https://www.c-span.org/video/?528117-1/openai-ceo-testifies-...
- Mazie Hirono, Junior Senator from Hawaii, has very thoughtful questions. Very impressive.
- Gary Marcus also up there speaking with Sam Altman of OpenAI.
- So far, Sen. Hirono and Sen. Padilla seem very wary of regulating AI at this time.
- Very concerned about not "replicating social media's failure", why is it so biased and inequitable. Much more reasonable concerns.
- Also responding to questions is Christina Montgomery, chair of IBM's AI Ethics Board.
- "Work to generate a representative set of values from around the world."
- Sen. Ossoff asking for definition of "scope".
- "We could draw a line at systems that need to be licensed. Above this amount of compute... Define some capability threshold... Models that are less capable, we don't want to stop open source."
- Ossoff wants specifics.
- "Persuade, manipulate, influence person's beliefs." should be licensed.
- Ossoff asks about predicting human behavior, i.e. use in law enforcement, "It's very important we understand these are tools, not to take away human judgment."
- "We have no national privacy law." — Sen Ossof "Do you think we need one?"
- Sam "Yes. User should be able to opt out of companies using data. Easy to delete data. If you don't want your data use to train, you have right to exclude it."
- "There should be more ways to have your data taken down off the public web." —Sam
- "Limits on what a deployed model is capable of and also limits on what it will answer." — Sam
- "Companies who depend upon usage time, maximize engagement with perverse results. I would humbly advise you to get way ahead of this, the safety of children. We will look very harshly on technology that harms children."
- "We're not an advertising based model." —Sam
- "Requirements about how the values of these systems are set and how they respond to questions." —Sam
- Sen. Booker up now.
- "For congress to do nothing, which no one is calling for here, would be exceptional."
- "What kind of regulation?"
- "We don't want to slow things down."
- "A nimble agency. You can imagine a need for that, right?"
- "Yes." —Christina Montgomery
- "No way to put this genie back in the bottle." Sen. Booker
- "There are more genies yet to come from more bottles." — Gary Marcus
- "We need new tools, new science, transparency." —Gary Marcus
- "We did know that we wanted to build this with humanity's best interest at heart. We could really deeply transform the world." —Sam
- "Are you ever going to do ads?" —Sen Booker
- "I wouldn't say never...." —Sam
- "Massive corporate concentration is really terrifying.... I see OpenAI backed by Microsoft, Anthropic is backed by Google. I'm really worried about that. Are you worried?" —Sen Booker?
- "There is a real risk of technocracy combined with oligarchy." —Gary Marcus
- "Creating alignment dataset has got to come very broadly from society." —Sam Senator Welch from Vermont up now
- "I've come to the conclusion it's impossible for congress to keep up with the speed of technology."
- "The spread of disinformation is the biggest threat."
- "We absolutely have to have an agency. Scope has to be defined by congress. Unless we have an agency, we really don't have much of a defense against the bad stuff, and the bad stuff will come."
- Use of regulatory authority and the recognition that it can be used for good, but there's also legitimate concern of regulation being a negative influence."
- "What are some of the perils of an agency?"
- "America has got to continue to lead."
- "I believe it's possible to do both, have a global view. We want America to lead."
- "We still need open source to comply, you can still do harm with a smaller model."
- "Regulatory capture. Greenwashing." —Gary Marcus
- "Risk of not holding companies accountable for the harms they are causing today." —Christina Montgomery
- Lindsay Graham, very pro-licensing, "You don't build a nuclear power plant without a license, you don't build an AI without a license."
- Sen Blumenthal brings up Anti-Trust legislation.
- Blumenthal mentions how classified briefings already include AI threats.
- "For every successful regulation, you can think of five failures. I hope our experience here will be different."
- "We need to grapple with the hard questions here. This has brought them up, but not answered them."
- "Section 230"
- "How soon do you think gen AI will be self-aware?" —Sen Blumenthal
- "We don't understand what self-awareness is." —Gary Marcus
- "Could be 2 years, could be 20."
- "What are the highest risk areas? Ban? Strict rules?"
- "The space around misinformation. Knowing what content was generated by AI." —Christina Montgomery
- "Medical misinformation, hallucination. Psychiatric advice. Ersatz therapists. Internet access for tools, okay for search. Can they make orders? Can they order chemicals? Long-term risks." —Gary Marcus
- "Generative AI can manipulate the manipulators." —Blumenthal
- "Transparency. Accountability. Limits on use. Good starting point?" —Blumenthal
- "Industry should't wait for congress." —C. Montgomery
- "We don't have transparency yet. We're not doing enough to enforce it." —G. Marcus
- "AGI closer than a lot of people appreciate." —Blumenthall
- Gary and Sam are getting along and like each other now.
- Josh Hawley
- Talking about loss of jobs, invasion of personal privacy, manipulation of behavior, opinion, and degradation of free elections in America.
- "Are they right to ask for a pause?"
- "It did not call for a ban on all AI research or all AI, only on very specific thing, like GPT-5." -G Marcus
- "Moratorium we should focus on is deployment. Focus on safety." —G. Marcus
- "Without external review."
- "We waited more than 6 months to deploy GPT-4. I think the frame of the letter is wrong." —Sam
- Seems to not like the arbitrariness of "six months."
- "I'm not sure how practical it is to pause." —C. Montgomery
- Hawley brings up regulatory capture, usually get controlled by people they're supposed to be watching. "Why don't we just let people sue you?"
- If you were harmed by AI, why not just sue?
- "You're not protected by section 230."
- "Are clearer laws a good thing? Definitely, yes." —Sam
- "Would certainly make a lot of lawyers wealthy." —G. Marcus
- "You think it'd be slower than congress?" —Hawley
- Copyright, wholesale misinformation laws, market manipulation?" Which laws apply? System not thought through? Maybe 230 does apply? We don't know.
- "We can fix that." —Hawley
- "AI is not a shield." —C. Montgomery
- "Whether they use a tool or a human, they're responsible." —C. Montgomery
- "Safeguards and protections, yes. A flat stop sign? I would be very, very worried about." —Blumenthall
- "There will be no pause." Sen. Booker "Nobody's pausing."
- "I would agree." Gary Marcus
- "I have a lot of concerns about corporate intention." Sen Booker
- "What happens when these companies that already control so much of our lives when they are dominating this technology?" Booker
- Sydney really freaked out Gary. He was more freaked out when MS didn't withdraw Sydney like it did Tay.
- "I need to work on policy. This is frightening." G Marcus
- Cory admits he is a tech bro (lists relationships with investors, etc)
- "The free market is not what it should be." —C. Booker
- "That's why we started OpenAI." —Sam "We think putting this in the hands of a lot of people rather than the hands of one company." —Sam
- "This is a new platform. In terms of using the models, people building are doing incredible things. I can't believe you get this much technology for so little money." —Sam
- "Most industries resist reasonable regulation. The only way we're going to see democratization of values is if we enforce safety measures." —Cory Booker
- "I sense a willingness to participate that is genuine and authentic." —Blumenthal