“Our best current model of the human brain works like this; We made a computer program based on this model; Therefore this computer program is in effect the same sort of thing as a human brain” does not seem like an especially strong argument tbh
Thousands of 19thC people worked SO hard to convince their contemporaries that this was somehow possible in an abstract way. To see our contemporaries so eager to swallow the myth is frankly unexpected and yet entirely unsurprising.
I find that essay so frustrating.
The main point is correct — the human brain is not a Church-Turning device!
But then he insists that the centuries-old figure of speech “drawn from memory” is a metaphor derived from digital computing.
I feel like the worst thing Alan Turing did was establish an arbitrary metric for "full humanity" by positing a machine that could rapidly generate bullshit answers indistinguishable from human bullshit answers, that later generations viewed as received Wisdom that cannot be questioned.
That we still use a thought experiment goal from a very smart man in 1950, when "computers" were room-sized machines with vacuum tubes that could just compute trajectories, & zeitgest futurism-robots looked like this, to act as the benchmark for "intelligence" seems like a quasi-religious impulse.
Like, all it did was create a weird contest and arms race to build a reflexive language-game-simulator. As it turns out, "intelligence" has, like, other vectors.
good thing he didn't actually do that!
he just said "what if a machine could fool a human?" aaaand that's it.
it was a very revolutionary thing for his time.
Oh, I don't blame Turing. It's more the kind of Moses-like-reverence paid to the idea of him after his death that can't see "the Turing Test" as an essentially dumb & meaningless metric.
There is probably a list of horrible claims of legitimacy by analogy to something presumably great. One of my recent favorites is the Shapley value. It's great because, game theory, you know, like in that movie with the Gladiator guy
And that's with models and other non LLM technologies where most attention has been given to exploring capabilities and minimal attention has been given to optimization ---
Except nobody is making this argument (in fact amusingly enough LLMs are non-biomorphic plus biobrains certainly don't implement backpropagation for state updates) and computer is a vast term that encompasses so many systems anything that ever tangentially processed "information" might be one
It is not the complete argument because it is missing the explanation what the correct model of the human brain would. If we build the system with syntethic cells - would that do any better? This reply of mine isn't meant to be the last word in this matter, it can't be, the science is too young.
The brain certainly is not a touring machine and sucks at tasks that touring machines are good at. A brain, the neocortex to be precise, is for the most parts a neural network. Touring machines can emulate neurons with software. If these neurons are coded to work similiar to real neurons there is
much ground to cover before we can make any further statements.
The main problem is that the main figures behind this A.I.-Complex are not the trust worthiest people. There are linked to Elon Musk a knowm hoaxter and con man, lying about technological achievements and possibilities. It all needs ro
This was a fascinating article; thank you for sharing it. It’s incomplete of course as it provides no real movement forward, but one can’t disagree with the fundamental premise.
Good stuff.