CPSC grads are all aware of the story of the Mechanical Turk. I got mine just before the "AI Winter" when all the 70s-80s projects collapsed into being of minor commercial value. That's always the test: Money is the measure of all things, in business...
I went to the Master's party for my own T.A., hers being in AI of the time (1983). She said flatly that there was no product on Earth at the time capable of *understanding* a sentence as well as a three-year-old child - that being the point where kids start using whole sentences that connect together. Even then, comprehensible sentences could be generated, but the human was doing all the *understanding*. I finally plodded (and skimmed) through Stephen Wolfram's accessible but deeply detailed explanation of the LLM:
...and it still isn't *understanding*. Wolfram himself sells his "natural language understanding" system, which really does get into that problem. But even his, is maybe getting past age 3, possibly 4, with the connection of words to "meaning".
The current products are mostly creating imitations of saleable material - and "imitations" are not something there's a hot market for, in clothing, music, or words. What LLMs can really pump out, are "corporate bafflegab" with a minimum of meaning, and those jobs will die unlamented. There's too much of such "content" now.
Yup. I think that sums up my thinking pretty accurately. I think there are avenues where this computing power is going to make a huge difference — maths, physics, medicine, etc — but I imagine there's going to be a backlash to the low-quality, high-volume content that will soon bury us.
Wonderful piece, Justin, thank you. I just wonder, when current AI "creations" are done mining the "understanding" of previous generations, will they eventually encounter nothing but their own recreations? Corruption of civilization in the second degree?
The snake will eventually eat its own tail, absolutely. I think we've already seen a few LLMs that have declined in quality the more info is being added into them. But I think that will serve to discredit them as a good stand-in for humans. Sure there will be chatbots and automated messages, but I genuinely don't see us spending a significant quantity of our days talking to LLMs. Maybe I'll be proven totally wrong, here.
I saw an Instagram post about Dudesy's George Carlin "AI" and I fell for it completely. I was punk'd! So were many others who, like me, commented in the thread that it was an insult to Carlin's memory. Well-played, Dudesy's creators. I don't really get the motivation, but whatever turns your crank I guess (pun intended).
An interesting and thought provoking piece! Most of us probably haven’t paid much attention to the pitfalls of AI or for that matter, the ways that AI could be used for good. Thank goodness for journalists like you who can help us to learn more about it and to pay closer attention to this important issue!
Sad!
CPSC grads are all aware of the story of the Mechanical Turk. I got mine just before the "AI Winter" when all the 70s-80s projects collapsed into being of minor commercial value. That's always the test: Money is the measure of all things, in business...
I went to the Master's party for my own T.A., hers being in AI of the time (1983). She said flatly that there was no product on Earth at the time capable of *understanding* a sentence as well as a three-year-old child - that being the point where kids start using whole sentences that connect together. Even then, comprehensible sentences could be generated, but the human was doing all the *understanding*. I finally plodded (and skimmed) through Stephen Wolfram's accessible but deeply detailed explanation of the LLM:
https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-and-why-does-it-work/
...and it still isn't *understanding*. Wolfram himself sells his "natural language understanding" system, which really does get into that problem. But even his, is maybe getting past age 3, possibly 4, with the connection of words to "meaning".
The current products are mostly creating imitations of saleable material - and "imitations" are not something there's a hot market for, in clothing, music, or words. What LLMs can really pump out, are "corporate bafflegab" with a minimum of meaning, and those jobs will die unlamented. There's too much of such "content" now.
Yup. I think that sums up my thinking pretty accurately. I think there are avenues where this computing power is going to make a huge difference — maths, physics, medicine, etc — but I imagine there's going to be a backlash to the low-quality, high-volume content that will soon bury us.
Wonderful piece, Justin, thank you. I just wonder, when current AI "creations" are done mining the "understanding" of previous generations, will they eventually encounter nothing but their own recreations? Corruption of civilization in the second degree?
The snake will eventually eat its own tail, absolutely. I think we've already seen a few LLMs that have declined in quality the more info is being added into them. But I think that will serve to discredit them as a good stand-in for humans. Sure there will be chatbots and automated messages, but I genuinely don't see us spending a significant quantity of our days talking to LLMs. Maybe I'll be proven totally wrong, here.
I saw an Instagram post about Dudesy's George Carlin "AI" and I fell for it completely. I was punk'd! So were many others who, like me, commented in the thread that it was an insult to Carlin's memory. Well-played, Dudesy's creators. I don't really get the motivation, but whatever turns your crank I guess (pun intended).
I fell for it too, initially! I probably should've mentioned that detail.
An interesting and thought provoking piece! Most of us probably haven’t paid much attention to the pitfalls of AI or for that matter, the ways that AI could be used for good. Thank goodness for journalists like you who can help us to learn more about it and to pay closer attention to this important issue!