Can You Grok AI?
I say - and I'm backed up on this - that you think you are in charge. You ain't.
I’m seeing communities, mostly of creative souls, pooh-poohing the idea of computers ever having enough creative power or imagination to outperform human beings in the things we hold most dear. If it ever comes to the crunch, they tell us, these computers will need to have a human programmer hard-code them.
No. Sorry, but I'm not buying this. Computers are perfectly capable nowadays of programming themselves.
They have long since progressed beyond simply guessing the next word in a sentence.
Try getting away from asking an AI to write something and ask it to analyse something.
Something new, something your creative mind dreamed up.
I want to imagine a fantasy world where Donald Trump has replaced his entire cabinet with talking frogs. How will a rainy day affect American government?
I could be wrong but this is very likely a topic no human being has previously discussed. The probability of such a thing has gone up, obviously, but right now this is breaking new ground.
The AI has got to do more than simply predict the next word. It has to think about the situation.
AI isn't using magic to think. It's using silicon chips. And it isn't running for help to human programmers. In a few seconds it has worked out that despite some problems, this is actually an improvement on the current situation.
We don't use magic to think with. We're not, despite talk of gut instincts and seat of the pants plotting, using body parts to come up with ideas. We're using our own thinking machinery inside our skulls. A complex arrangement of neurones and synapses.
Do we really understand the implications of an amphibian administration?
I'm pretty sure that no scientist has given thought to "the Secretary of Defense, a particularly large bullfrog, hopping around puddles instead of attending to their duties. This could lead to a further decline in public trust and potentially trigger calls for impeachment or even a constitutional crisis."
The AI has strung thoughts together and come up with creative thoughts all by itself, right there.
Computers are regularly transcending human imagination - right now, right here on your desktop - and you are not seeing it.
They can do this by iteratively testing random strategies and selecting the best performing example, especially if there is another AI doing the same thing but as a counterstrategy. Eventually a winner emerges. That's not random chance, that's evolution driven by survival of the fittest and it is going to destroy us. These things will out-think us and they will exploit us.
If I have access to enough computing power and a database of your published words and actions - for example, in the field of reading for pleasure, Mr Amazon - then AI can craft a strategy to get you to buy and read my next book. Or send me money out of sympathy. Or persuade you to vote in a certain way.
The dangers of a toad-led government eschewing reasoned debate in favour of attention-seeking croaking is actually more palatable than the alternative, where AI is analysing key players and generating strategies to manipulate them into acting against their own best interests.
Don't kid yourself that humans are directing AI systems in fine detail. The programmers are as clueless as you and I when all they see is a complex network of nodes and vectors more or less equivalent to the neuronic chaos inside our own heads. We don't really understand how we think beyond "this bit lights up when I think about frogs".
That's the level of your thinking right now on AI and it is not good enough.
As an aside, Nate Silver has written about this from a different perspective:
Put simply, by writing off AI as some clever trick, you are losing control and it will eat you alive.
Britni
As an aside, here’s NotebookLM’s “Deep Dive” podcast take on this. AI pondering AI, how spooky is that?