Tidy up, make dinner, mow the lawn, play catch with the kids — these are just a few of the chores Elon Musk promises his robot Optimus will one day perform.
The arrival of his bipedal droid and other forms of artificially intelligent machines like it will eventually usher in a new “age of abundance,” the Tesla CEO predicted. Thanks to their labor, the only goods and services that will be scarce in the future will be those we as a species deem necessary to limit.
“People have no idea,” a frustrated Musk said in April. “This is going to be bigger than the car.”
A robot in every home, in other words.
If Tesla generated $50 billion in 2021 from its core business—a number set to grow further this year—that means Musk believes sales of Optimus would eventually top that. After all, who doesn’t dream of their very own mechanical butler?
With bold pronouncements like these, it’s no wonder that expectations are running sky high for what the visionary entrepreneur may unveil at Tesla’s “AI Day” on Friday. Many believe he will finally pull the cloth off the Optimus bot that he first showed as a mock-up in August 2021.
As keen as they are to see it themselves, many robotics experts Fortune spoke with are worried. The Tesla CEO has a long history of over-promising and under-delivering when it comes to artificial intelligence. For example, U.S. road transportation regulator NHTSA has serious concerns over the safety of his Full Self-Driving (FSD) feature, which allows drivers to take their hands off the wheel as long as they watch the road.
However, Musk is this generation’s Steve Jobs—the man famous for creating a reality distortion field around him. And if on Friday he inflates expectations that cannot subsequently be met, IT services and consulting firm Cognizant warned it could lead to public disillusionment with robotics and A.I. more generally.
“Too much hype not a good thing,” said Babak Hodjat, its chief technology officer for artificial intelligence. “The bar is being set too high here.“
As affordable as a cheap car
If anyone can do it, the optimists retort, it’s Musk. The ultimate technologist is blessed with an uncanny knack for predicting humanity’s future needs.
Musk first helped revolutionize online payments with PayPal, before making electric vehicles wildly popular with the Tesla Model S, and now plans to “occupy Mars” with SpaceX. When Russia invaded Ukraine, Kyiv reached out directly to him to restore internet service by using his Starlink, a service whose satellites in low Earth orbit are now the best hope for Iranian women to coordinate their protests for freedom.
Musk claims that, by comparison, the feat of building a droid is almost simple. He already has the necessary expertise in computer vision as well as a silicon brain that can process a robot’s environment—that is, the Full Self-Driving (FSD) computer that is currently installed in every Tesla. All he needs is to continue training this chip with the help of his new Dojo neural net cluster — the focal point of last year’s AI Day.
Since the mechanics of the 5’8”-tall Optimus will be considerably less complex than a Tesla, he reasons they won’t be any more expensive than a cheap car, at least once they are industrialized. For that price, you will get a sleek droid walking on two legs that can respond to commands, navigate its environment and even deadlift 150 pounds.
“The missing things are basically real-world [artificial] intelligence and scaling up manufacturing. Those are two things that Tesla is very good at,” Musk said. “So then basically we just need to design the specialized actuators and sensors that are needed for a humanoid robot.”
Notably, Musk is expected to stop short of giving the robot a face similar to Hanson Robotics’ Sophia or the Ameca bot from Engineered Arts. Beyond adding complexity, robotics experts warn giving a machine too human a look can backfire: people may be inclined to anthropomorphize a droid, for one, while mimicking facial expressions can be bizarre and unsettling if done poorly. Not having a face works like a warning sign then, reminding people that they are dealing with a machine.
Pieter Abbeel, a professor of robotics at the University of California at Berkeley, praised Musk for tackling the challenge of creating useful robots, and argues the Tesla CEO is right to frame it as more of an artificial intelligence problem—a software issue—rather than primarily a hardware problem.
“I think [robotics] is a space that has been honestly under the radar compared to self-driving,” said Abbeel.
While experts like Abbeel warmly welcomed the attention Musk can bring to robotics, they warned against being all too easily dazzled by whatever he shows. Make no mistake: under the hood, they say, A.I. is still in its infancy.
“Elon’s timelines are quite optimistic compared to reality,” said Abbeel, who co-founded an A.I. company called Covariant that creates the software behind some of the world’s most capable robotic arms for use in warehouses.
First off, the technology simply does not exist to create a robot with the agility to accomplish delicate tasks ranging from washing dishes to folding clothes and putting away groceries. Applying precisely the appropriate amount of force at all times across different tasks is something machines struggle with.
“At the moment, robot dexterity is not even up to the level of a human infant,” Natan Lepora, a professor of robotics and artificial intelligence at the University of Bristol, told Fortune.
One of the best human-like robot hands in existence is called the Shadow Hand, which uses artificial tendons to achieve remarkably life-like movements. But robotics experts who have worked with it say it is not well-suited for a mass consumer product because it needs frequent maintenance to keep the tendons working properly.
And it’s not just the hands or face that are highly complex. Even walking on two legs is challenging. Many bipedal robots have a uniquely odd staccato walk that fails the test of mimicking a human.
When Boston Dynamics, the Massachusetts Institute of Technology spin-off famous for its Atlas humanoid robot, went commercial, it switched tactics. Only its Spot, a quadruped, and Stretch, a wheeled sibling with a single maneuverable arm, are sold to businesses.
