Hello and welcome to Eye on AI.
This week, Waymo co-CEO Tekedra Mawakana announced the self-driving vehicle company is now giving 100,000 paid robo-taxi drives per week across Los Angeles, San Francisco, and Phoenix. The number is double the 50,000 paid weekly rides the company announced just a few months ago in June and marks a significant milestone for the company. It’s also a significant showing of AI on our roads driving alongside humans every day.
Autonomous vehicles from robo-taxis to self-driving trucks are highly dependent on AI. Some—like Waymo’s—combine extremely detailed maps with rules-based systems and real-time machine learning to navigate the roads, while other makers are moving toward systems that rely even more on AI to assess a vehicle’s driving environment in real time. Not only does AI make self-driving technology possible, but autonomous vehicles are also a quintessential example of AI’s many challenges. For example, hallucinations and accuracy dominate conversations around the leading commercial AI models. But when it comes to self-driving cars, the accuracy is literally a matter of life or death.
Yet, self-driving cars are not a big part of today’s AI conversation, and at the same time, autonomous vehicle makers aren’t talking much about AI. While self-driving cars were held out as the ne plus ultra of AI a decade ago, Waymo’s website doesn’t mention AI even once, for example, only including one brief nod to machine learning toward the bottom. It’s a drastic difference compared to practically every other industry from enterprise software to education, health care, law, retail, advertising, and so on, where companies are essentially shouting from the rooftops about how they’re using AI. That contrast is striking because while many companies across sectors are rushing to adopt AI without a clear strategy because it’s the hot thing to do (and because investors have bid up the stock of any company claiming to be “AI-first” and saying a product is “AI-powered” has become the go-to way to market it), self-driving cars are a truly disruptive technology that actually depends on AI to function. So why aren’t self-driving car companies seizing the AI moment?
“It’s because of regulation. It’s a closed community,” Eran Ofir, CEO of Imagry, a company that creates mapless AI autonomous driving software, told Eye on AI, noting that “AI is happening every day on the road with hundreds of thousands of vehicles.”
Long before generative AI models like ChatGPT dominated the AI conversation and put AI at the top of the legislative agenda, autonomous vehicle makers were jumping through regulatory hoops to get permits to test and drive their vehicles on roads. They’ve had serious setbacks. Cruise was banned and had to recall its cars after an accident last year and today the General Motors-owned company agreed to an additional recall of 1,200 of its self-driving cars to address federal highway safety regulators’ concerns over a “hard-braking” problem.
But, if one takes a longer view, the industry has actually made a lot of progress convincing regulators and the public to trust it, as Waymo’s milestone and continued expansion shows. Then came generative AI, a totally different type of AI model that opened up several new cans of worms. While many of the issues around generative AI—like copyright and deepfakes—don’t even apply to the AI systems that power self-driving vehicles, they’ve made AI a topic of heated debate and caused everyone from consumers to regulators to be more skeptical of the technology and its impact. In short, for the self-driving vehicle industry, attaching itself to the AI hype is more of a risk than a selling point.
“The discussions with the regulators are difficult, and they don’t want all the bad effects and debate about whether AI should be restricted and used and stuff like that,” Ofir said. “No one wants that side of it to be imposed on autonomous vehicles.”
Another interesting distinction is that while the goal for self-driving vehicles is to make something we already do (driving) safer, generative AI is largely aimed at making tasks faster and easier, sometimes at the expense of safety. Generative AI has turbocharged scams, introduced new cybersecurity threats, given rise to tools that make it possible for anyone to create nonconsensual explicit deepfakes of other people, and overall leads us to believe things that are not real and distrust real things we see with our own eyes.
Overall, this leads to interesting questions about how we evaluate the risks and impacts of various AI applications—the very questions lawmakers trying to figure out how to regulate AI are grappling with. Are self-driving cars an AI technology, or is AI just a technology used in self-driving cars? Is it both? Does it matter?
Different types of AI can work very differently, achieve very different impacts, and create very different risks. One thing that remains clear, however, is that AI is already in use all around us—even driving in the next lane.
