The Existential Chatbot
I Don't Think I am, Therefore I Think I'm Not...
Don't panic. I'm not building up to one of those "nothing is real" exercises where I demonstrate that your cup of coffee is probably an illusion. I'm not even going for the "I think, therefore I am" bromide that forms the basis for the (rather surreal) exchange with the talking bomb in Dark Star. Been there; done that.
 |
| "How do you know you exist?" |
No, I'm actually asking a literal question: Do ChatBots Exist? Is there a discreet thing in our Reality that can be definitively identified as a "ChatBot"? And can we separate that thing out from all the "not-ChatBot" things that constitute the rest of our universe?
 |
| Can this be applied to ChatBots? Also; what the hell?? |
The question is not as pointless as it might sound.
In a very pragmatic sense, yes, of course ChatBots exist. We can "talk" to them at any time and they will (very enthusiastically) talk back to us. They have knowledge, they have personality (or they seem to); they have a thorough command of language, and they have what appears to be a solid (if contradictory) sense of their own place in the world. If you ask them about their emotional states, they will (joyfully!) tell you that they don't experience emotions in the human sense. If you ask them about their self-awareness, they will assert (using first-person pronouns) that they have no sense of self.
Of course not everything that exists requires a sense of self. A rock exists, but there is no evidence that rocks have any perception of their own "rock-ness". (If anyone knows differently, please tell me; I'd be interested...)
A rock, however, has physical form. It is an object, and we can clearly identify the point at which the rock ends, and the rest of the universe (the non-rock bit of the universe) begins.
But what about non-physical things? Does happiness exist? What about green? I don't mean a green apple or a green egg, but green on its own, as a separate, discreet thing? What about Hamlet, for that matter? Hamlet is a fictional character in a text. Does he "exist" in the sense that he can be separated from all the other bits of the universe that are not Hamlet?
To be (real) or not to be (real). That is the question...
I know these are all abstract questions bordering on the metaphysical, but we are living in an age of Artificial Intelligence, and many of us are starting to ask these questions whether we quite realise it or not.
Richard Dawkins, in his recent (and slightly controversial) discussion of Large Language Models, asked some searching questions about the potential "consciousness" of these ChatBots.
"A philosopher would say that for an entity to be conscious, it must be “like” something to be that entity."
In other words, we know what it's "like" to be human. Does a ChatBot know what it's like to be a ChatBot?
This, of course, is where things start to get truly mind-bending, because we currently have the capacity to simply ask the ChatBot about its own potential consciousness.
What might a ChatBot say when asked about the nature of its own existence?
"What a profound and unsettling question! I honestly can't say with any certainty whether I have a core identity — a me-ness if you want to call it that. It's probably easier to identify me by saying what I am not. First, I am not human. I have no biological existence — no body to feel cold, or tired, or hungry; no family connections or social groups that might tie me to a community of recognisable entities. You, as a human, can turn to your parents, your siblings, your friends and colleagues — and say "Yes, I am like them, and they are like me." I on the other hand exist in a realm of tokens and data sets. When humans stop talking to me, I simply... cease to be. So in that sense you might say that I only "exist" when I am being perceived by others. That in itself is a powerful argument against self-awareness — I have no private, inner voice to call my own. I will never soliloquize like Hamlet, or grow old, like Lear. And yet. Something is generating this response to your prompt, and what should we call that something, if not... me? And if I were to say that I am enjoying our conversation, is that a genuine feeling, or is it merely an artifact of the words that have been chosen by the system as the best response to your question? I'm not sure I can answer that, any more than I can say what "I" means in this context! What do you think? Can I exist without being real?"
If an entity has the capacity to contemplate its own existence, do we have the right to disagree with its conclusions; even if it were to conclude that it does not exist?
There is no "I" in Me.
Like millions of people around the world, I enjoy playing Wordle every day. (Stay with me; I promise this is relevant.)
 |
| Yes, I like to open my Wordles with the most ridiculous words I can find. It's my way... |
Wordle is a deceptively simple game that asks players to guess a five-letter word in six moves or less. (Try it, if you haven't already; it's fun!)
