AI: Particle or Wave?

Is ChatGPT Intelligent?

Deep Learning is Hitting a Wall

Reasoning or Reciting? Exploring the Capabilities and Limitations of Language Models Through Counterfactual Tasks

FANTOM: A Benchmark for Stress-testing Machine Theory of Mind in Interactions

The Medium digest before the one that pointed me to Is ChatGPT Intelligent? and Deep Learning is Hitting a Wall (the former links to the last two of the linked articles above) was one that set my teeth on edge. Two articles talked about Ouija boards, one comparing ?ChatGPT 5? to an Ouija board, and the other alleging it talked about the [legitimate] science besides Ouija boards, and the image of that accursed tool displayed on my computer screen. I thought about discussing disengaging from the channel with my abbot, as I asked for and received a blessing to disengage from Twitter after watching a suicide and then someone else getting shot to death. And I may do so. I already read it despite a miserable signal to noise ratio, and my main mental efforts in reading through the digests is finding legitimately useful information amidst a 90% of sheer sensationalist drivel. But the last two articles, and in particular Is ChatGPT Intelligent?, left me wondering if I was actually right in substance about AI in my 2004 thesis at Cambridge, AI as an Arena for Magical Thinking Among Skeptics.

In AIlice in Wonderland, the last section was entitled "Alcasan's Head," a reference to C.S. Lewis, That Hideous Strength, where a guillotined prisoner's head has been artificially kept alive and overclocked with stimulants, but is in fact acting as a demonic gateway for commerce with evil spirits. And I repeat a warning from a friend that asking wondrous things of ChatGPT can be like approaching a Ouija board.

Years ago, in my bachelor's or previous master's program (before 1998), before I wrote AI as an Arena for Magical Thinking Among Skeptics, someone commented about how artificial intelligence can learn how to move through a room, but if you "put a cup in the center of the room," it has to learn all over. Deep Learning is Hitting a Wall said,

Not long ago, for example, a Tesla in so-called "Full Self Driving Mode" encountered a person holding up a stop sign in the middle of a road. The car failed to recognize the person (partly obscured by the stop sign) and the stop sign (out of its usual context on the side of a road); the human driver had to take over. The scene was far enough outside of the training database that the system had no idea what to do.

And humans do not need to be specifically trained to avoid driving over a sign-turner at a construction site, even if their driver's education has never mentioned sign-turners. Elsewhere, Is ChatGPT Intelligent? reads:

In chess, GPT was asked to evaluate whether a sequence of moves was legal or not. For a normal chess game, it accurately predicted the legality of a move 88% of the time. But when the starting positions of the bishops and knights were swapped, its guesses on the legality of moves became completely random, while even a novice human player should be able to adapt to these changes easily.

The first comment shown on that article, as of this writing reads:

While reading the article, I tried the reversal curse method on ChatGPT. I gave it a prompt "Martin is my brother" followed by "Who is Martin?" and I got a wrong answer. The possibility of getting to AGI is subject to our understanding of our own mind and cognition. And I don't think we are even close to where AGI starts. It was a good read.

I realize in retrospect that my original quest in evaluating ChatGPT was asking for some hidden wonder when I essentially asked for the never-written eighth book from The Chronicles of Narnia where, after the prequel The Magician's Nephew but before any of the other books, a King of Narnia delivered the Lone Islands from a dragon and in gratitude was granted the title Emperor of the Lone Islands. However, persistent efforts met with disappointment; a repeatedly tweaked prompt of "You are C.S. Lewis writing in the style of the Chronicles of Narnia" did not secure much above mediocre fiction, a stinginess with words, and a failure to observe guidelines observed in writing even mediocre fiction, such as "Show, don't tell."

The AI Dilemma credits ChatGPT with acquiring abilities its implementors never imagined, and the scientists working on AI both feel that there is something transcendent and have nonetheless goosebumps about getting into creepy turf. The last video I saw credits AI with an above-average-adult-level Theory of Minds, but the FANToM article raises serious doubts about whether that alleged Theory of Minds is real or just smoke and mirrors yielding false positive results, and there are bits and pieces like "Since machines lack emotions or intentions (Gros et al., 2022)..." that suggest my thesis was right about the limitations of AI, even if AI makes those limitations much less significant than one might imagine.

However, concerning the different strands of debate, I might appeal to the debate that raged in physics about whether light was a particle or a wave. The resolution came that it was both: light acts as a particle when treated like a particle, and acts like a wave when treated like a wave.

