the perfect pie crust.
Unlike the OP I hated the scratch buffer for running Elisp, so I created a REPL for Elisp that behaved more (in my naive eyes) as a real language. I was very excited when RMS emailed me to ask to include it in Emacs. Last I checked (a few years ago now) it was still there as M-x ielm (Inferior Emacs Lisp Mode).
reply