Optimus might not be prime as experts claim humanoid robots are many years from perfection
Published on Sep 06, 2025 at 4:20 AM (UTC+4)
by Claire Reid
Last updated on Sep 02, 2025 at 4:52 PM (UTC+4)
Edited by
Kate Bain
Robotics expert Ken Goldberg claims that truly capable humanoid robots are a long way off.
The UC Berkeley roboticist said that despite the hype, people have no reason to fear a humanoid robot takeover, at least for now.
He said that, unlike AI language models (LLMs) that are advancing rapidly, this won’t be the case for humanoid robots.
He said there was a ‘100,000-year data gap’ that stops robots from gaining real-world skills as quickly as AI chatbots are gaining language fluency.
DISCOVER SBX CARS – The global premium auction platform powered by Supercar Blondie
A robotics expert has said truly capable humanoid robots are years away
There have been some huge advancements when it comes to robots in the last few years, from Tesla’s Optimus bot that the company says has been put to work in its offices, to this headless robot ‘dog’ that can outrun a human.
However, robotics expert Ken Goldberg says it will be years before we have truly capable humanoid robots that are able to mimic tasks and jobs carried out by humans.
Goldberg published two papers in the journal Science Robotics, in which he said that there has been a bit of ‘hype’ around what robots will be able to do and how quickly.
The main issue, according to Goldberg, is dexterity and being able to manipulate objects.
“Things like being able to pick up a wine glass or change a light bulb,” Goldberg told UC Berkley News. “No robot can do that.”
He went on to explain that because we have seen AI systems carry out tasks that appear more complex, like playing chess, people think a robot should be able to pick up a glass, which appears to be a lot more straightforward.
But even the simple act of picking up a glass or changing a lightbulb is a lot more involved than you might think and is ‘extremely difficult’ for robots to be trained to do.
So, if nothing else, wine tasters and electricians should have nothing to worry about for a while.
The 100,000-year data gap that’s impacting training robots
In the paper, Goldberg used the term ‘the 100,000-year data gap to help try and explain the difference between the language capabilities of AI compared to dexterity in humanoid robots.
To get to that figure, Goldberg said he looked at how much data exists on the internet and then worked out how long it would take a person to read it all: about 100,000 years.
That 100,000 years of text is what is used to train AI LLMs.
But when it comes to training robots, there is nowhere near as much data.
He also suggested that training robots to do real-life skills would likely require even more data than the 100,000 years of text used to train AI chatbots.
The expert said that while he does think that truly skilled and capable humanoid robots could exist in the future, it’s not going to be soon.
“I’m not saying it’s not going to happen, but I’m saying it’s not going to happen in the next two years, or five years or even 10 years,” he said.
DISCOVER SBX CARS: The global premium car auction platform powered by Supercar Blondie
Claire Reid is a journalist who hails from the UK but is now living in New Zealand. She began her career after graduating with a degree in Journalism from Liverpool John Moore’s University and has more than a decade of experience, writing for both local newspapers and national news sites. Claire covers a wide variety of topics, with a special focus on cars, technology, planes, cryptocurrency, and luxury.
