Can an Artificial Intelligence (AI) be the same as a person? Earlier this year I gave a talk to local Catholic group on the subject of science and belief. My point in the talk was that the things we think about science and belief are influenced very heavily by the world, by the conventional wisdom, that surrounds us. I used two examples in this talk — the first chapter of Genesis, and life on other worlds — but I think there could have been an AI connection to the talk as well.
Regarding Genesis, my point was that the sorts of Genesis-related science and faith discussions that we hear about tend to connect to the idea that the world was created in six twenty-four-hour days, 6000 years ago. Such science and faith discussions might be in support of the 6-24-6000 idea, or they might be in opposition to it, but if Genesis is part of the discussion, 6-24-6000 is probably there. And it is there because that is what we hear about; 6-24-6000 dominates the conventional wisdom. As I have discussed several times here at Sacred Space Astronomy, there is a conflict between the first chapter of Genesis and the fact that science has revealed that stars are larger than the moon. We don’t hear about that. We hear about 6-24-6000, even though in Genesis 1 the sun is not created until the fourth day, and therefore the mornings and evenings of the first three days clearly must be metaphorical in some sense; with no sun there could be no mornings and evenings as we know them to demark the first three “days” in Genesis 1. By contrast, there is no metaphoric fuzziness to the question of the moon and the stars; Genesis calls the moon one of the great lights of the night sky, but astronomy says that the moon is very small versus the stars. So why doesn’t that moon-and-stars question dominate Genesis-related science and faith discussions? I would say it is because we can’t escape the conventional wisdom that tells us that the Genesis-related science and faith question is 6-24-6000.
Regarding life on other worlds, discussions tend to connect to the idea that science has shown that the universe is so vast that there is sure to be intelligent life out there that we are likely to run into (or already have run into), and what will that mean for society and religion, etc.? Discussions don’t tend to connect to other things that science has shown, like the diversity of stars and planets within the universe (they are not all earths and suns), or like the fact that the ancient idea of spontaneous generation (the idea that life arises commonly and regularly from inanimate matter, and that you can observe this happening for yourself if you know where to look) was bogus. This matters for the idea of life on other worlds, and Dennis Danielson and I have significant discussion of it in our book A Universe of Earths: Our Planet and Other Worlds from Copernicus to NASA. Whether or not the universe is diverse, and whether or not life is a regular consequence of matter, is going to bear on the question of whether or not lots of planets have life on them. And if there are not lots of planets with life, then the vastness of the universe will make finding any that do have life very unlikely. But the conventional wisdom is that science has shown that the universe is so vast that there is sure to be intelligent life out there, and that idea is almost impossible to get away from.
I think there is something similar going on with discussions of Artificial Intelligence (AI). Consider this, stated by Sam Altman, CEO of OpenAI, in December of 2022, shortly after the release of OpenAI’s ChatGPT.
i am a stochastic parrot, and so r u
That is,
I am a stochastic parrot, and so are you.
His term “stochastic parrot” referred to a 2021 paper by Emily M. Bender et al. about Large Language Model AIs such as ChatGPT. In the paper, the authors had stated:
Contrary to how it may seem when we observe its output, an LM [language model] is a system for haphazardly stitching together sequences of linguistic forms it has observed in its vast training data, according to probabilistic information about how they combine, but without any reference to meaning: a stochastic parrot.
A pair of ordinary six-sided dice are stochastic — they are random, but you can say that a roll of seven is statistically more likely than a roll of two. Likewise cards — you cannot predict what cards you will be dealt, but you know that you are less likely to be dealt four aces than to be dealt four face cards, and less likely to be dealt four face cards than four number cards. And likewise words — “Mary had a little lamb, his fleece was white as…” is rather more likely to be followed by “snow” than by “cream cheese”.
So an LM produces a “seemingly coherent” (as Bender et al. say) text in response to some prompt or question, but what it is really doing is mindlessly squawking words according to some algorithm and a large pile of data that can be statistically processed — it is a stochastic parrot. Altman asserts that we are all stochastic parrots — nothing different from an AI; nothing more than seemingly coherent.
The same week that I gave the talk mentioned at the start of this post, I got into a conversation with a talented and accomplished friend of mine who essentially asserted the same thing. He said there was no difference between what went on in our heads and what went on in an AI. This is a point of some distress for him, because it logically follows that an AI, if no different from a human, should be able to do anything a human can do, and he was concerned about the future of his two young daughters and whether there would be any career opportunity open for them since anything they could do could be done by an AI.
I think this presumed equivalence between humans and AIs, this presumption that we are all stochastic parrots, is another one of those ideas that has a lot to do with a sort of conventional wisdom that surrounds us. It does not withstand careful scrutiny.
