Regulating Artificial Intelligence: A Conversation with Yoshua Bengio and Scott Wiener(Episode #379)

Sam Harris speaks with Yoshua Bengio and Scott Wiener about AI risk and the new bill introduced in California intended to mitigate it. They discuss the controversy over regulating AI and the assumptions that lead people to discount the danger of an AI arms race.

Yoshua Bengio is full professor at Université de Montréal and the Founder and Scientific Director of Mila - Quebec AI Institute. Considered one of the world’s leaders in artificial intelligence and deep learning, he is the recipient of the 2018 A.M. Turing Award with Geoffrey Hinton and Yann LeCun, known as the Nobel Prize of computing.

He is a Canada CIFAR AI Chair, a member of the UN’s Scientific Advisory Board for Independent Advice on Breakthroughs in Science and Technology, and Chair of the International Scientific Report on the Safety of Advanced AI.

Scott Wiener has represented San Francisco in the California Senate since 2016. He recently introduced SB 1047, a bill aiming to reduce the risks of frontier models of AI. He has also authored landmark laws to, among other things, streamline the permitting of new homes, require insurance plans to cover mental health care, guarantee net neutrality, eliminate mandatory minimums in sentencing, require billion-dollar corporations to disclose their climate emissions, and declare California a sanctuary state for LGBTQ youth. He has lived in San Francisco's historically LGBTQ Castro neighborhood since 1997.

Twitter: @Scott_Wiener

August 12, 2024

[briancase]

Thanks for this sober and adult discussion. It's important, and the timing is good.

[progrob]

Happy to hear more content with Joseph :)

[jaarneal]

Woah, you got Bengio himself! Can't wait to listen to this.

[phillipcarr]

Who will regulate Ai research in North Korea, Iran and Russia ?

[jynxkizs]

Regulations can still have extra paperwork, costs, and other hassels, even if the risk isn't real.

[GenericAF]

Hey Sam! To be clear I’m not an ai expert just just a fan of intelligence. I think we have to separate intelligence from ai until it’s actually intelligent. Answering a question right isn’t the only qualifier for intelligence. Know why it’s being asked. Know who’s asking it and the purpose. Those are in essence just as important to intelligence and more so to safety. We consider intelligence as knowing. Knowing the right answer is one aspect out of a hundred or a thousand. That’s why as of now you can’t assume ai is intelligent.
Regulations I’m not a fan of especially by politicians. Instead of regulations there should be tests. Test, not about the right answers but all the things encompassing the answers.

[Zleec]

It is amazing that he doesn't realize that as soon as he says "AI can still be used (by a human) to do bad stuff" he is talking about something completely different than AI having an inherit risk in itself.

[mbarker_lng]

Currently, the job market is flooded with out of work programmers and it is due to AI but not in the way everyone thinks. It is not that our jobs are being done by some MCP-like AI (+1 if you get that ref), its that the big tech firms are pivoting their projects to AI; either R+D or integrating AI into their shiny new project. Their mode of operation is to shutter current projects, lay everyone off, then hire Ai-centric programmers. So now, you apply to a job, AI or not, and find out it has 100+ applicants before a single day has passed. Good luck having anyone even see your resume. :(

[Bronco]

Sam, id love to see you have conversations with Claud or one of the major AIs and test it and give your honest thoughts on some of the difficult questions about consciousness and such.

[MarcioSouza]

Great timing, I needed new Sam Harris content to fall asleep to (not a joke, try it!)

[anuzis]

For folks who aren't familiar with Sam's background reasoning on this topic I highly recommend checking out his TED talk from 7 years ago exploring whether we should expect we can build artificial intelligence without losing control of it. Much of the reasoning applied then is still reasonable to apply today.

[dmanson]

“Let’s use the Govt to keep us safe and free.”

[paddydiddles]

It’s a law of nature - anything that has great potential (for good) can also have great potential for harm

[itsallgoodman]

Sam will you address the pernicious audience capture of the Joe Rogan podcast

[retromograph]

It doesn't take $100 million to train a human with genius level intelligence, so why, with the inevitable advances in the tech, would it take that for an AGI? How many pictures of an elephant does a toddler need to see to know what an elephant is? Maybe 10? Certainly not 100, 000.

[abartonicek]

I'm more and more convinced that all this AI safety talk is just a marketing grift designed to oversell VC's on the capabilities of LLMs. "These things are super dangerous, moving so fast, human-level intelligence is just behind the corner - quick, invest in us so you don't get left behind."

No-one ever discusses the big question of how you actually get true general intelligence out of a probabilistic model of language.

[Ellerich]

What if in a few years the performance threshold in the legislation is easily available to all AI programmers even for the developer working in his basement in Indonesia? Wouldn't it then impact small and large companies?

[palmpat]

I don't think anyone, regardless of expertise, truly believes that AI doesn't pose unfathomable problems on a wide variety of levels. As long as there are economic incentives, there will be those that benefit and those who will suffer, either by the psychological pollution it will cause, actually malicious and/or greedy applications, or virtually unlimited kinds of mistakes possible in its development.
Then there are those who have the resources to believe they can escape the negative consequences, and perhaps they will be able to.

[visceral]

Scott Wiener needs to be regulated for the good of humanity.

[ZippyLeroux]

When did Sam come back to youtube? oh well, back catalogue FTW!