I do think, though, that too much is expected in this post from 'generative AI'. Generative AI has serious trouble with producing *meaningful* results. The results are well-structured and they fit the subject, but the results are not trustworthy, nor is there any sign that they will be. Reading OpenAI's article on GPT Few-shot learning (https://arxiv.org/pdf/2005.14165.pdf) for instance, it is clear that — other than producing language that is well-structured and 'fitting' — the results on being *trustworthy* in terms of content/meaning remain very poor. Generative AI seems magical, not because the systems are intelligent, but because we humans not that much. We are easily fooled/misled.
As someone who regularly relies on data, this rings so true to me: "...it’s more like only Google is positioned to know. Only Google has the capacity to know. But given the magnitude of their traffic, the company only notices those things that staff have been assigned staff to look for. "
Thank you for making that observation. It's not enough to have the data - someone has to actually be looking at and analyzing it. It always comes back to people's time and what they are focused on.
I definitely agree with the concern, but it's also exactly this space where there's room for legislative and regulatory intervention, to forestall the creation and perpetuation of (more) invincible black boxes. E.g., requiring companies to disclose, at least, the source of training data and incorporated data for any generative AI or algorithm - it would at least allow a way to think about backwards-engineering them, as well as a potential avenue for redress of IP or copyright claims from artists (a royalty structure like with radio play or songwriting credits seems at least plausible in this context).
Yep, agreed. And I wouldn't be surprised if copyright law/copyright-holding interests ends up playing a central role in narrowing the scope of this stuff.
There need to be a few cases to remind companies that AI may be acting as a agent for the company and that leaves the company responsible. It isn't about looking inside the black box and trying to decide if the AI model is doing things by explicit design. The issue is whether it is making biased decisions, providing false information, leaking private information or the like. It's too easy for companies to say, well, that's what the data says and evade responsibility.
Just wondering, is this sentence missing a word and meant to say “less transparent”?
“ AI could be the engine driving us towards a future where the main hubs of information and communication become transparent, less responsive, less manageable, and more socially harmful.”
Good stuff, as usual.
I do think, though, that too much is expected in this post from 'generative AI'. Generative AI has serious trouble with producing *meaningful* results. The results are well-structured and they fit the subject, but the results are not trustworthy, nor is there any sign that they will be. Reading OpenAI's article on GPT Few-shot learning (https://arxiv.org/pdf/2005.14165.pdf) for instance, it is clear that — other than producing language that is well-structured and 'fitting' — the results on being *trustworthy* in terms of content/meaning remain very poor. Generative AI seems magical, not because the systems are intelligent, but because we humans not that much. We are easily fooled/misled.
See https://ea.rna.nl/2022/12/12/cicero-and-chatgpt-signs-of-ai-progress/
As someone who regularly relies on data, this rings so true to me: "...it’s more like only Google is positioned to know. Only Google has the capacity to know. But given the magnitude of their traffic, the company only notices those things that staff have been assigned staff to look for. "
Thank you for making that observation. It's not enough to have the data - someone has to actually be looking at and analyzing it. It always comes back to people's time and what they are focused on.
I definitely agree with the concern, but it's also exactly this space where there's room for legislative and regulatory intervention, to forestall the creation and perpetuation of (more) invincible black boxes. E.g., requiring companies to disclose, at least, the source of training data and incorporated data for any generative AI or algorithm - it would at least allow a way to think about backwards-engineering them, as well as a potential avenue for redress of IP or copyright claims from artists (a royalty structure like with radio play or songwriting credits seems at least plausible in this context).
Yep, agreed. And I wouldn't be surprised if copyright law/copyright-holding interests ends up playing a central role in narrowing the scope of this stuff.
There need to be a few cases to remind companies that AI may be acting as a agent for the company and that leaves the company responsible. It isn't about looking inside the black box and trying to decide if the AI model is doing things by explicit design. The issue is whether it is making biased decisions, providing false information, leaking private information or the like. It's too easy for companies to say, well, that's what the data says and evade responsibility.
Just wondering, is this sentence missing a word and meant to say “less transparent”?
“ AI could be the engine driving us towards a future where the main hubs of information and communication become transparent, less responsive, less manageable, and more socially harmful.”
?
it is indeed, will fix that now. Thanks.