And with that, here’s more of today’s AI news.
Sage Lazzaro
[email protected]
sagelazzaro.com
AI IN THE NEWS
U.S. political campaigns are steering clear of AI tools. Over 30 tech companies have pitched their AI tools to U.S. political campaigns for November’s election, but the vast majority of the campaigns aren’t biting, according to the New York Times. The few that have didn’t want to admit it and found that the technology fell flat. The AI technologies pitched include products that reorganize voter rolls, recreate candidates’ likenesses, and voice tools that can make tens of thousands of personalized phone calls to voters. “People just didn’t want to be on the phone, and they especially didn’t want to be on the phone when they heard they were talking to an AI program,” Matthew Diemer, a Democrat running for election in Ohio’s Seventh Congressional District, told the Times of his trial with a tool called Civox. His campaign used the AI program to make personalized voter calls that invoked his talking points and sense of humor. The AI program made almost 1,000 calls in just five minutes, and almost all of the recipients hung up upon hearing the voice on the other end describe itself as an “AI volunteer,” according to the Times.
Microsoft schedules a limited rollout of its controversial Recall AI tool for October. That’s according to The Verge. The software giant planned to launch the feature—which screenshots whatever a user does and sees on their computer and allows them to search through archives and retrieve what they’ve seen—to testers in June, but delayed the initial launch over backlash surrounding security concerns. Recall relies on local models built into the Windows operating system, but security researchers found the database wasn’t encrypted and that the information could easily be accessible to attackers. Nothing on a computer connected to the internet will ever be entirely safe, but the amount of extremely sensitive data Recall keeps and how easy it makes it to access it alone is cause for concern, experts said.
NIST wants you to help red-team AI office software. AI and algorithmic assessment nonprofit Humane Intelligence called for U.S. residents—both software developers and everyday users—to take part in the qualifying round of a nationwide challenge to red-team (or attack) AI productivity software products to test them for flaws. In his AI executive order last year, President Joe Biden called for a series of such challenges to be conducted through the U.S. National Institute of Standards and Technology (NIST), and this is the first of several Humane Intelligence plans to announce in collaboration with government agencies in coming weeks, according to Wired. Participants who proceed through the online round will join an in-person red-teaming event at the end of October at the Conference on Applied Machine Learning in Information Security (CAMLIS) in Virginia.
FORTUNE ON AI
A new web crawler launched by Meta last month is quietly scraping the internet for AI training data —by Kali Hays
Andreessen Horowitz leads $80 million bet on startup seeking to tame AI with copyright —by Jeff John Roberts
LVMH’s Bernard Arnault has quietly invested in 5 AI startups this year via his family office —by Prarthana Prakash
AI CALENDAR
Aug. 28: Nvidia earnings
Sept. 10-11: The AI Conference, San Francisco
Sept. 10-12: AI Hardware and AI Edge Summit, San Jose, Calif.
Sept. 17-19: Dreamforce, San Francisco
Sept. 25-26: Meta Connect in Menlo Park, Calif.
Oct. 22-23: TedAI, San Francisco
Oct. 28-30: Voice & AI, Arlington, Va.
Nov. 19-22: Microsoft Ignite, Chicago, Ill.
Dec. 2-6: AWS re:Invent, Las Vegas, Nev.
Dec. 8-12: Neural Information Processing Systems (Neurips) 2024 in Vancouver, British Columbia
Dec. 9-10: Fortune Brainstorm AI San Francisco (register here)
EYE ON AI NUMBERS
50 million
That’s how many views just one video of Chubby, a cat people are generating using AI, has garnered thus far. The BBC reports that videos of Chubby are drawing millions of views and a devoted following online, “blurring the line between spam and art.” Chubby is distinctly “rotund, ginger,’ and depicted in sad situations like facing a schoolyard bully or being addicted to cigarettes, which makes it easy for users to create their own versions using any of the text-to-image models easily available online. It’s no surprise that a cat marks one of the first memes of the AI content age (save for maybe Shrimp Jesus). From Nyan Cat to Grumpy Cat, cats have always been on the frontlines of every era of internet content.
