The AI Daily Brief: Artificial Intelligence News and Analysis - 76% of Americans Think AI Might Kill Us

Episode Date: August 10, 2023

A new survey from the AI Policy Institute shows that Americans are extremely concerns about AI and support regulation to slow it down. Before that on The Brief: Disney explores AI; DARPA starts a $20M cybersecurity challenge; and a New Zealand Grocery AI app goes horribly awry. Today's Sponsor: Supermanage - AI for 1-on-1's - https://supermanage.ai/breakdown ABOUT THE AI BREAKDOWN The AI Breakdown helps you understand the most important news and discussions in AI.  Subscribe to The AI Breakdown newsletter: https://theaibreakdown.beehiiv.com/subscribe Subscribe to The AI Breakdown on YouTube: https://www.youtube.com/@TheAIBreakdown Join the community: bit.ly/aibreakdown Learn more: http://breakdown.network/

Today on the AI breakdown, we're looking at a survey that suggests that Americans are very nervous about AI. Before that on the brief, DARPA has announced a major new AI cybersecurity competition. The AI breakdown is a daily podcast and video about the most important news and discussions in AI. Go to Breakdown Network for more information about our YouTube, our Discord, and our newsletter. Welcome back to the AI breakdown brief, all the AI headline news you need in around five minutes. We kick off today with a story that. that hopefully is reflective of a shift in the phase that we're in, moving from just talking about whether we should be dealing with AI-related issues to actually dealing with AI-related issues.

Yesterday, the White House and DARPA announced a new artificial intelligence cyber challenge, through which they will give out millions of dollars in grants to help teams figure out how to update infrastructure for a new era. The White House writes, The Biden-Harris administration today launched a major two-year competition that will use AI to protect the United States' most important software, such as code that helps run the internet and our critical infrastructure. The AI Cyber Challenge, AIXCC, will challenge competitors across the United States to identify and fix software vulnerabilities using AI. Led by DARPA, this competition will include collaboration with several top AI companies, including Anthropic, Google, Microsoft, and OpenAI,

who are lending their expertise and making their cutting-edge technology available for this challenge. The competition, which will feature almost 20 million in prizes, will drive the creation of new technologies to rapidly improve the security of computer code, one of cybersecurity's most pressing challenges. So one of the things that's interesting about this is that they're inviting both teams that can pay for themselves to participate, but they're also making about $7 million in grants available for small businesses who want to participate who otherwise might not be able to. They have registration for a open track, which doesn't involve that funding, as well as a funded track, which does give companies access to that. All of that kicks off

this fall, and then teams spend a year in qualifiers after which the top 20 teams advance to the semifinals, which happened next August. During the semifinals, the top 20 teams are whittled down to the top five teams, each of who get an additional $2 million, which leads to the finals a year later in August 25, when the winners are announced with first getting 4 million, second getting 3 million, third getting 1.5 million. Now, intuitively, I have some amount of skepticism around this sort of funding or competition structure leading to the best results, but that might be just because we're not used to seeing this approach. Maybe this is actually a really good way to incentivize people to work on these types of issues, rather than whatever they might be doing in their commercial

startup anyways, especially to the extent that this program ends up being kind of a feeder into those larger AI labs as well. In other words, it's not a bad way to spend a couple years working on an important problem, getting access to some government money, and showing off your skills to companies that might hire you or your team later. Either way, for me, I am mostly glad to just see actual action happening in this space, and it'll be really interesting to see how this plays out. Next up today, we have a couple of updates from stories we've covered recently. The first surrounds China and AI chip orders. We've talked a lot recently about how there is increasing pressure on the Biden administration to make even further restrictions on the AI chips that are

available to China. Since chip restrictions went into place over the last couple years, Nvidia started to produce a chip that was lower powered so that it could still be available to the Chinese market. Now, with all the scuttle butt that the U.S. might clamp down even farther, apparently the big Chinese companies have been placing an absolute flurry of orders for AI chips. According to sources close to Nvidia, Baidu, Tencent, Bite Dance, and Alibaba have placed orders worth $1 billion for the rest of 2023 and around $4 billion for 2024. Our next update comes from the big Zoom dust up. You'll remember that over the weekend, people noticed that a Zoom terms of service update

seemed to give Zoom a heck of a lot of access to train their AI on customer data. Now, quite quickly, the noise got to the level where they had to have an official response, adding a line in the terms to make it very clear that users would have to opt in for that data collection to occur. And presumably they hope that would be that, but it wasn't. Now the Zoom CEO has had to come out and actually may a culpa it and say that it was a mistake. Eric Yuan, Zoom CEO said in a LinkedIn post, given that many friends have reached out to me asking about what had happened over the last two days, I want to set the record straight. One, it is my fundamental belief that any company that leverages customer content to train its AI without customer consent will be out of business overnight. Two, given Zoom's value of

