Why Are We Obsessed with Building Robots That Look Like Us?

In partnership with

Learn AI in 5 minutes a day

This is the easiest way for a busy person wanting to learn AI in as little time as possible:

1. Sign up for The Rundown AI newsletter

2. They send you 5-minute email updates on the latest AI news and how to use it

3. You learn how to become 2x more productive by leveraging AI

Sign up to start learning.

Twelve years ago, Boston Dynamics unveiled one of the first humanoid robots, sparking widespread excitement—and expectations. The general belief was that, very soon, we’d have robots pouring our coffee, folding our laundry, and perhaps even becoming part of our daily lives. But fast forward to today, and most of these humanoid machines are still stuck in demo mode. Despite billions of dollars poured into robotics startups, the majority of them have yet to deliver on the original promise. It makes you wonder: are humanoid robots more about viral videos than actual productivity?

Human Shape ≠ Human Efficiency

Let’s start with the basics: humans are not very efficient machines. Walking 100 meters, for example, uses about 7 calories—roughly the energy in a bite of an apple. And while that’s not much, the way we walk is mechanically inefficient. We lift, accelerate, and stop each leg individually—about 200 times over that distance. Comparatively, if we could just roll, we’d use about a third of the energy.

Moreover, walking is a remarkably complex process. It requires coordination between the brain, spinal cord, muscles, tendons, and joints. Our inner ear helps us maintain balance, and visual cues constantly adjust our posture. Trying to replicate this delicate dance in a machine is mind-bending. Robots need cameras, LiDAR, force sensors, and massive processing power to simulate this—and even then, their batteries only last a few hours and they can barely lift more than a skinny human.

So Why Build Robots That Look Like Us?

One word: anthropomorphism. We’re naturally inclined to relate more to things that resemble us. The Greeks imagined humanoid automata, the Chinese told stories of human-shaped machines centuries ago, and in modern pop culture, Hollywood has strongly influenced our imagination—think C-3PO and R2-D2.

There's also a psychological factor. Japanese roboticist Masahiro Mori coined the term “uncanny valley” in 1970 to describe how increasing human likeness in robots initially improves our connection to them—but only to a point. As they become too humanlike, our reaction flips from admiration to discomfort and eeriness. This response is deeply rooted in evolutionary psychology—our instincts trigger a form of repulsion when we sense something is almost human, but not quite right.

Staying on the “safe” side of the uncanny valley, or risking crossing it, is a strategic dilemma for robotics companies. But these humanlike designs haven’t exactly proven practical.

The Reality Behind the Viral Clips

Despite the wow factor in videos of robots dancing, vacuuming, or pouring coffee, most of these clips are highly choreographed. Many are shot in controlled environments, edited to hide failures, and involve pre-programmed routines—not real autonomous behavior. Some demonstrations, like Tesla’s Optimus robot, were even shown to be partly human-controlled behind the scenes.

When it comes to real-world utility, almost none of these robots are ready. Boston Dynamics' Atlas, for instance, may have gone viral for backflips and parkour, but it's never left the lab. Their revenue-generating robots are the doglike Spot and the warehouse-oriented Stretch—not humanoids.

Other companies like Figure AI and Agility Robotics have made some progress. Figure AI’s robot briefly worked on a BMW assembly line, and Agility’s Digit moved boxes in a warehouse. But these are still early pilot programs—and ironically, those tasks are already done efficiently by non-humanoid machines.

The Real Problem: Moravec's Paradox

The core challenge in robotics isn’t just mechanical—it’s cognitive. Moravec’s Paradox states that tasks which are easy for humans (like perception and movement) are extremely difficult for machines, while tasks we find difficult (like math or playing chess) are comparatively easy for AI.

For example, solving a basic math problem takes a computer milliseconds—but handing it a shape-sorter toy is an entirely different story. A one-year-old can learn through curiosity, trial, and error. A robot, by contrast, has no curiosity. It has to be explicitly programmed with layers of instructions like “understand game,” “see object,” “grasp object,” and so on—each requiring enormous complexity.

A Costly Obsession

Despite billions in funding, no company has announced a release date for a general-purpose humanoid robot that’s affordable. The goal of producing a robot for under $20,000 remains elusive. In truth, most of these startups rely more on PR and viral demos to drive fundraising and attention than actual product deployment.

So, do we really need humanoid robots? For now, the answer might be no. The tasks where robots are most successful—like moving boxes or vacuuming floors—are already being done more efficiently by purpose-built machines that don’t look anything like us.