The more far-reaching the different pieces of information to combine, and the more interestingly they get combined the more unique and creative the product is. Polymaths have always made important breakthroughs in fields for this reason and their importance seems to only be increasing.
I wrote a paper 40 years ago on an idea I came up with for a different translation of Exodus 6:3, bounced it off my professor, he said make sure you search all the indeces, you'll probably find some obscure journal printed that idea 100 years ago.
Well, 40 years later, finishing the paper, I did find an obscure journal that came close, 90 years ago.
And AI has been amazingly helpful in research and writing - some of them are quite good at ancient languages, including hieroglyphics and cuneiform.
Your "thrive in the unrecorded" point is the one I keep coming back to. If intelligence is mostly search, the most valuable thing left is what never made it into the cauldron: your own notes, your history, your context, the needs you've noticed that nobody wrote down.
That changes where the action is. In a recent piece I put it this way: "Models still matter. But I expect the center of gravity to move toward what surrounds them: personal knowledge, useful applications, trusted relationships and meaningful control."
The catch is that your unrecorded edge stops being yours the moment it has to be uploaded to someone else's model before it can be searched. The centaur works best when the human half keeps its own memory, on hardware it controls, and decides what to share.
If your edge is the query, it matters a lot who keeps the record of every query you ask.
In my own work I see a lot of knowledge that exists just in a single person or small group of people. It's specialized, and not written down anywhere. My concern is that bredth of knowledge and experience (applied wisdom) gets lost, and that as AI takes over more and more jobs, humans don't pass that knowledge on, and never develop it again.
Tomas, great insight because it rings/resonates true! All the correlated data AI can come up with in a swoosh by pushing and popping ( am ex Bell Labs ) but the stuff unrecorded ( the extra data in your brain ) only you can have an Einstein moment with.
Most intelligence is compression,and decompression, so that you can hold more than one thing in your mind. LLMs consistently compress and decompress poorly, leading to errors.
Daiju visited the master Baso in China. Baso asked: "What do you seek?"
"Enlightenment," replied Daiju.
"You have your own treasure house. Why do you search outside?" Baso asked.
Daiju inquired: "Where is my treasure house?"
Baso answered: "What you are asking is your treasure house."
Daiju was enlightened! Ever after he urged his friends: "Open your own treasure house and use those treasures."
——-
** Everything Is Best **
When Banzan was walking through a market he overheard a conversation between a butcher and his customer.
"Give me the best piece of meat you have," said the customer.
"Everything in my shop is the best," replied the butcher. "You cannot find here any piece of meat that is not the best."
At these words Banzan became enlightened.
[ Commentary: everything is a best answer, i.e., search result; from them, can you figure out the search questions the Universe was answering? And from many of these, better questions in general? As in reverse engineering how proteins get made (drug discovery or causes of disease) ]
Different topic: It would be interesting to see how these historical developments of 'ICT' have influenced the way we organize our decision making at societal en especially political level. Is there a link between revolutions and wars and our ICT '(r)evolutions'?
Most of what my own setup does all day is search, really. A small model sorts thousands of posts and only a handful ever reach me. Reading this made me feel less strange about how little of it is writing.
An uplifting read on AI is well received, especially when it is human centered, rather than techno-hype sales pitches for mega-scalers looking to fatten the pre-public offering valuation for the stock market.
Well done, Thomas.
As a self identified generalist and reneisceance inspired student of life, I would of course agree with the sentiment of this article.
Arnaud Bertrand published an interesting article on the Chinese perspective of AI development and purpose regarding humanity, alignment and perspectives, today. Also, very good.
Cool, though I would say search is more like erudition, or memory recall, rather than intelligence. We all know someone who seems to know it all but is otherwise dull.
Intelligence is the part where you can compress the result of the search into less bytes, and link it to more simple concepts. We've all been impressed by someone who was able to explain complicated stuff by linking it to simple things - which is pretty much what explaining something is. The compression is pretty much what AI training does. Linking, it does not so well yet.
Come to think of it, the compression is still vastly less good than an average human - except for things like math, AI still needs lots of bytes to express something a human child could express in much less (like, how many bytes to code "a cat on the floor" in AI vs human brain?). Or wattage - our 30W brains still vastly outperform any 30W AI on almost any task of value.
Hey this POMA sponsor seems to be doing good work. I think we really need AIs that properly cite their source, so we can check their summary. Gemini was doing that decently initially, but less so now. Would guess acknowledging your sources would eventually make it easier to make you pay for them, so perhaps we can't expect shareholder-owned entities to push much there. But can't see how we can have creators keep creating unless something like that occurs, and fast. That should be the default for all chatbots, though unfortunately that doesn't seem to be POMA's business model.
The Go chart cuts against the polymath section. Decision quality rose after AlphaGo because players studied it. That's depth, not breadth. Depth on Demand, my latest essay, puts the human edge in direction rather than recombination.
I don't know, I think AI will be able to do taste really well eventually, and insofar as it can't, it will be linked to the facts here. Eg, unrecorded human trends are hard to anticipate
The more far-reaching the different pieces of information to combine, and the more interestingly they get combined the more unique and creative the product is. Polymaths have always made important breakthroughs in fields for this reason and their importance seems to only be increasing.
This is why it's important to listen to the non-experts. "Why aren't you modeling this?" is a valuable question.
I wrote a paper 40 years ago on an idea I came up with for a different translation of Exodus 6:3, bounced it off my professor, he said make sure you search all the indeces, you'll probably find some obscure journal printed that idea 100 years ago.
