What The Departure of Tesla's Ai Chief Tells Us
The biggest news in artificial intelligence in July 2022 was undoubtedly the unexpected resignation of Andrej Karpathy, head of Tesla AI and Autopilot.
While Karpathy may not be a household name, the 35-year-old scientist is a big deal within Tesla. He oversaw the artificial intelligence department and was the public face of Tesla's proud Autopilot program.
Musk poached the founding member of the artificial intelligence research firm OpenAI in 2017, and he became integral to fulfilling two of Musk's most ambitious promises: fully self-driving cars and humanoid robots.
A graduate of Stanford University with a PhD in computer science, Karpas speaks like a typical intellectual, passionate about training deep neural networks (a kind of computer brain) rather than mundane problems related to business.
Karpas did not explain his departure, which he announced on Twitter, and has no concrete plans for what he will do next. While his departure is not necessarily a sign of major problems in Tesla's AI department, it also never bodes well for the near-term prospects of any related future projects.
The departure of a central figure in Tesla's AI department is also a reminder to us: pay attention to the reality behind the hype in AI.
Self-driving technology requires complex decision-making and perception capabilities and is a useful proxy service for measuring whether artificial intelligence is on the verge of equaling or exceeding human capabilities. For several years, self-driving cars have seemed on the verge of becoming a mainstream technology as AI researchers conquer milestone after milestone.
In 2019, Musk claimed that Tesla would have 1 million robotaxi on the streets within a year - which didn't happen.
While other specialized self-driving technology companies are already offering fully autonomous robotaxi in cities like Phoenix or San Francisco, many experts believe that self-driving fleets are still many years away from having a significant impact on transportation in most cities.
The problem is that driving is a much more difficult task than most artificial intelligence engineers imagine.
There are too many things that can happen on the road. Even in countries with well-marked highways, predicting and training AI for every possible scenario is a huge challenge, not to mention that much of the world looks more like India - with cars, motorcycles, rickshaws, donkey carts and even elephants all coming together to form chaotic bumper-to-bumper venues.
Human drivers successfully navigate such situations every day, but this environment is far beyond the scope of what today's AI technology can handle.
On top of that, the advanced sensors and artificial intelligence-based computer vision systems used in self-driving cars have trouble coping in rain, fog and snow. That's why Cruise and Waymo are testing their robotaxi in sunny places like Arizona and California (both companies have been asked to suspend operations in the event of heavy rain or fog in San Francisco).
Neither traditional car companies, such as General Motors, which has pledged to sell self-driving cars by 2025, nor technology companies, such as Apple, which has been trying to develop self-driving cars for nearly a decade, have figured it all out yet.
Cruise, a self-driving car company backed by GM, also caused a major traffic jam in San Francisco in late June. At least a dozen of Cruise's self-driving Chevrolet Bolts were found blocking the intersection of Gough and Fulton streets for hours.
Humanoid robots, such as Musk's "Optimus" robot, which he promises to launch in September, are another sci-fi staple that has captured the imagination of technologists.
Admittedly, all the innovative consumer gadgets on the market right now (from drones to AR glasses) do suggest that the concept of " robots" doesn't seem so far-fetched. But creating a robot that can do a range of things around the house is in some ways a more daunting problem than autonomous driving.
Even the average apartment has different kinds of furniture and floor surfaces to navigate, unpredictable conditions from pets to toddlers, and a plethora of possible tasks to learn that can quickly exceed the capabilities of most of today's most advanced artificial intelligence systems.
It's not that there hasn't been rapid progress in robotics, it's just that the most capable robots are still currently dedicated to a specific task, such as: picking up objects; moving things around a warehouse.
While Musk is likely to surprise the world in September - he promises to show the Optimus Prime prototype - he is known to have a habit of boasting that he has been saying every year since 2014 that achieving fully autonomous driving will only take another year.
Currently, there are a host of legal and regulatory issues that need to be understood and addressed for both self-driving cars and mobile robots. The National Highway Traffic Safety Administration's investigation of a crash involving Tesla's self-driving technology is just a small preview of the scrutiny to come.
We are indeed living in a golden age of artificial intelligence compared to previous decades. So, I believe the entire industry is eagerly awaiting what Karpas will do next. But calm down, we are still a long way from the desired future.