Listen to this story
In September this year, Tesla presented a glimpse of their general-purpose humanoid robot ‘Optimus’. Unfortunately, as many humorously remarked, the launch was “suboptimal”. The demo couldn’t do much to excite the audience for what’s coming next. However, many roboticists and experts in the field believe that if sufficient resources are thrown into it, the vision of an “all-purpose” robot is practical. Dennis Hong, director at Robotics & Mechanisms Laboratory, said this about the Optimus project:
While Hong disagrees with the proposed timelines under which Tesla will be able to mass produce humanoid robots, he agrees with the notion that the future is humanoid robots.
The human-like robot, also dubbed as ‘Tesla Bot’, will incorporate the AI-driven autopilot technology used in their self-driving cars. The focus on Optimus, Musk believes, will help realise the dream of Artificial General Intelligence (AGI). However, there are detractors to the same. An important aspect that forms a point of commonality between the detractors is the lack of usefulness and the practical impossibility of building “intelligent” robots.
Subscribe to our Newsletter
Join our editors every weekday evening as they steer you through the most significant news of the day, introduce you to fresh perspectives, and provide unexpected moments of joy
Special-purpose AI vs General AI
Andrej Karapthy, former director of AI at Tesla, acknowledged the position that argues for designing robots that cater to particular problems, but added that “getting a data engine and everything behind it is an incredibly tough problem”. Instead, going for general intelligence, which although is not perfect at performing one given task, has the factor of generality to use natural language prompts and perform tasks across different domains. He added, “Fundamentally, building a robot is not is not different from building a car.”
However, Chris Anderson, CTO of Kitty Hawk, has a conflicting view on this. He argues that “robotics” is too broad a term to be used, essentially calling it a “category error”. The attempt towards creating universal robotic solutions, he says, comes with too much complexity, and rarely is the best solution for any one specific real-world problem. Therefore, for him, cars cannot be approximated to drones, which in turn cannot be approximated to industrial arms or vacuum cleaners.
Anderson’s argument resonates with Gary Marcus, who in writing about the Optimus unveiling, said, “There was no clearly outlined vision for what Optimus would do, nor much justification for why Tesla is building the robot in this specific way.” Marcus believes that underneath the hoopla that Musk created with his claims on what this robot would be able to achieve, there is no clarity over which direction, and specifically for what real-world applications, would the bot be used.
Hence, a vision that could perfectly place why humanoid robots were needed in place of anything else was lacking in the Optimus demo presentation. The lack of value obtained despite continuous investment in the concept of a human-like robot was also seen in Google’s Francois Chollet’s tweet:
Tesla Bot: A demoware?
Moreover, conversation over the Tesla Bot also extends to how realistic is the mission of moving towards AGI. To have a robot that can perform multiple tasks, each consisting of many sub-tasks, will require an enormous amount of training data, which Musk claimed during the event, Tesla will have. However, many critics, including Marcus, said the implied subpremise of Musk’s claim is that “lots of people will buy Tesla’s $20k robot, leading to the collection of a lot of data, but that’s speculative at best, and years away even in an optimistic scenario”.
Marcus, in writing for Scientific American, also makes an interesting point with respect to the current state of AI research. He says that most of the biggest teams of researchers in AI are no longer found in academia where peer review is the judge, but are seen instead in corporations. Here, researchers can evade the peer review process, directly taking their findings or model to a press release. In the software industry, there is a term for this kind of strategy, called “demoware”, which means “a software designed to look good for a demo, but not necessarily good enough for the real world”.
It makes one wonder if the Tesla Bot demo, showing no real-world applications, but built simply because Tesla has the technology to do so, was aimed at boosting its stock. However, Musk was quick to address this, replying to a tweet from an investor who was interested in knowing the long-term global economic implications of Optimus, who if anything didn’t remotely use the words ‘Tesla stock’.
Will Musk keep his promise?
The claim that Tesla will be able to mass produce humanoid robots in a short period of time is similar to the promise he makes every year of achieving full autonomy in their cars “the next year”— it seems to undermine the complexity of the task and the amount of time it will take to fully perfect it.
In addition, generalisation shouldn’t be the culmination of all efforts towards bringing AI-driven solutions. As Niedermeyer says, “automation is a spectrum”. While we don’t have a general-purpose humanoid robot, there are several simple ones adding economic value every day. Going all-in on General AI, whose most vocal advocate has been Elon Musk, shouldn’t take the focus away from designing solutions to specific problems.