Well, 40 years later, finishing the paper, I did find an obscure journal that came close, 90 years ago.
And AI has been amazingly helpful in research and writing - some of them are quite good at ancient languages, including hieroglyphics and cuneiform.
There's going to be an explosion of history with AI!
Your "thrive in the unrecorded" point is the one I keep coming back to. If intelligence is mostly search, the most valuable thing left is what never made it into the cauldron: your own notes, your history, your context, the needs you've noticed that nobody wrote down.
That changes where the action is. In a recent piece I put it this way: "Models still matter. But I expect the center of gravity to move toward what surrounds them: personal knowledge, useful applications, trusted relationships and meaningful control."
The catch is that your unrecorded edge stops being yours the moment it has to be uploaded to someone else's model before it can be searched. The centaur works best when the human half keeps its own memory, on hardware it controls, and decides what to share.
If your edge is the query, it matters a lot who keeps the record of every query you ask.
In my own work I see a lot of knowledge that exists just in a single person or small group of people. It's specialized, and not written down anywhere. My concern is that bredth of knowledge and experience (applied wisdom) gets lost, and that as AI takes over more and more jobs, humans don't pass that knowledge on, and never develop it again.
If the knowledge is valuable, it does get passed!
As a techno-pessimist, this article gave me some degree of hope. Thanks.
Tomas, great insight because it rings/resonates true! All the correlated data AI can come up with in a swoosh by pushing and popping ( am ex Bell Labs ) but the stuff unrecorded ( the extra data in your brain ) only you can have an Einstein moment with.
Most intelligence is compression,and decompression, so that you can hold more than one thing in your mind. LLMs consistently compress and decompress poorly, leading to errors.
Two famous Zen stories tie in with this:
** Open Your Own Treasure House **
Daiju visited the master Baso in China. Baso asked: "What do you seek?"
"Enlightenment," replied Daiju.
"You have your own treasure house. Why do you search outside?" Baso asked.
Daiju inquired: "Where is my treasure house?"
Baso answered: "What you are asking is your treasure house."
Daiju was enlightened! Ever after he urged his friends: "Open your own treasure house and use those treasures."
——-
** Everything Is Best **
When Banzan was walking through a market he overheard a conversation between a butcher and his customer.
"Give me the best piece of meat you have," said the customer.
"Everything in my shop is the best," replied the butcher. "You cannot find here any piece of meat that is not the best."
At these words Banzan became enlightened.
[ Commentary: everything is a best answer, i.e., search result; from them, can you figure out the search questions the Universe was answering? And from many of these, better questions in general? As in reverse engineering how proteins get made (drug discovery or causes of disease) ]
Different topic: It would be interesting to see how these historical developments of 'ICT' have influenced the way we organize our decision making at societal en especially political level. Is there a link between revolutions and wars and our ICT '(r)evolutions'?
https://unchartedterritories.tomaspueyo.com/p/pax-mercatus
https://unchartedterritories.tomaspueyo.com/p/internet-blockchain-kill-nation-states
:-) I could have known you had covered this topic already. Thanks for sharing. I'll read your posts and come back if I have any questions or remarks.
Most of what my own setup does all day is search, really. A small model sorts thousands of posts and only a handful ever reach me. Reading this made me feel less strange about how little of it is writing.
Doesn’t a good query take depth, though? You can only ask a new question if you know which ones have already been asked.
True
An uplifting read on AI is well received, especially when it is human centered, rather than techno-hype sales pitches for mega-scalers looking to fatten the pre-public offering valuation for the stock market.
Well done, Thomas.
As a self identified generalist and reneisceance inspired student of life, I would of course agree with the sentiment of this article.
Arnaud Bertrand published an interesting article on the Chinese perspective of AI development and purpose regarding humanity, alignment and perspectives, today. Also, very good.
Cool, though I would say search is more like erudition, or memory recall, rather than intelligence. We all know someone who seems to know it all but is otherwise dull.
Intelligence is the part where you can compress the result of the search into less bytes, and link it to more simple concepts. We've all been impressed by someone who was able to explain complicated stuff by linking it to simple things - which is pretty much what explaining something is. The compression is pretty much what AI training does. Linking, it does not so well yet.
Come to think of it, the compression is still vastly less good than an average human - except for things like math, AI still needs lots of bytes to express something a human child could express in much less (like, how many bytes to code "a cat on the floor" in AI vs human brain?). Or wattage - our 30W brains still vastly outperform any 30W AI on almost any task of value.
Hey this POMA sponsor seems to be doing good work. I think we really need AIs that properly cite their source, so we can check their summary. Gemini was doing that decently initially, but less so now. Would guess acknowledging your sources would eventually make it easier to make you pay for them, so perhaps we can't expect shareholder-owned entities to push much there. But can't see how we can have creators keep creating unless something like that occurs, and fast. That should be the default for all chatbots, though unfortunately that doesn't seem to be POMA's business model.
The Go chart cuts against the polymath section. Decision quality rose after AlphaGo because players studied it. That's depth, not breadth. Depth on Demand, my latest essay, puts the human edge in direction rather than recombination.
Do you think taste works the same way, or is that the part search can't really cover?
I don't know, I think AI will be able to do taste really well eventually, and insofar as it can't, it will be linked to the facts here. Eg, unrecorded human trends are hard to anticipate
Taste linked to facts makes sense. Trends nobody wrote down are hard to predict for anyone, AI included.