Fascinating reading; thank you for writing these newsletters!
My impression is that there are two distinct definitions of the term "AI". On one hand, "AI" refers to earlier-generation algorithmic and machine learning systems branded "AI" largely for marketing purposes. Such systems are relatively mature, can significantly impact our lives, and clearly require thoughtful regulation. Even earlier drafts of the AI act provided that!
On the other hand, the term "AI" now also refers to a quite different set of powerful and rapidly-evolving technologies, including foundation models, derivatives, and probably incarnations that we can not yet imagine. Unlike the technology previously branded "AI", this new crop looks like genuine machine intelligence or its immediate predecessor. While also perhaps worthy of regulation, this technology is likely to continue undergoing radical evolution. As a result, attempting to write an enduring law to cover it seems doomed to fail; specifically, what "it" is could change abruptly, making a regulation obsolete or nonsensical. As an analogy, imagine trying to write the definitive computer regulation in the late 1980s only to see the world wide web appear!
So regulating older technologies called "AI" is actually quite important to society, and drafts of the Act do that. Regulating newer technologies is ALSO entirely worthy of consideration, but more challenging. The critical point is that the need for the latter does not diminish the need for the former. Yes, GPAIS (as you call them) are indeed flourishing, but they are not replacing many of the earlier-generation decisions systems branded "AI", which continue to affect our lives.
Given this, I think the EU would do well to promptly pass regulation covering systems other than GPAIS. They were on a great track and should wrap it up. Separately and in parallel, the EU should take up regulation of GPAIS with no illusion that that will be a one-time effort.
In short, I'm afraid that the aesthetic desire to produce a single, unified "AI act for the ages" could leave us with unnecessary delays or no act at all. I find this very frustrating!
Can't wait until AI takes half of all jobs so I can collect my UBI check and focus on my art!
Fascinating reading; thank you for writing these newsletters!
My impression is that there are two distinct definitions of the term "AI". On one hand, "AI" refers to earlier-generation algorithmic and machine learning systems branded "AI" largely for marketing purposes. Such systems are relatively mature, can significantly impact our lives, and clearly require thoughtful regulation. Even earlier drafts of the AI act provided that!
On the other hand, the term "AI" now also refers to a quite different set of powerful and rapidly-evolving technologies, including foundation models, derivatives, and probably incarnations that we can not yet imagine. Unlike the technology previously branded "AI", this new crop looks like genuine machine intelligence or its immediate predecessor. While also perhaps worthy of regulation, this technology is likely to continue undergoing radical evolution. As a result, attempting to write an enduring law to cover it seems doomed to fail; specifically, what "it" is could change abruptly, making a regulation obsolete or nonsensical. As an analogy, imagine trying to write the definitive computer regulation in the late 1980s only to see the world wide web appear!
So regulating older technologies called "AI" is actually quite important to society, and drafts of the Act do that. Regulating newer technologies is ALSO entirely worthy of consideration, but more challenging. The critical point is that the need for the latter does not diminish the need for the former. Yes, GPAIS (as you call them) are indeed flourishing, but they are not replacing many of the earlier-generation decisions systems branded "AI", which continue to affect our lives.
Given this, I think the EU would do well to promptly pass regulation covering systems other than GPAIS. They were on a great track and should wrap it up. Separately and in parallel, the EU should take up regulation of GPAIS with no illusion that that will be a one-time effort.
In short, I'm afraid that the aesthetic desire to produce a single, unified "AI act for the ages" could leave us with unnecessary delays or no act at all. I find this very frustrating!