Article illustration 1

Just a few years ago, humanoid robots were synonymous with clunky, unreliable machines, epitomized by viral fails like Boston Dynamics' Atlas stumbling during early trials. But the landscape is shifting rapidly. According to exclusive reporting by WIRED senior writer Will Knight, OpenAI is doubling down on robotics—specifically humanoids—by recruiting top researchers in the field. This marks a dramatic resurgence after the company shuttered its robotics division in 2021, and it underscores a pivotal moment in AI: the race to artificial general intelligence (AGI) may hinge on mastering the physical world, not just digital conversations.

Why Humanoids? The Path to Embodied Intelligence

Humanoid robots, designed to mimic human form and function, have long been dismissed as gimmicks. Yet, as Knight explains, advances in hardware—like more responsive motors and actuators—have transformed them from comedic novelties into viable platforms. The real game-changer, however, is AI. Large language models (LLMs) like GPT-4 excel at abstract reasoning but lack real-world grounding. "ChatGPT can solve complex math problems, but it can't make you a cup of coffee," Knight noted in a recent episode of WIRED's Uncanny Valley podcast. Humanoids bridge this gap by forcing AI to learn through physical interaction, potentially creating "world models" that understand cause and effect in environments built for humans—from factory floors to living rooms.

OpenAI's renewed focus isn't happening in a vacuum. The company is hiring specialists to tackle core challenges like robotic manipulation and locomotion, with job listings hinting at ambitious projects beyond their earlier work, such as training algorithms to solve a Rubik's cube with a robotic hand. This pivot follows the underwhelming reception of GPT-5, suggesting OpenAI sees embodied AI as the next frontier. As Knight reports, "Developing a better understanding of the physical world would be crucial for consumer hardware, like smart glasses or whatever they're building next."

The Competitive Rush: From Factories to Hype Cycles

OpenAI joins a crowded field of players betting big on humanoids. Tesla, with its Optimus bot, leverages real-world data from self-driving cars, while startups like Figure AI and Agility Robotics showcase robots performing tasks like moving boxes in Amazon warehouses. Chinese firm Unitree, now filing for an IPO, has demonstrated low-cost humanoids in boxing matches—a far cry from the cautious, teleoperated demos of the past. Industry projections are sky-high: the humanoid robotics market, valued at $5 billion today, could explode to $5 trillion by 2050.

But as Knight cautions, the hype often outpaces reality. Demo videos of robots folding laundry or feeding pets are frequently staged or unreliable, masking the immense difficulty of operating in unstructured environments. "Getting these things to be super reliable is an enormous challenge," he said, drawing parallels to the slow, fraught evolution of self-driving cars. Initial applications will likely be narrow and industrial—Boston Dynamics' electric Atlas robot is set for Hyundai factory trials this year—rather than the sci-fi vision of home butlers. As Knight put it: "Manufacturing and ecommerce are where you'll see them first, not your kitchen."

Challenges and Ethical Quicksand

The road to practical humanoids is riddled with obstacles. Hardware limitations persist; current robotic hands can't match human dexterity, and AI models struggle with the "hallucinations" that, in physical systems, could lead to dangerous failures. Training data is another hurdle—unlike the vast text corpora fueling LLMs, there's no equivalent dataset for real-world actions. And as Kylie Robison, WIRED's senior AI correspondent, highlighted, military applications loom large. During a tense exchange with Tucker Carlson, OpenAI CEO Sam Altman downplayed involvement in defense, but the potential for weaponized humanoids raises urgent ethical questions.

Despite the challenges, the momentum is undeniable. If successful, humanoids could revolutionize labor-strapped industries like logistics and manufacturing, with Amazon already testing Agility's robots for repetitive tasks. But as this race heats up, the biggest takeaway is philosophical: AGI might not emerge from code alone. As Knight reflects, "Building something with human-like intelligence requires it to do things in the real world." For now, though, the robots remain firmly tethered—sometimes literally, as Robison wryly noted, to meat hooks.

Source: WIRED's reporting and podcast episode "Move Aside Chatbots, AI Humanoids Are Here," featuring Will Knight, Michael Calore, and Kylie Robison.