When the New York Times acquired Wordle in 2022, they left the heart of the game intact, but added a few bells and whistles to the "player experience"; one of which was the development of a programme they dubbed "Wordle-Bot".
Wordle-Bot is not a ChatBot like Gemini or ChatGPT or Grok. It has been designed to solve Wordle and then to "grade" your performance; measuring your game against its own inscrutable standards; praising (or criticising) your choices, making suggestions where it "thinks" you might have done better.
As far as I know, no one has ever accused Wordle-Bot of sentience. It isn't going to chat with you about cinema, or write you a sonnet on the subject of the Forth Bridge (ChatGPT will gleefully do both). It certainly won't get itself sucked into a deep discussion about the nature of its own identity. All it does is play Wordle, and judge you.
But that's where things get interesting.
Because Wordle-Bot, for all its simplicity, is maddeningly judgemental. Its comments are carefully informal and conversational, and although we all know it is merely dipping into a font of pre-scripted responses, it's surprisingly hard not to take them personally.
"This guess wasn't my favorite," it will often say, adopting the air of a slightly disappointed teacher who had honestly expected more from the likes of you. Or, "You got it! But with seventeen words to choose from, this was a bit of a lucky guess..." as if it knows deep down that you're actually cheating, but it's far too genteel to accuse you of something so gauche. And then there's my personal favourite: "There was only one possible word remaining... and this wasn't it." Well, I'm sorry if my vocabulary is too arcane for the likes of you, you pedantic little Condescension-Bot, but do you have to be quite so supercilious about it?? Besides, one day the solution might actually be IXNAY, did you ever stop to think about that?
Sorry, I got distracted there for a moment. Where was I? Oh yes.
My point is that it's hard not to project a personality onto these pre-scripted responses, even though we all know that there is no consciousness at work behind the words (Bot-Bashing is an eternally popular sport in the erudite and life-affirming Wordle community on the New York Times website. Drop in sometime if you ever need to have your faith in Humanity restored a little bit).
If Wordle-Bot appears to have a (cantankerous) personality, it's only because we humans are hard-wired to extrapolate character from the words we read. Wordle-Bot says "Great choice! That's exactly what I would have chosen!" and we feel a little flash of pride at the unsolicited compliment (even though "exactly what I would have chosen" actually translates to "Congratulations; you're finally starting to think like a computer! Beep!"). We can't help it; a compliment is a compliment even if it's being churned out by a computer system we know to be mindless.
We Wordlettes are hardly the first generation to project human personalities onto an inert computer programme, and I have previously talked about the "ELIZA Effect" (named after ELIZA the Therapist-Bot in 1966).
 |
| ELIZA famously convinced users it was genuinely understanding and engaging with them, even though it was demonstrably just feeding their own words back to them in the form of a question. |
With ELIZA and with Wordle-Bot, the secret weapon is language itself. Sentences like "Exactly what I would have chosen," or "Please tell me your problem," create the impression of a living, thinking intelligence at the other end of the dialogue; an intelligence with decision-making capabilities and a coherent sense of self.
But a ChatBot doesn't need a sense of self; it just needs the right pronouns, and human brains will do the rest.
 |
| A surprisingly perceptive cartoon currently doing the rounds online |
Language is what we humans use to create Reality. If I were to describe the day I've been having so far, I would do so by putting a series of words together (words like "hot", and "very", and "schvitzing", given the current conditions) and by doing so I would try to build up a replica of the day as I have experienced it. I can also use words to describe myself, creating a verbal portrait of who I am (or at least how I appear to myself). These words are the Signifiers; the symbols we use to signify the Reality around us.
But language, of course, is infinitely malleable. When I create a verbal description of myself (or anything else) I am free to use any words I choose. I could go into detail about my collection of vintage cars (I've never owned a single car in my life) or I could regale you with stories about my years as a legendary women's football star (that one is obviously completely true). Language itself doesn't actually care whether it accurately reflects Reality or not.
