I was asking my AI thinking companion (this time Grok 4.1) about hidden truths that AI uncovers about us, humans.
That was after a fun accident - i used a prompt to draw a light warrior fighting an evil monster, got what i got, and then i told him the warrior is purifying the evil from the world and the warrior became female.
Why? what changed? I asked grok. So our human collection of imagery and art is bombarded with biases and stereotypes, some at racism level. Quite a fun revelation about ourselves. You should try it.
So i asked for more secrets.
It spitted "The “next-token oracle” effect".
Say Whaaaat??
At the end of the day, ANY AI model, with whatever reasoning, cognitive abilities, smart and awesome, eventually is "just" predicting the next token (word or half word, usually).
But wait, it can solve AMAZING things no human can do like the protein folding problem, do already a better job that humans like better than docs in diagnosis, lawyers at contract review, coding, chess, translations, summarizing, and even beats psychologists at therapy scripts.
Soooo.... thats it? all of our leading and great minds are just people better than average auto-completers?
Dreyfus model of skill acquisition actually defines that. In super short - the difference between the novice and the expert, is the (huge) amount of exposure to the profession, field, subject, or just item or platform.
With enough exposure us human starts to have unconscious deep understanding and internal supercomputing powers we often refer as intuition, "gut feeling" or detached decision making.
2 simple examples:
Driving - experienced driver can go via rough road while speaking to someone and actively consciously thinking hard about another problem, math, coding, or shopping list. So ZERO effort and conscious thinking about the road, he even cant remember his drive and obstacles. So how he does it?
Detached decision making - internal, unconscious, faster and better than the young conscious driver thinking about every obstacles.
Math - know those (normal) kids that just calculate at the market super fast? They just enjoy doing math at home, like 845684+448552 = ???. They LOVE it, doing it all day long, just for fun.
Their brain is so trained and math calculation they will solve something like this faster than you can write in in the calculator.
Even simple stuff, is 300ml beer at 5$ is cheaper that 500ml at 8$? How fast do you just say 300*3=900=15$ VS 500*2=1000=16$? And then how fast you think the missing 100ml is 5/3=1.6$?
YEP. That is kinda the premise of the base tech for any AI model called Neural Network, we worked hard to mimic our brain's Neural Network and the results are those bots we call AI. We are still 100 years at least from a really real AI, but those next-word-predictors we have today are really AMAZING!
And that is why the reflect us do good, for better or worse, or just understand ourselves. If AI can do a better job than my doctor, or better than me at coding, then what am I doing? what is my doctor doing?
If AI is just predicting the next word - then i guess that is what I do as well...
So what is Experience that we relay on so much? In Grok's words:
“most likely next token” given everything that came before it in your life.
My attempts to ask Grok to challenge me all were around humans Abstract thinking like Einstein's work, and other great inventors. Point. But eventually they all described in a new manner (math manner with this example) our reality that we all experience every day.
Novel problems and thinking? Not for my liking. Ask anyone to describe an alien. I am an orthodox jew so I dont believe in such a thing. But as an exercise, what would they look like? how would they communicate?
Even the most creative people will suggest seeing by water waves and hearing by nuclear micro-self-induced-explosions, ALL concepts we KNOW.
Just learn about the octopus - the most strange creature on earth than human, we cant even decide if those each self-decision-taking tentacles are independent brains, meaning he has 9 brains, or just extensions of the 1 brain but with independent cognitive abilities. That is how limited we are on a lightly foreign creature.
I NEVER saw even 1 description of any really real detached of our knowledge about aliens. Always very similar to what we know.
Therefor all humans has Abstract thinking is, again, supercomputing our vast context of knowledge and experience. In AI we call it Training.
Us human has a natural ability to bind different subjects to each other. Even if they dont really bind. If you are married you know it - its absolutely YOUR fault that the tomato's price was high today, as your behavior this morning got her to go to her sister to talk BEFORE going to the supermarket.
But that is a feature, not a bug. ANY great chef is doing that exactly every day, testing the bindings of new things. All kids are chef.....
But eventually its all "just" pattern-matching by trial and error. AI is great in pattern-matching given enough tokens/data/instructions ect.
So yes, we can do trial and error, and we can do better job with newer/abstract things (did you know chocolate cake tastes better with few drops of salts before putting in the oven?). But AI does much better job on vast data and can recognize patterns where we cant. And with enough data he can do just like us and better even with newer/abstract things.
Humans has almost 100 years of experience tokenizing text and images (turning them to math or values than computers can understand).
We have Zero experience about all that with the 3D real world and senses stuff. How you describe taste, smell, touch? We described dog, cat, skies, ect.
Therefor any claim about AI's inability about those fields, its just a tokenization issue.
On the 1 hand, its humbling. We are all just bunch of token oracles.
On the other hand, its AMAZING HOW MUCH INTELLIGENCE is compacted to just token prediction!
Our brains multi-function ability to master a skill and operate it in light speed supercomputing in the unconscious parts to enable us to always expand our learning and thinking and so awesome!
I have 2 thoughts from here.
About AI - think of how much more we can achieve with it if we could mimic internal "unconscious " and external "conscious " thinking? If its changes us from disabled infants to top level experts, it could make our AI models maybe be able to solve Time itself? PS as an orthodox jew I dont believe in time travel, but yes in space-time seeing/visioning.
About humans - beyond the fact that anyone can become a world leading expert on something (that he likes/do), think about exploring non-prediction abilities, that can really super-jump us. Think about spiritual, emotions, ACTIVE learning and thinking (unlike reactive - going to school...).
The face we can gather knowledge from one another by listening, still like AI Training, and he can even mimic compassion. But real Love, real spiritual experience, that is all "ours".
And while the big giant do work on that, we still own long term doing of anything, like really mentoring our kids and employees. AI can teach better than me A math problem. But he cant daily mentor and encourage the kid to be a math hero.
What do you think? leave a comment.
Need AI consult? Send Whatsapp or Email