September 1, 2026
A new study covered by Robohub, led by researchers at Drexel University, the U.S. Air Force Academy, and George Mason University, used brain imaging to show something that should worry anyone building a customer-facing humanoid: when an expressive, conversational robot makes a mistake, people don't just shrug it off -- they become measurably more suspicious of it than they would of a plainer, less chatty machine that made the same error. The more human the robot seems, the more it gets held to a human standard, and the harder it falls when it misses that bar.
This is not a trivial UX footnote. It's a warning about sequencing. Companies racing to make humanoid robots feel warm and relatable -- expressive faces, conversational banter, humanlike gestures -- are making a bet that likability buys forgiveness. The brain data suggests the opposite: likability raises the stakes. A robot that seems like a peer gets judged like one, and a fumbled handoff or a misheard instruction reads as a character flaw rather than a bug. We flagged this exact tension a few days ago when we wrote about why robots that look human have to act human, too -- this new research gives that argument a neuroscience backing it didn't have before.
For a business evaluating a humanoid robot for a warehouse floor, a hotel lobby, or a retail counter, the practical takeaway is blunt: don't buy expressiveness you can't back with reliability. If a robot's personality outpaces its competence, you're not building trust -- you're setting a trap that springs the first time it stalls or misreads a request. The safer path, at least for now, is a robot whose behavior matches its actual capability, even if that makes it less charming in a demo video.
Meanwhile, New Atlas reported on something almost comically different: Pollen Robotics and Hugging Face's Microduck, a small waddling robot that can pick things up and learn new tricks, built explicitly as an open, hackable platform for people learning robotics rather than a polished commercial product. It's not trying to pass as human. It's not managing anyone's expectations of empathy or conversation. It's a tool for tinkering, and that's precisely why it's interesting.
Put next to the trust study, Microduck makes an accidental argument: maybe the fastest way to build real capability in legged, manipulating robots isn't to chase humanlike expressiveness at all, but to get an open, low-stakes platform into as many hands as possible and let the skills -- grasping, balance, learning new tasks -- improve through sheer volume of experimentation. A duck that occasionally fails to pick something up doesn't trigger suspicion; it triggers a shrug and a firmware update. That's a much cheaper feedback loop than the one facing a six-foot conversational humanoid deployed in front of paying customers.
For business leaders, the lesson from pairing these two stories is about where to place your risk. If you're evaluating automation for a role that involves any human-facing interaction, the expressiveness of the robot is not a feature to maximize -- it's a liability to manage carefully, and the research increasingly agrees. If you're evaluating automation for back-of-house, repetitive manipulation tasks, boring and reliable platforms -- duck-shaped or otherwise -- are probably a better bet than something built to seem likable. We've made a version of this same point about automation generally: the goal isn't to impress people, it's to pay off in ways you can measure, and a robot's charisma has never once shown up on a P&L.
None of this means expressive humanoids are a dead end -- there's a real argument that as reliability improves, the warmth becomes an asset rather than a risk, and companies that get the sequencing right could earn real loyalty. But right now, the order matters: competence first, personality second. Which of your customer-facing processes would actually benefit from a robot with a personality, and which ones would you rather hand to something as unglamorous and dependable as a duck?
Sources