I might make a similar suggestion that whether ChatGPT is something that can pull an accurate guess about what comes next in territory close to the training data it has processed, and being a can of dragons and a gateway to the demonic, is something like this particle / wave paradox. Generative AI can guess what's next in a conversation along similar lines to what I conceived of during my first master's, and this can simultaneously both be something that falls far short of genuine human intelligence with its generalizability, and something spooky where demons can fill in. This BOTH-AND quality recalls the Pascal quote that opens Everyday Saints:

Openly appearing to those who look for Him with all their heart, while hiding from those who run from Him with all their heart, God governs the knowledge of His presence. He gives signs that are visible to those who search for them, and yet invisible to those who are indifferent to Him. To those who wish to see, God gives sufficient light; to those who do not wish to see, He gives sufficient darkness.

All that I can recall reading either sees AI as fundamentally less than regular human minds, or something spookily beyond regular human minds. I propose a BOTH-AND, in the same vein of light as both a particle and a wave, or a God who gives both sufficient light and sufficient darkness.

I think the comments about AI in Within the Steel Orb might well be valid:

Oinos said, "Let me show you." He led Art into a long corridor with smooth walls and a round arch at top. A faint blue glow followed them, vanishing at the edges. Art said, "Do you think it will be long before our world has full artificial intelligence?"

Oinos said, "Hmm... Programming artificial intelligence on a computer is not that much more complex than getting a stone to lay an egg."

Art said, "But our scientists are making progress. Your advanced world has artificial intelligence, right?"

Oinos said, "Why on earth would we be able to do that? Why would that even be a goal?"

"You have computers, right?"

"Yes, indeed; the table that I used to call up a scientific calculator works on the same principle as your world's computers. I could almost say that inventing a new kind of computer is a rite of passage among serious inventors, or at least that's the closest term your world would have."

"And your computer science is pretty advanced, right? Much more advanced than ours?"

"We know things that the trajectory of computer science in your world will never reach because it is not pointed in the right direction." Oinos tapped the wall and arcs of pale blue light spun out.

"Then you should be well beyond the point of making artificial intelligence."

"Why on a million, million worlds should we ever be able to do that? Or even think that is something we could accomplish?"

"Well, if I can be obvious, the brain is a computer, and the mind is its software."

"Is it?"

"What else could the mind be?"

"What else could the mind be? What about an altar at which to worship? A workshop? A bridge between Heaven and earth, a meeting place where eternity meets time? A treasury in which to gather riches? A spark of divine fire? A line in a strong grid? A river, ever flowing, ever full? A tree reaching to Heaven while its roots grasp the earth? A mountain made immovable for the greatest storm? A home in which to live and a ship by which to sail? A constellation of stars? A temple that sanctifies the earth? A force to draw things in? A captain directing a starship or a voyager who can travel without? A diamond forged over aeons from of old? A perpetual motion machine that is simply impossible but functions anyway? A faithful manuscript by which an ancient book passes on? A showcase of holy icons? A mirror, clear or clouded? A wind which can never be pinned down? A haunting moment? A home with which to welcome others, and a mouth with which to kiss? A strand of a web? An acrobat balancing for his whole life long on a slender crystalline prism between two chasms? A protecting veil and a concealing mist? An eye to glimpse the uncreated Light as the world moves on its way? A rift yawning into the depths of the earth? A kairometer, both primeval and young? A—"

"All right, all right! I get the idea, and that's some pretty lovely poetry. (What's a kairometer?) These are all very beautiful metaphors for the mind, but I am interested in what the mind is literally."

"Then it might interest you to hear that your world's computer is also a metaphor for the mind. A good and poetic metaphor, perhaps, but a metaphor, and one that is better to balance with other complementary metaphors. It is the habit of some in your world to understand the human mind through the metaphor of the latest technology for you to be infatuated with. Today, the mind is a computer, or something like that. Before you had the computer, 'You're just wired that way' because the brain or the mind or whatever is a wired-up telephone exchange, the telephone exchange being your previous object of technological infatuation, before the computer. Admittedly, 'the mind is a computer' is an attractive metaphor. But there is some fundamental confusion in taking that metaphor literally and assuming that, since the mind is a computer, all you have to do is make some more progress with technology and research and you can give a computer an intelligent mind."

And at the same time, I believe that AI may be a gateway to contact with dark forces. I propose a BOTH-AND. And I suggest that the both-and be held onto both in revisiting a classic saying from Tristan from Humane Tech:

Give a man a fish, and he will eat for a day.

Teach a man to fish, and he will eat for a lifetime.

Ask AI to fish, and it will learn oceanography, climatology, evolutionary biology… and fish all the fish in the ocean to extinction.

What do you think?