The Illusion of Robotic Progress: Why a Backflip Doesn’t Mean Intelligence
Despite viral videos showcasing impressive feats, experts caution against overestimating the current state of humanoid robots. The dazzling displays of robotic athleticism often mask a fundamental struggle with everyday tasks, raising questions about the true path to artificial general intelligence.
The hype surrounding robotics reached a new level recently, with footage circulating of robots running, performing backflips, dancing, and even kickboxing. However, speakers at the Fortune Brainstorm AI conference on Tuesday warned against being misled by these acrobatic demonstrations. As one attendee explained, “What looks hard is easy, but what looks easy is really hard,” echoing a long-held observation from computer scientist Hans Moravec dating back to the late 1980s. Moravec and his contemporaries discovered that computers excelled at complex calculations but faltered at tasks easily mastered by young children.
The Gap Between Spectacle and Substance
The core issue, according to Deepak Pathak, CEO of robotics startup Skild AI, lies in the environment. Robots currently demonstrate proficiency in controlled environments, where parameters are predictable. A video showcased a Skild robot skipping down a sidewalk, but Pathak emphasized that the robot was “apart from the ground, not interacting with anything.”
The challenge arises when robots attempt tasks requiring real-world interaction. Picking up a bottle or navigating stairs demands constant visual correction – a process deeply ingrained in human general intelligence. “That interaction is the root reason for human general intelligence, which you don’t appreciate because almost every human has it,” Pathak stated.
A partner at Sequoia Capital further cautioned that viral videos often lack transparency. “They don’t show how the product was trained, nor whether it can operate in an uncontrolled environment,” she said. “The challenge for you as a consumer of all these videos is to really discern what’s real and what’s not.”
From Programming to Learning: The Next Evolution
Despite these limitations, optimism remains regarding the future of robotics. The field is undergoing a significant shift, moving away from pre-programmed instructions toward machine learning. “Robots used to be driven more by human intelligence,” Pathak explained. “Somebody super smart would look at [a task], and…pre-program the robot mathematically to do it.” Now, the focus is on enabling robots to “learn from experience,” allowing them to adapt to more complex and unpredictable situations without constant reprogramming.
This transition promises to unlock new possibilities. Current robotics firms are “still constrained by having robots that are only built for specific things,” but a platform with greater general intelligence could tackle tasks currently too dangerous for human workers. Consumers could also benefit, moving beyond single-function household robots to systems capable of handling a wider range of chores. Qualcomm CEO Rene Haas previously noted at Brainstorm AI that the adaptability of humanoid robots will likely make them superior to traditional robotic arms in factory settings.
Navigating the Social and Economic Impact
The rise of advanced robotics inevitably raises concerns about job displacement. However, Pathak offered a sanguine perspective, highlighting potential benefits such as increased safety by removing humans from hazardous jobs and addressing the growing labor shortage in blue-collar and manufacturing sectors – a key obstacle to re-shoring advanced manufacturing to the U.S.
Ultimately, Pathak envisions a future where automation liberates humans from mundane work. “There lies a scenario, a good scenario, where everybody is doing things that they like,” he said. “Work is more optional, and they are doing things that they enjoy.” However, he acknowledged the need for societal strategies to ensure the equitable distribution of the benefits derived from increased automation.
