Representatives from tech giants including Meta (META), Amazon (AMZN), Microsoft (MSFT), and Alphabet (GOOGL) met at the White House to discuss regulating artificial intelligence. The companies agreed to voluntary safeguards to better manage the risks posed by AI.
After the meeting concluded, Meta’s President of Global Affairs Nick Clegg told Yahoo Finance Live’s Seana Smith and Allie Garfinkle that he believes all the tech companies at the meeting want to move AI technology forward quickly, but responsibly, adding that Meta is “absolutely determined to follow-up and honor these commitments.”
Clegg doesn’t share the fears that some have of AI, saying that some of predictions “have slightly run ahead of the technology.” Clegg does say concerns about things like deep fakes and the manipulation of audio and video are “legitimate,” but that “the technology which generates that unwelcome material is also the technology we will use to actually identify and control it.”
On regulation, Clegg said it is “definitely” needed, but that companies are not “going to sit around an wait for legislation,” noting that’s part of the reason why these voluntary commitments are being made.
One of the big concerns about AI is how it can be used to create misinformation or disinformation that could impact an election. Clegg argues that AI can help battle misinformation and disinformation too, “whether it’s created by a human being or by a machine.”
00:00:37 What the White House conversations were about
00:02:17 Clegg discusses AI fears
00:04:20 Meta’s stance on regulating AI
00:05:34 Clegg on Meta’s open-source AI large language model
00:07:35 Discussion on AI and misinformation ahead of the 2024 election
Video Transcript
– Well, the Biden administration announcing it has reached a deal with seven tech giants to safeguard artificial intelligence. The White House saying that the move will ensure that AI products are safe before introducing them to the public. And the companies say that they will earn the public’s trust.
Our next guest just wrapped up a meeting with President Biden. Joining me now, Meta President of Global Affairs, Nick Clegg, along with our own Allie Garfinkle.
Nick, it’s great to have you here. And I know you’re very busy. What can you tell us about the discussions with President Biden? What exactly was discussed within the meeting and how you plan to enforce some of the guidelines that you laid out?
NICK CLEGG: So the conversation with President Biden and his team was really around the following. AI is developed. It’s not new. AI has been around for years. But it’s now evolving very fast with these so-called new large language models, like ChatGPT, where you can ask it to write your school history essay or your wedding speech. And it comes out with perfect text. It’s moving very fast.
And like with every kind of fast moving technology, which is technology is bigger than any single company or indeed any single country, there’s always legitimate questions about how to strike the right balance between moving fast with these innovations but doing so responsibly. And I think all of the companies, including Meta, we kind of agree with that. We want to strike the right balance about moving forward.
These US companies have a real role of world leadership in this technology. But to do so in a way that people feel is transparent so people can see for themselves how these systems work. To make sure that they’re properly stress-tested so you have so-called red teaming. To make sure that wherever possible, it’s not always going to be possible, you can detect and inform people where content has been generated by AI systems rather than by human beings. To publish research, to work collaboratively across the sector and so on.
And that was the topic of the conversation. That’s the subject of those voluntary commitments that all the companies have made. And I think, certainly I can speak for Meta, we are absolutely determined to follow up and honor these commitments because it’s, beyond anything else, it’s in our own interests to do so. We want to make sure that any technology that is deployed that we come up with is deployed as responsibly and safely as possible
– Nick, there has certainly been many dramatic predictions when it comes to AI, right. You’re talking– you talk about how it’s all about striking that balance. But there has been predictions everywhere, from massive job loss to misinformation, disinformation, and then all the way to some saying that it could actually trigger the end of the world.
I’m curious to get your take outside of those dramatic predictions. What is your number one risk or the number one threat that you’re worried about when it comes to AI?
NICK CLEGG: So I think quite a lot of the sort of existential, you know, it’s going to end humanity and turn us all into paperclips by next Tuesday predictions, I think they’ve kind of slightly run ahead of the technology. The technology is absolutely not in any position to do any of that.
In the end of the day, you’ve got to remember, these large language models, they’re like– they’re like some giant autofill. They kind of just guess the next word or actually they turn the words into tokens in a sentence in response to a human prompt.
So I think those concerns can be set aside. I think some of the concerns around deepfakes, the manipulation of audio and video and images and so on, I think that’s a legitimate area of concern. But I would say that the technology which generates that unwelcome material is also the gen– is also the technology we will use to actually identify and control it.
So looking at our own experience in Meta for instance, AI has been a fantastic weapon in our own defenses against misinformation, disinformation, hate speech, and so on.
I’ll give you a very concrete example, hate speech on Facebook, the prevalence of it is now down to 0.02%. So that means that for every 10,000 bits of content you might see if you’re scrolling endlessly on Facebook, you might see two bits of hate speech. And that is down by over 50% over the last couple of years purely because of advances in AI.
So it’s an adversarial space. In exactly the same way that I can create problems, it can also solve those same problems. And that’s exactly the kind of thing that companies like Meta should, and we do, we must continue to invest in.
