AI pioneer Yoshua Bengio says governments must move fast to ‘protect the public’(ft.com)
ft.com
AI pioneer Yoshua Bengio says governments must move fast to ‘protect the public’
https://www.ft.com/content/b4baa678-b389-4acf-9438-24ccbcd4f201
10 comments
This isn't really something that can be legislated. The cat is out of the bag, it's not being put back in. And closing the systems that are massive and popular will simply encourage replacements that have less restrictions. We can't even manage to legislate away the demon everyone agrees is a demon (child exploitation).
Further, we are not talking about an easy to identify item like child pornography, we are talking about systems that are mostly invisible, even to professionals.
To me, these are unsolvable issues. It also seems obvious enough that I question the motives of every person involved in the legislation push.
Before you downvote, please present a potential solution for discussion.
Further, we are not talking about an easy to identify item like child pornography, we are talking about systems that are mostly invisible, even to professionals.
To me, these are unsolvable issues. It also seems obvious enough that I question the motives of every person involved in the legislation push.
Before you downvote, please present a potential solution for discussion.
Crypto software has also been legislated in the past and everyone ended up using weak crypto for quite a long time. Looking back, maybe it wasn't that important anyway since most of the web still ran without TLS and you wouldn't read on a weekly basis about password breaches.
If "GPT 4 or higher" would be restricted then a lot of large businesses wouldn't use it, startups/smaller companies wouldn't work on services around that. The common knowledge about it wouldn't be as high as unrestricted.
It's not optimal but maybe good enough for the next few years. Also bringing back the comparison with crypto, maybe in the mean-time everyone involved is better equipped in dealing with all that.
If "GPT 4 or higher" would be restricted then a lot of large businesses wouldn't use it, startups/smaller companies wouldn't work on services around that. The common knowledge about it wouldn't be as high as unrestricted.
It's not optimal but maybe good enough for the next few years. Also bringing back the comparison with crypto, maybe in the mean-time everyone involved is better equipped in dealing with all that.
I believe the goal is to only restrict systems that have capabilities that do not yet exist (GPT- 4.5, or so).
GPT-4, ChatGPT, or Wizard-Vicuna, etc are not really the concern to many experts regarding the type of safety legislation that is more seriously being considered.
I don't downvote, but I dislike your attitude about problems not having solutions.
You're saying "nothing" can be done. Should we same the same about everything? Climate change, nuclear proliferation etc?
You're saying "nothing" can be done. Should we same the same about everything? Climate change, nuclear proliferation etc?
For what it's worth, I hate being the problem with no solutions person. I don't think it's particularly useful and when I can think of potential solutions I will push for them at my own expense. Though honestly, those expenses are typically short lived and turn into benefits in the long term.
This problem however seems different from other problems. I don't see a solution and fear the target of legislation is inadequate and misplaced. Though I don't know where it should go either.
Edit: I believe what will happen is that legislation will focus on empowering powerful people and disempowering poor people. Which based on the article will be stated to be for the benefit of marginalized groups while only benefiting members of said groups when they are already in positions of power. As is normally the case.
And what about Russia? What can we do to stop them from using AI maliciously? Not to suggest they will, it's just the commonly stated villain.
Even a solution of first arrival doesn't inspire confidence. As any sufficient system is likely to escape and render us as we have rendered cows.
Edit: I propose many solutions for climate change that are more severe than most people will accept. Things like incredibly expensive garbage collection (because I can get by with zero garbage and so can everyone else) and taxes on new item purchases(scaling exponentially for wants instead of needs) with incentives for products to become more focused on longevity. And many others.
This problem however seems different from other problems. I don't see a solution and fear the target of legislation is inadequate and misplaced. Though I don't know where it should go either.
Edit: I believe what will happen is that legislation will focus on empowering powerful people and disempowering poor people. Which based on the article will be stated to be for the benefit of marginalized groups while only benefiting members of said groups when they are already in positions of power. As is normally the case.
And what about Russia? What can we do to stop them from using AI maliciously? Not to suggest they will, it's just the commonly stated villain.
Even a solution of first arrival doesn't inspire confidence. As any sufficient system is likely to escape and render us as we have rendered cows.
Edit: I propose many solutions for climate change that are more severe than most people will accept. Things like incredibly expensive garbage collection (because I can get by with zero garbage and so can everyone else) and taxes on new item purchases(scaling exponentially for wants instead of needs) with incentives for products to become more focused on longevity. And many others.
I don't think there's a solution AFA regulating AI directly. I do think there are solutions to existing problems that need to finally be implemented.
We need better individual identity/security/privacy technologies, frameworks and rights to not have some dippy broker use your voice as your fingerprint. That's necessary regulation and infrastructure made more necessary by AI not direct regulation of AI. Weakness in the international phone systems and the ability to send arbitrary legal offers via email without any trust anchor are similar too.
We need better individual identity/security/privacy technologies, frameworks and rights to not have some dippy broker use your voice as your fingerprint. That's necessary regulation and infrastructure made more necessary by AI not direct regulation of AI. Weakness in the international phone systems and the ability to send arbitrary legal offers via email without any trust anchor are similar too.
> The near-term dangers of the technology include misinformation, insidious bias and discrimination against marginalised groups and minorities
All of the same could be said of the internet itself when AOL started its outreach. By god, anybody can publish anything-- and just like that, it becomes truth!
They said the same about the printing press, and probably that heretic Martin Luther too.
The FUD is reaching "eternal damnation" levels of histrionics. You keep saying we're in danger of this apocalyptic event, but nobody seems to know what the hell you're talking about. What exactly do you know about this tech that we don't?
More and more it looks like minority groups are trying to wrest control of the technology away from the masses, citing vague dangers like people misleading each other, such that we once again have to depend on a handful of ordained priests to interpret the Word of God and deliver us from evil. We trust priests, right?
Human interests have flooded the streets with guns such that children are murdering each other in schools. Human laws force women to carry pregnancies to term even if it kills them. Human-engineered addiction goes unchecked. But we're told a fucking NPC they created is an existential threat to humanity-- and that their historically-reckless corporate creators should be entrusted with exclusive governance over it.
From what I've seen, current instruct-based models are frighteningly adept at fraud and manipulation, but we downplay both as "hallucination." Nobody citing safety is talking about the danger of trusting a clearly-mentally-ill entity as a source of truth though, because that would lead to fear and distrust of the product...
All of the same could be said of the internet itself when AOL started its outreach. By god, anybody can publish anything-- and just like that, it becomes truth!
They said the same about the printing press, and probably that heretic Martin Luther too.
The FUD is reaching "eternal damnation" levels of histrionics. You keep saying we're in danger of this apocalyptic event, but nobody seems to know what the hell you're talking about. What exactly do you know about this tech that we don't?
More and more it looks like minority groups are trying to wrest control of the technology away from the masses, citing vague dangers like people misleading each other, such that we once again have to depend on a handful of ordained priests to interpret the Word of God and deliver us from evil. We trust priests, right?
Human interests have flooded the streets with guns such that children are murdering each other in schools. Human laws force women to carry pregnancies to term even if it kills them. Human-engineered addiction goes unchecked. But we're told a fucking NPC they created is an existential threat to humanity-- and that their historically-reckless corporate creators should be entrusted with exclusive governance over it.
From what I've seen, current instruct-based models are frighteningly adept at fraud and manipulation, but we downplay both as "hallucination." Nobody citing safety is talking about the danger of trusting a clearly-mentally-ill entity as a source of truth though, because that would lead to fear and distrust of the product...
Destabilize democracy? It’s already been destabilized
Just wait a little more, sweet regulatory prince.