Consider that an algorithm is a lot like something from an IRS tax worksheet (“enter amount from line 15a; if amount is greater than line 14d, subtract…”) — that is, an algorithm is a recipe of sorts that says, for example, that if x is greater than y, then x is to be subtracted from z, otherwise twice y is is to be subtracted from z, and blah blah blah…. You can do your taxes by hand, on paper, in which case you yourself work through the algorithm. Or, you can use tax software, in which case the computer works through the algorithm. Either way, your taxes are calculated.
If the computer is doing your taxes, then there are electric switches at work. Modern electronic computing technology is based on binary logic — 0 and 1; open switches (0) and closed switches (1); either/or; logic gates that mean that if this switch is opened, that switch will close, etc.; the numbers 0-1-2-3-4-5 being 0-1-10-11-100-101, etc.; the numbers of transistors (switches) on a computer chip doubling every two years (the famous “Moore’s Law”). All those things are part and parcel of modern computing technology. If you are using a modern computer to do your taxes, then you are essentially using a gazillion switches opening and closing under certain programmed conditions.
However, we can imagine all this happening in many ways that do not involve modern electronic technology at all. We might imagine, instead of electric switches, a purely mechanical system, with gears and cogs and levers going this way and that, serving to hold and process information. The first calculating machines were mechanical. The computer that Charles Babbage envisioned and designed in the nineteenth century was to be mechanical:
You could use a Babbage-style system to do your taxes. So, couldn’t a mechanical Babbage-style system conceivably be used to run a Large Language Model, a stochastic parrot, rather than modern electronic technology?
Would Altman be asserting that we are all stochastic parrots if ChatGPT ran on a Babbage computer? Consider — the hype surrounding modern electronic technology is going to distort our perception regarding all things AI. Many of us can’t put down our modern electronic technology even when we are behind the wheel of a car and our life and health, and the lives and health of others, are at stake. Would anyone be persuaded by the idea that a less-hyped technology, like a big mechanism of gears and levers, was equivalent to us? Or can Altman only say what he says because years of hype and electronic AIs in popular culture (like HAL 9000 or Lt. Cmdr. Data) have established a conventional wisdom that renders modern electronic technology an acceptable stand-in for a mind?
Perhaps Altman would still make his assertion. But what is the basis for asserting that we are no different than a collection of switches (or gears)? The following bit of conventional wisdom does not hold water: “Well, human beings are just collections of cells and we think we’re sentient and have feelings, so why can’t a vast collection of switches, or gears, be sentient and have feelings? And if that collection of switches is a stochastic parrot, why are we not?”
It does not hold water because we know what there is to know about a switch or a gear. Even if you grant that a human being consists of nothing but cells, and that there is no soul and no God and nothing beyond the material world, we don’t even know how a living cell is formed from inanimate matter or what differentiates it from its material components. Here we connect back to the idea of spontaneous generation and its demise at the hands of science in the nineteenth century.
This matters for claims that a computer algorithm can be the same thing as something that’s alive. Once we thought life was a regular, natural product of matter. Now we don’t, by which I mean we do not ever see life arising from lifeless matter in nature and we cannot make it happen in a laboratory (both of which were thought to be observed in the heyday of spontaneous generation). In a certain sense, how to make a living thing, even a simple cell, is a mystery to us. A switch, by contrast, is no mystery. Our best scientists can’t build a cell; a kid can build an electric switch from some nails, a paper clip, and a short piece of two-by-four. A cell is a complex thing whose workings elude us in a very crucial way; a switch is not. So what is the rational basis (not the conventional wisdom basis) for equating the two?
This is not to say that we can’t use an algorithm and technology to mimic something that is alive. Large Language Model AIs can certainly mimic a human mind. They might mimic it well enough to do many jobs. Moreover, they might mimic it so well that how we interact with them matters. For example, being abusive toward something that responds like a human would respond is probably not a good thing, because it stands to reason that we might not be able to avoid transferring such abusiveness over to our interactions with real humans. Consider that even Bender et al. themselves seem unable to avoid portraying a language model in human-like terms, even as they call it a “stochastic parrot”. Note that they write of the “linguistic forms it has observed in its vast training data” rather than, for example, the “forms it has processed” or “sifted”.
But the fact is, it makes sense to think of a collection of switches as a stochastic parrot, and not to think of it in human-like terms. The fundamental difference between a collection of switches and a living thing means there is no real basis for saying that an AI might be the same as a person, such that we would all be stochastic parrots, or that everything we can do could be done by an AI — even if it fits the conventional wisdom to say otherwise.
Click here for other AI-related posts.
P.S. The gears-and-levers parrot at top? Made it with AI!