ALLIE GARFINKLE: Nick, Allie here. What is Meta’s stance on regulating AI?
NICK CLEGG: Oh, I definitely think you need regulation. I think with new technology like this, of course, it needs to be framed by democratically agreed laws. There are so many complex issues around, use of data, copyright, intellectual property, how do you make sure these systems are safe and transparent, how do you make sure they’re properly stress tested, what kind of regulatory bodies you need to invent if they’re not– if you don’t have sufficient ones already.
And all of that needs to be, of course, decided upon by legislators in this town in DC. That’s why we’re working at Meta, and all the companies are working closely with Senator Schumer and others as he’s leading an effort to– across parties to see what political agreement there might be around this.
But it doesn’t– we’re not going to sit around and wait for legislation. And that’s partly the point of these voluntary commitments. I think they will– I think they will end up sort of providing the guts of a lot of legislation both here and in other parts of the world.
But why it was so useful at the White House, it kind of pushed all the companies to be involved in all of this, is it means that we were starting to be able to agree on some basic swim lanes on what is and what is not a desirable well before legislation is actually passed.
ALLIE GARFINKLE: To that end, Nick, you know, Meta has made its large language model open source. Won’t that in theory make it more difficult to regulate, more difficult to manage?
NICK CLEGG: No. I mean, the history of open sourcing, and the whole internet has been built on open source technologies. And we’ve been open sourcing AI tools for years over the last decade. I think we’ve open sourced over 1,000 AI databases and models.
The basic spirit of it is that if you put these technologies– because they’re very expensive to build. There are only few companies that have got the very expensive computing capacity to build them. If you make them available to academics, researchers, developers, entrepreneurs, that, in turn, creates a kind of flywheel of innovation, which– which would not be possible if it was only– if these systems were kept under lock and key by the companies themselves.
And I should stress, of course, we wouldn’t release a model like this if we felt that it posed some existential threat to the world. So I explained earlier, these large language models, we will look back on them in several years to come.
The large language models of today are still relatively primitive. They are, as the name implies, they’re large language models. They’re literally just guessing the next word in a string of– string of words.
But we think right now, given the relative immaturity of these large language models and given all the stress testing we’ve done, we had 350 people so-called red teaming these models for several months before we released them to others, means that we can strike that balance of yes, innovating, but also making that innovation available to more people, which in turn hopefully leads to far more innovation in the AI economy as a whole.
– Nick, when it comes to misinformation on AI, there’s been lots of talk about how this could potentially influence or what the run-up to the election could look like now that we have AI and the mass adoption of AI. When it comes to Meta’s role in this and how you’re approaching the 2024 election, because now you also have threads which has over 100 million users, what does your playbook look like this time around?
NICK CLEGG: I think the key thing I’d like to say about this is that the way we build our systems, not least our AI systems, which– which– which make sure that bad content and nefarious content, whether it’s disinformation, whether it’s foreign attempts to interfere in elections, whether it’s hate speech and so on, those systems will tackle content regardless of where it’s created, whether it’s created by a human being or by a machine.
So whilst the volume may change, the fundamental architecture of our protections, which we’ve built up over the years and not least because of the experience of what happened back in 2016 and because of the immense advances in AI– we’ve spent $16 billion over the last few years on those kind of defenses and protections. We have tens of thousands of people working on this around the clock. Those AI systems will tackle misinformation, disinformation, where we can identify it regardless of its source.
And that’s, I think, important to remember. It’s not as if– as I say, the velocity and volume may change. But the fundamental power we have is because we are a platform that can control the distribution of that content. As long as we can identify it, we can then restrict or even remove that content if we find it.
ALLIE GARFINKLE: Nick, this all totally makes sense to me. But my question is, how do we know it’s going to work? Let’s fast forward 18 months. Let’s say, the election is over. How do we know that Meta has succeeded in mitigating misinformation?
NICK CLEGG: Well, look, my crystal ball is no clearer than yours. All I can point to you is– I can’t tell you exactly what’s going to happen in the future and particularly not in something as unpredictable as elections. What I can reassure you about is that– do remember, we are a global platform. Over 95% of our users are outside the United States. And we have elections every single month in many different parts of the world. So it’s not as if we only ever have to prepare ourselves for the US election. We have teams. We have highly sophisticated automated models, which are working to deal with elections nonstop all-year round, month in, month out. And with each election, we learn more because it’s a very adversarial space.
So I do think we’re getting better. We have constant practice in doing this. We have made massive investments, I think world-beating and certainly industry-leading technology to deal with misinformation, disinformation and so on. We give users as much information as we possibly can.
And as I said earlier, it doesn’t really matter from our point of view, whether the material, if it’s unwelcome and we want to block it or restrict it, is produced by a machine or a human being. We will try and tackle it.
But look, so I can only tell you that I think we try our level best, and we do so on an ongoing basis. But exactly how that will play out, well, that’s something that obviously we can speak about after the– after the elections themselves.