care and transparency, we would absolutely never train AI models with customers' content without getting their explicit consent. Three, then why did we make a TOS change in March 2023? We had a process failure internally that we will fix, period. Let me be crystal clear. For AI, we do not use audio, video, screen share, or chat content for training our AI models without customer explicit consent. Four, many of our users have asked me what Zoom's next steps would be after I shared this

post on August 8th. We have two generative AI features on free trial. Even though we have several customers who have opted in to share their content, we haven't used any customer content to train our AI models. After considering all of your feedback, we are committing to all of our customers that we will never use any of their audio, video, chat, screen sharing, attachments, or other communication like poll results, whiteboard, and reactions to train our AI models. We will change our general terms of service, free trial terms of service, and our product

to reflect this decision very soon. So this is a big response. This isn't just a note saying, oops, we messed up. It's Zoom feeling like they are so backed into a quote. corner that even if customers opted in, they didn't feel like they had the trust to any more ask for that consent. You have to think that this is a shot across the bow for AI companies who have just got a sense at this point that there is growing acrimony towards data collection for AI training purposes. Over in the world of TikTok, that platform is trying to make it

easier to identify which content is generated by AI. It is apparently now giving users a simple toggle to let them identify if they used AI in the process of creating their videos. Now, in addition to allowing users to identify that their content is AI generated, it also warns that they must disclose that AI-generated content or risk having it taken down. Speaking of content and AI, the latest big entertainment company to explore artificial intelligence appears to be Disney. According to Reuters, Disney has created a task force to study AI and look at how it might be applied across numerous divisions ranging from their parks to movies. Reuters points to 11 current job opening seeking candidates with expertise in AI and machine learning

that they note touch, quote, virtually every corner of the company, from Walt Disney Studios to the company's theme parks and engineering group, Walt Disney Imagineering, to Disney branded television and the advertising team. Now, interestingly, one source of Reuters said that it was a cost-cutting measure, or rather that they are looking to AI as a way to bring down the cost of production, which has soared in recent years. Again, perhaps reflecting less Disney strategy and more where mainstream media discourse is, every article that has been picked up about this

has included the idea that this is for cost-cutting, right in the title, even though that's just speculation from one source and isn't necessarily reflected in any of those job titles. Again, I think this has a lot to say about where Americans are in general with AI, perhaps more so than it has to say about Disney itself. Lastly today, one from the uh-oh files. A grocery chain in New Zealand has tried to do something cool. They tried to create an AI meal planner app that would take the leftovers or random things that a customer had in their fridge and try to come up with creative ideas for recipes that could make better use of those goods so they didn't go to waste. This is a really cool thing in a time when everyone's trying to pinch

pennies and cut down their budgets, right? Well, the problem came when people started adding items to their list that were not necessarily food items. One example picked up by the Guardian was a recipe that the AI app dubbed aromatic water mix. The bot recommended the recipe as, quote, the perfect non-alcoholic beverage to quench your thirst and refresh your senses. Serve chilled and enjoy the refreshing fragrance. Unfortunately, the recipe that it gave would create chlorine gas, the inhaling of which can cause lung damage or even death. Now, of course, the supermarket said that this was just pranksters being bad, and that they were disappointed to see a, quote, small minority have tried to use the tool inappropriately and not for its intended purpose, and that of course they

would, quote, keep fine-tuning their controls. But we have to remember, whether it's a savi mealbot or something much more powerful, the one thing that we can count on is people being people and trying to mess with it. When that leads to ant poison and glue sandwiches, bleach-infused rice surprise, and methanol bliss French toast, there may be a problem. That's going to do it for today's AI Breakdown Brief. If you enjoyed it and you haven't signed up yet for the newsletter, go check it out. You can get it on the AIB breakdown.Behive.com, that's B-E-H-I-I-V, and get all of this type of content in the morning first thing via email. Thanks for listening to and watching, and I'll be back soon with the main AI breakdown.