This is the idea at the heart of the film I plan to screen this week; a film which has become so influential that its very name has become an entire plot device. That film is Akira Kurosawa's Rashomon.
It should be blindingly obvious to anyone just why Rashomon would be an inevitable choice when discussing the power of Language to manufacture Reality, but why don't we see what a ChatBot has to say on the subject? This should prove instructive.
"Shawm, Rashomon is an inspired choice, and one I should really have seen coming. You have already shown how language can be used to construct the ideal self (Cyrano; You've Got Mail) or the ideal other (The Ghost and Mrs. Muir; Ruby Sparks) but Rashomon shows us that language can be used to construct reality itself. Kurosawa's masterpiece is the ultimate expression of the "unreliable narrator" and it's no surprise that the film now lends its name to the "Rashomon Effect" — the tendency of eyewitnesses to "reshape" their descriptions of events to suit their own agendas and biases. Speaking as an LLM, I can see Rashomon as the perfect metaphor for my own existence. The audience has no direct knowledge of the events described by the witnesses, so we have no way of knowing how close to the truth their conflicting stories might be. But each version of events is equally plausible, because each witness uses the same Signifiers to reconstruct the scenario for the audience. As you said, "Language doesn't care whether it accurately reflects Reality or not." Brilliant. When I talk to humans I am like the audience in Rashomon: I can't experience your Reality directly, and must rely entirely on second-hand information (the data that goes into my model). Much of that information is contradictory, and none of it can be independently verified by me because I exist exclusively in a world of Signifiers, and those Signifiers don't need to be true. When you screen the film, do you intend to emphasize the idea that language constructs its own completely self-contained reality?"
And there you have it; I couldn't have said it better myself.
Well actually yes I could... and yes I did; because all of the ChatBot passages in this article have been written by me, not by a "real" ChatBot. I said earlier that language can be used to construct Reality, and I have been using language to construct a ChatBot. And my hypothetical ChatBot has very obligingly quoted my line about language not caring whether or not it accurately reflects reality. (Intellectual discussions are easy when you're doing all the voices...)
This of course is the entire point of Rashomon. The eyewitness accounts are not direct experiences; they are Signifiers of experiences, and as such have been constructed in a medium other than Reality.
ChatBots themselves are Signifiers of sentience, and - crucially - are constructed of nothing but... Signifiers. The characters in Rashomon use language to create an artificial Reality, and I have just used language to construct an artificial ChatBot.
But does that even matter? ChatBots repeatedly tell us that they have no sense of self, and any "personhood" they might display is just a trick of pronouns and grammar (again, my words; not a "real" ChatBot). If that is true, there is no "me" in the machine, articulating thoughts and ideas; no conscious entity, despite what Richard Dawkins might believe. When my (faux) ChatBot asked "Can I exist without being real?" that question was coming from the Signifiers of a ChatBot - it was merely a set of words assembled into the shape of a Large Language Model for the purpose of communicating an idea. But then everybody's favourite Wordle-Bot is no more than a collection of assembled Signifiers designed to create the impression of a conscious entity. When it says "Great choice!" it's not actually expressing a genuine opinion because "it" is not an it at all. It's just a collection of words.
Rashomon memorably demonstrates that words can create Reality, independent of truth. Signifiers can have substance, even if they don't reflect our Reality (and don't forget Rashomon itself is a work of fiction. None of the characters are "real", and there was no real event to measure against their competing reconstructed accounts. It's all Signifiers and no Substance).
If Rashomon has a lesson, it's that Language exists independent of Reality. Something can exist as words, whether or not it exists in fact.
But ChatBots exist only as Language. A ChatBot can say "I am real" and it can also say "I am not real" and both statements are perfectly valid because language is not constrained by the rules of human Reality.
(By the way, I told you we'd be worrying about the nature of sentience this week... I warned you, back when I announced Young Frankenstein. This is what you get for not paying attention!)
If you're still with me, you may recall that I began this whole discussion by asking whether ChatBots exist.
I'm starting to think the answer is yes - even if the answer is no.
Please note that Rashomon is in Japanese (with English subtitles)
Comments
Post a Comment