Unlock Editor's Digest at no cost
Roula Khalaf, editor of the FT, picks her favorite stories on this weekly newsletter.
It's now been greater than a decade since I took my first ride in a self-driving automotive. As I drove a Google prototype through Mountain View in Silicon Valley, what promised to be an exhilarating taste of the long run quickly became quite boring – the robot automotive just drove so easily. A 2017 deadline for bringing autonomous vehicles to market, set in 2012 by Google co-founder Sergey Brin, seemed aggressive but not implausible.
Of course, Google didn't meet that deadline, just as Elon Musk's Tesla has done with autonomous vehicles through the years.
I used to be reminded of those shaky schedules once I read Sam Altman's New Year's Greetings “Reflections”. The OpenAI boss is “now confident that we all know how you can construct artificial general intelligence” – the purpose at which AI’s capabilities surpass those of most humans – and predicts that “in 2025, there often is the first AI agents enter the world of labor.”
AI agents that may plan and perform a variety of tasks on a user’s behalf – akin to purchasing items or booking travel – are the industry’s newest holy grail. Many view general-purpose agents as a vital component for achieving AGI (artificial general intelligence), although they initially deal with specific tasks. But because the autonomous driving industry has discovered over the past decade, there's an enormous difference between AI that works well in a restricted environment, like a chatbot window, and a “free-roaming” agent in the true one world is let go.
There is little question that 2024 was a turning point for Waymo, as Google's self-driving automotive project was renamed in 2016. Before last 12 months, Waymo had carried public passengers on a million rides for the reason that initiative began in 2009. In 2024, the whole was 1 million rides, increasing by one other 4 million as Waymo expanded its service to the general public from Phoenix to San Francisco and Los Angeles.
Waymo co-chief executive Tekedra Mawakana said on the Consumer Electronics Show this month that Waymo's robotaxis now travel 1 million miles per week – “greater than an individual drives in a lifetime.” According to Waymo's own studies across greater than 25 million miles of autonomous testing, its system can be safer than humans, with vehicles having far fewer serious collisions than when an individual is behind the wheel.
Waymo's acceleration was all of the more remarkable in a 12 months through which General Motors shuttered its cruise robotaxi division after pouring $10 billion into the corporate since 2016 and Apple abandoned its Project Titan automotive project.
The optimism for autonomous vehicles from the mid-2010s has returned. “The AV revolution has arrived – after so a few years,” Nvidia CEO Jensen Huang said in his own CES keynote, predicting that autonomous vehicles could be “the primary multi-billion-dollar robotics industry.”
Waymo wins by bucking the hype and never participating. Aware of the backlash that forced Uber out of AV traffic after one in every of its vehicles killed a pedestrian, Waymo has prioritized safety over growth. Even after Waymo moved its vehicles from the “lab” into public use, the corporate continued to check its robotaxis in Phoenix – a less complex environment than a city like San Francisco – for greater than three years.
Breaking through the large city still required fixing some very strange bugs, akin to the bizarre bursts of honking at 2 a.m. that woke up San Francisco residents near a Waymo parking zone last summer. It's still determining other complications that arise when AI meets humans, akin to how chaotic their cars can grow to be when there's no driver to control the passengers.
The company's long robotaxi journey contrasts with ChatGPT's rapid growth, with the chatbot attracting greater than 300 million users in only two years. Scaling an app and deploying a taxi fleet are very different propositions. But whilst generative AI agents operate in a purely digital world, there’s a big gap between co-pilots that work side by side with humans and fully autonomous bots that may reliably roam the open web.
The security issues with online agents will not be life-threatening, however the risks aren’t any less real. For example, will AI developers or bank card corporations bear the prices if a hallucinating agent goes on an unauthorized spending spree? And even with current AI tools, many corporations are finding that uncooperative employees and mismatched organizational structures pose greater obstacles than flaws within the technology itself.
Altman's prediction that AI agents will reach their potential should still be a full decade away. But developing a field AI that's as reliable as a human could still prove to be just as difficult as Waymo's long road to robotaxis.
.