Before we get into the main AI breakdown, I want to tell you about today's sponsor, Supermanage. If you work in a professional setting, you probably have some version of a one-on-one meeting, either with the people that work for you or the people that you work with. Unfortunately, all too often, those one-on-one meetings become glorified catch-up calls. Don't you wish you could jump right to the stuff that really matters? That's where Supermanage comes in. Supermanage AI magically distills your team's public Slack channels into a real-time brief on any employee, any time. Catch up on contributions, work in progress, challenges they're facing,

sentiment, everything you need to show up ready for a truly meaningful conversation. And it's completely free. Visit supermanage.a.i forward slash breakdown today to start making the most of your one-on-ones. And thanks again to Supermanage for sponsoring the AI breakdown. Welcome back to the AI breakdown. Today we are talking about a new survey that is getting a lot of attention, and which appears at first glance to suggest that Americans are extremely worried about AI. So this was first written up in Axios. The piece was called exclusive poll Americans distrust AI giants. The piece kicks off. Majorities of American voters from both parties are worried about AI risks and support federal regulation to control those risks according to new polling by a new

AI-focused think tank, the Artificial Intelligence Policy Institute shared exclusively with Axios. And boy, there are some big banner numbers here. The first poll question that Axios shares is how U.S. adults say they feel about growth in artificial intelligence. Mostly excited? Only 7% of respondents. Somewhat excited, 14%. Totally neutral, 16%. Somewhat concerned 32%. And mostly concerned, 30%, nearly a third that are mostly concerned, and another third that are somewhat concerned. That was far from it when it comes to interesting numbers from this study, though. Here are some of the highlights, according to the organization that sponsored the study. 72% of voters prefer slowing down the development of AI compared to just 8% who prefer speeding

development up. 62% of voters are primarily concerned about artificial intelligence, while just 21% are primarily excited about it. 86% of voters believe AI could accidentally cause a catastrophic event, and 70% agree that mitigating the risk of extinction from AI should be a global priority alongside other risks like pandemics in nuclear war, 82% of voters don't trust tech executives to regulate AI, while voters support a federal agency regulating AI by a more than three to one margin, including two to one among Republicans. 76% of voters believe artificial intelligence could

eventually pose a threat to the existence of the human race, including 75% of Democrats and 78% of Republicans. Now, Daniel Colson, the executive director of this new Artificial Intelligence Policy Institute, wrote, the data is clear. Americans are wary about the next stages of AI and want policymakers to step in to develop it responsibly. At a time when nearly every issue is polarized, there's a broad consensus among Americans that policymakers need to decide what path to AI development should take. So before we discuss these poll results even further, let's talk about the organization that sponsored this survey. The AI Policy Institute's webpage says, translating public concern into policy. American voters are worried about risks from AI technology. The AI Policy

Institute's mission is to channel public concern into effective regulation. We engage with policymakers, media, and the public to shape a future where AI is developed responsibly and transparently. Now, as you can see, the perspective of the AI Policy Institute is very clear. They are not a neutral institution, and don't pretend to be. They are a body that exists because they are concerned about AI, because they believe that others are also concerned about AI, and because they want that to turn into actual effective regulation. Now, a little bit down their homepage, there is a section called slow deliberate progress. That phrase, I think, from the little that I've read on this website, seems to sum up what the goal is here. In other words, it doesn't appear to me to be some

neo-Luddite organization that doesn't want any advancement in artificial intelligence. Its concern appears to be around the financial incentive and the market competition around AI more than anything else. The AI frontier, they write, is being pushed forward rapidly by corporations pouring billions into training powerful AI models. The speed of this advancement has outpaced our understanding of these systems, leaving us in the dark about their capabilities, behavior, and the risks they pose. The AIPI points out that the leaders of all of the big labs have signed that letter saying that mitigating the risk of extinction from AI should be a global priority. And they even hint at what they see as perhaps the most significant first moves that

the government could make. They write, By regulating the data centers necessary for developing cutting-edge AI models, and by mandating the demonstration of an AI model safety prior to its deployment, government has the opportunity to significantly mitigate approaching threats. So that's who this poll is coming from. Now, of course, it wasn't the AIPI who was outdoing. That was done by UGov, who does tons and tons of polls across a million different topics, and are broadly speaking considered good at what they do, and not likely to nudge results in one direction, just to confirm the priors of whoever sponsoring the survey they're doing. On the website, they shared a set of the actual questions,

rather than just the summarized results. And the only one that felt at all leading to me was this agree. Agree or disagree, tech company executives can't be trusted to self-regulate the AI industry. That could have just as easily have been written agree or disagree. Tech company executives can be trusted to self-regulate the AI industry, but by putting the emphasis on the can't, it kind of nudges the respondent to be in the mindset of skeptical in the first place. Still, I share those things. only so that you have an appropriate grain of salt and treat this survey for what it is, which is one data point. Now, broadly speaking, and you've probably heard me say this on the show before, I agree pretty wholeheartedly with the idea that broadly speaking, Americans have a sense

