While videos of robots performing parkour and backflips dominate social media, industry insiders say the acrobatics are a misleading indicator of progress. Industry executives are Fortune Brainstorm Artificial Intelligence The conference in San Francisco in early December argued that the real revolution in robotics is not physical agility but the ability of robots to “think” for themselves – an ability that ultimately brings them closer to conquering mundane but seemingly difficult tasks, such as opening doors or climbing stairs.
For the past 70 years, robotics has relied on a specific paradigm: intelligent humans using complex mathematics to preprogram machines to perform specific tasks. Sequoia Capital Partners believes this approach is now outdated Jennifer Zhan and Skild AI CEO Deepak Pathak, in conversation with wealthEllie Garfunkel. The industry is undergoing a seismic shift, with bots like the large language models (LLMs) behind tools like ChatGPT learning directly from data and experience rather than following strict code.
“Changes in robotics used to be driven more by human intelligence,” Pathak said, noting that the new wave is defined by models that can generalize and learn. “What’s changed now is these models or these robots can now learn from data.”
In July 2024, Zhan Wei Sequoia’s Blog About Pathak’s deep credentials in the field and what makes him unique as a robotics CEO: his computer vision and deep learning skills. In contrast, traditional robotics focuses on collecting specific data to train robots to perform specific tasks. Pathak and his partner Abhinav Gupta built the underlying model using large-scale data. Chan writes that Pathak was from a small town in India and made national headlines when he was admitted to the Indian Institute of Technology Kanpur without leaving his rural hometown. He learned programming by hand-coding at home and running his programs in limited time at local coffee shops. He later pursued a Ph.D. Obtained a PhD in artificial intelligence from Berkeley, joined the Facebook Artificial Intelligence Research Center, and co-founded Skild.
Jan and Pathak’s conversation with Garfinkel touches on a paradox in artificial intelligence, namely Moravec’s Paradox: Things that look difficult are often easy, and things that look easy are often very difficult.
Why a backflip is easier than a door
A robot performing a backflip essentially requires controlling its body in free space, a physics problem that computers have been adept at solving for decades. “It’s actually much easier to program a robot to do a backflip than to climb stairs,” Garfinkel noted, a point two panelists agreed with.
The real challenge—and the holy grail of “body intelligence”—lies in interacting with the messy real world. Climbing stairs or picking up a glass requires a robot to constantly use vision to correct its movements in response to changing circumstances. This “sensorimotor common sense” is the root of human general intelligence, and it’s the barrier that new “brain” software is trying to break down.
Investors and executives see this as a market opportunity that rivals the recent explosion in generative artificial intelligence. Zhan noted that just as OpenAI opened up the market for digital knowledge work, companies like Pathak’s Skild aim to open up the market for all manual labor. The goal is to create “universal intelligent software” that can act as the brain for any robotic hardware, thereby reducing costs by an order of magnitude.
Unlike the software world, however, robotics faces a unique obstacle: a lack of data. While LL.M.s are trained all over the internet, there is no equivalent database of robot physical interactions. Pathak believes the companies that deploy first will win by creating a “data flywheel” in which on-site robots generate the data needed to make systems smarter.
For consumers wondering when robots will do their laundry, the timeline is still playing out. Pathak and Jan predict that robots will proliferate first in industrial settings and “semi-structured” settings like hotels and hospitals, before moving into the more chaotic environments of private homes.
Despite concerns about job displacement, they believe the technology is necessary to solve the “3S” problems of the future: safety, shortages and social evolution. Robots are about to take over jobs that currently force humans to risk their lives or health. Additionally, with millions of job openings currently unfilled due to labor shortages, robots could fill the gap in essential blue-collar jobs. Ultimately, we hope that society will transform so that hazardous or drudgery jobs become dispensable, allowing people to focus on tasks they enjoy.