Mostafa ElSayed, co-founder and CEO of Automata, a London-based company that produces robots, believes Musk could potentially deliver a bipedal robot with an impressive human-like gait. But the Walt Disney Company, an early backer of robot technology through its animatronic theme parks, has already beaten him to it with robots able to perform incredible acrobatic feats.
“It’s extremely impressive, but that is not a useful product,” said ElSayed. “It can’t do anything for you.”
Useful would be working in a factory or warehouse, rather than a home, as that is where demand is headed.
Figures provided to Fortune by the International Federation of Robotics indicate unit sales of autonomous mobile robots surged 45% globally last year as businesses tried to make their logistics operations more resilient from factors like COVID lockdowns.
Clumsy means dangerous
But a robot does not need two legs working in a warehouse — in fact it’s a drawback.
“Boston Dynamics is the gold standard, with access to the latest gyroscopes and actuators from the Department of Defense. Yet still you can find plenty of videos of Atlas falling over,” said Alexander Kernbaum, interim director of SRI International’s Robotics Laboratory. “And that’s a 250 pound robot. So imagine on a bad day what could happen – it could destroy your wood floor, kill your pet, or worse—your kid.”
In his opinion, Musk should tone down talk of home robots and focus instead on next-generation factory robotics. But Kernbaum expects Musk will nevertheless show a biped, because that is what he promised to his fans even though it’s not what the market needs.
“Everybody wants a robot to do their laundry and clean up after them, but I don’t expect to see a Rosie the Robot maid in at least the next 10 years,” he said.
Noel Sharkey, emeritus professor of artificial intelligence and robotics at England’s Sheffield University, suspects Optimus is just an attempt by Tesla to burnish its reputation for technological leadership at a time when the company is facing increasingly intense competition from established car makers and startups.
Westinghouse already knew to tap into excitement over robots as far back as the 1920s and 1930s with its Televox, and more recently Honda’s humanoid robot, Asimo, added a space-age gloss to the Japanese brand’s cars and motorcycles even though it was never sold commercially.
“This is a billionaire’s toy we are talking about,” said Sharkey.
Tesla and Musk are not the only ones working on humanoid robots for mass consumption. Lei Jun, the founder of Chinese tech giant Xiaomi showed off his vision for what a humanoid robot looks like with CyberOne. And while that news may not have reached many Western ears, it will have impressed in China—Musk’s largest growth market.
Facing such competition, the Tesla CEO needs an eye-popping success after admitting how he was repeatedly fooled by false dawns when it comes to robotaxis. He is nearly three years behind delivering even one fully autonomous Tesla, let alone the million he promised.
When he used a much-anticipated product roadmap update in January to then declare Optimus was the most important new hardware Tesla was developing this year, his pivot away from new car models infuriated markets. For many professional investors, it looked like Musk was yet again attempting to distract people from painful delays to his promised products.
Tesla’s electric Semi, promised to companies like Anheuser Busch in December 2017, is already running a minimum of three years late, while rivals have already launched their electric trucks. The Roadster remains on the distant horizon, and Musk’s Cybertruck has also been marred by major setbacks.
It’s not just Tesla either. Neuralink still hasn’t tested its first brain implant in humans, and Musk is reportedly disappointed with its progress. And the hyperloop pods he predicts can transport people on in less than a half hour from NYC to Boston on magnetic skis inside a vacuum tube are so far just a fantasy.
Detractors like Stanphyl Capital’s Mark Spiegel call him a fraud as a result, and criticize prototypes as little more than pretty vaporware that will never make it to market.
A more generous take would be the demanding CEO is quite simply poor when it comes to accurately judging what his engineers can realistically accomplish in a short time frame.
Only weeks after Musk declared Optimus to be his top priority, he casually informed markets by tweet the widely-respected senior director spearheading his A.I. program was taking an extended sabbatical. Despite assurances Andrej Karpathy would return, he never did — not the first casualty of an oddly-timed Tesla sabbatical.
Even if he can build one—should he?
But what about the ethical argument the Tesla CEO has made, where human labor is no longer needed and people risk losing a purpose in this new age of abundance?
Musk after all has actively cultivated a reputation over recent years for warning everyone all the way up to Barack Obama that mankind must be protected from the threat posed by artificial intelligence.
“I tried to convince people to slow down A.I., to regulate A.I.—this was futile,” he said in 2017, admitting he had become fatalistic, because “nobody listened.”
In fact, Musk founded his company Neuralink, which wants to implant microchips in people’s brains that will allow them to communicate with computers through thought, largely because he said he has come to believe that it is the only way humanity will keep pace with advanced artificial intelligence and be able to control it.
“It’s going to be important from an existential threat standpoint to achieve a good A.I. symbiosis,” he said in August 2020.
It was therefore revealing when Musk was asked last August about how robots fit in Tesla’s mission of sustainable transport—effectively should he make them? The Tesla CEO responded quite simply either he would develop A.I.-enabled robots—or the competition would.
“We are just obviously making the pieces that are needed for a useful humanoid robot,” he revealed last August. “So I guess we probably should make it, and if we don’t, someone else would.”
Sign up for the Fortune Features email list so you don’t miss our biggest features, exclusive interviews, and investigations.