and have grok the message of the AI safety folks. In fact, at some points, I talked about how I think that community needs to focus a little bit more on solutions, because there's sort of been this head-spinning shift where the issue went from off-the-radar entirely to something that people were really concerned about. Now, why might that be? What might explain why Americans are inclined to be nervous about this? I think there are a few aspects of it. One has to do with how not theoretical this is. And what I mean by that is that usually when you have a big technology hype cycle, it's a small cohort of early adopters who are really enfranchised than actually doing the thing in question. Not so with generative AI, where hundreds and hundreds of millions, if not

billions of people have actually used these tools and do actually use these tools now day in and day out. The experience of going from not really thinking about AI at all, and maybe only having a sci-fi understanding of what that word or term even meant, and then having Chatchebtee do your homework for you better than you can, or write those emails for work better than you can, or teach you about things that you want to learn about better than any other resource that you can find, is the sort of zero to one moment that really makes people question everything they think they know about the world around them. Now, I don't think that's necessarily a priori good or bad, but what I think people got from those first experiences with chat GPT was a sense of the significance of this change.

No disrespect to NFTs or any other crypto thing, but it was unignorable for most people who actually use these tools in a way that most technology hype cycles that came before them just weren't. So that's factor one. To sum up, Americans were primed to think that this was a real and significant thing. Now, second, that came into the context of waning and frankly fast-declining trust in big tech firms. If you go back and look at polls from the early teens, tech companies were still seen as the new kids on the block, much more trustworthy, for example, than big financial institutions. The global financial crisis was still fresh in people's minds, and these cute new social apps were still just fun, kind of insignificant things. Now, the more and more of our

lives that have been consumed by those big tech platforms, the more and more trust has gone down. And frankly, people of different political persuasions have found lots of different reasons to dislike big tech equally, if oppositely. Brookings published a study summarizing longitudinal polls from American institutional confidence polls. They looked at the loss and confidence of different institutions between 2018 and 2021. Now, it's worth noting that in that period, institutions of all stripes lost confidence, which I think will resonate with many people's anecdotal or lived experience, but by a fairly significant margin, the institutions that lost the most confidence in that three-year period

included Facebook, Amazon, and Google. Just after that were colleges, which had lived through the admission scandal, press, which was being completely barraged by the then-sitting president, and after that, major companies, FBI, military, political parties, organized labor, organized religion, court, state government, nonprofits, local government, banks, executive branch, local police and Congress. Somehow, Congress had the lowest amount of loss in confidence, but my guess is that has more to do with how low its starting point was, rather than it just being more durable than these other institutions. In any case, the point is, chat GPT and all of these generative AI tools happened at a time when we were primed to not trust Big Tech, to not trust the CEOs of Google and Microsoft,

to stare squintily at Sam Altman and ask whether he's just another Sam. So you now have, one, the significance that comes from actually experiencing these tools, to the priming of the populace to dislike and distrust big tech. And then three. For the last three months, we have had basically a never-ending barrage of mainstream media coverage of extinction risk. This was jump-started, of course, by Jeffrey Hinton leaving Google and going on his small media tour and going on his media tour about his concerns. But for a while there, every day there was another headline about the idea that AI might kill us all. So that's a third factor. A fourth and final factor is that people are concerned even if AI doesn't kill us all, it might take our jobs. Now, this perspective has gotten louder

because of AI's central place in the Hollywood strikes happening right now, but it's also just a perpetuation and an amplification of a trend in some cases of automation taking away jobs that has been happening for decades, combined with the fact that this is the first time that white collar jobs have been threatened, and you have this really palpable stew of reasons why people are going to be scared, why they have reasons to mistrust big tech, and why they understand that whatever happens, something big is going to happen. The question really comes down to what is Congress going to do with this broad bipartisan support? What type of rules and regulations are actually going to make sense and serve to address these concerns, while also not just reactionarily shutting down a field

of technological innovation that could also come with huge benefits. I will say my general bias is to not be super optimistic about Congress's ability to get that right. But I will also say that there seems to be some pretty serious good faith efforts to get up to speed very quickly that could really bear fruit. The most hopeful thing might be that the fact that this is such a clear, popular issue among Americans gives representatives of both party the cloud cover to actually try to get it right. Is that being too optimistic? We will just have to wait and see. But anyways, that is the story with this poll that you are likely hearing about right now. If you haven't yet seen them, go to theaIPI.org, check out this new organization, see if

you think it's interesting, or if what they have to say makes you skeptical, do your own research, come to your own conclusions, and then come back and let's chat about it. That's going to do it for today's AI breakdown. Thanks as always for listening or watching, and until next time, peace.