Humanoid Robots are the Wrong Answer to the Right Problem
We all have a million things in our lives that we’d rather not have to deal with. From keeping up with emails, to countless chores around the house, there are plenty of reasons to dream of a world in which robots do the mundane for us. AI and robots are helping businesses to scale productivity across every imaginable industry and consumers are just as eager to eliminate the drudgery of chores in the home.
This desire to automate is driving massive investments in robotics, where even companies better known for their social, search, and ecommerce products, are getting into the field now. So when are we going to see a wave of robots coming to the rescue? And what will they look like?
(Disclosure: I work on the Robotics at Google team, supporting some of the great research you can see here. The views in the article are my own.)
To start, I’m very suspect of us ever achieving general purpose robots, where a single machine will perform any work that a human can. But let’s assume a robotics team somewhere has been given the goal to precisely automate the specific tasks that people do every day.
In the drawing below, it’s easy to imagine the work needing to be done in every room of this house. We’ll want to clean the bedroom window, the office requires dusting, the bathroom gets scrubbed, and food is prepared in the kitchen. The car too, has become an important part of our lives and needs maintenance.
Now let’s imagine that robotics team creating the one-size-fits-all solution to tackle this work for us. A robot designed so that it could even use the same tools and cleaning products that a person would.
Thanks to decades of Sci-Fi in the media, many people might immediately jump to the image of a large Rosie-like robot, as something right out of The Jetsons. There’s definitely been a long fascination with humanoid robots that seems to persist in the industry. At the very least we often see some kind of anthropomorphism.
Human shaped for a human world?
The rationale that I’ve heard many roboticists give, is that the modern world was designed around humans, and so we need human-shaped robots to interact with that world. At the surface, you certainly can’t fault that thinking as our products, tools, and infrastructure are indeed built around the typical human form and capability.
So maybe our future would have something like this, where a semi-humanoid robot comes into our home and takes over for the chores?
If they pick up the mop and broom, maybe they can also just hop into our car and drive it too. The goal of general purpose robots is of course to make them broadly useful across tasks. I’d certainly want my expensive home robot to work for me in as many ways as possible using the same equipment that I do, including my car.
If that argument were true, clearly we should be building humanoid chauffeur robots to jump in and drive the 1 BILLION vehicles out there on the road today. It would be absurd to replace the massive install base of human-operated cars and trucks, and we should build robots shaped like us to drive these vehicles while we hop in the back. The Chauffeurbots could even get out and fill the gas, plug in the charger, or perform vehicle maintenance.
Yet, that’s not what’s happening at all. There are billions of dollars being poured into autonomous car technology across hundreds of companies tackling the challenge. Not one of them is attempting to literally put a robot behind the wheel beyond automating tests tracks. Every autonomy company out there, even those led by roboticists dreaming of a future full of humanoids, is currently working on fully integrated self-driving cars instead.
Purpose built for maximum efficiency
The truth is that fixed function appliances dominate our lives when it comes to physical manipulation. Each product perfectly built to accomplish a fixed scope of work without compromise. We’ll likely see more of this as our kitchens, bathrooms, and closets all become more automated in the years ahead.
Want to fold clothes quickly? Which of these robots is doing the task better:
Do you really want your robot to clean the toilet and then go make you breakfast? I’d rather have a fully automated bathroom, appliances that take care of the cooking, and an integrated laundry stowage and washing system.
We can also take some cues from film and media. The lovable cleaning crew of the Axiom spaceship in Disney’s Wall-E, is full of purpose-built robots to tackle various chores on board. None of them grab nearby mops and brooms because we can assume humans have long forgotten how to use such things.
Humanoids are just plain creepy and immoral
When things go wrong with robots in Sci-Fi, they almost always take human form. Beyond the “robots out of control” trope, there’s the psychological factors that come to play as we can’t decide how to treat them, or how they should treat us. Just watch the first season of the TV show Humans to see a very plausible future where robots take on full human appearance.
Spoiler alert: it does not go well for the humans welcoming these robots into their homes. Surprise.
There’s a massive moral issue that comes with enslaving a humanoid that I just can’t get past. There’s no better example of this than the “Dissonance Theory” episode of Westworld, where robots are literally out on the plantation as slaves of the past. Is that really the right form factor for the job and the world we want to live in?
There is however one small outlier use case where humanoid robots are in fact the right form factor and moral choice: stunt and crash test dummies.
Disney, which I admit to having far too much influence in my thinking, seems to get it right again in this example of a humanoid robot standing in for a person performing stunts. It can pull off superhuman capabilities without harming real people, while trying to bring joy to use all through film and TV.
Superhuman or subhuman, but never human
In most cases, automating a given task can be done with something far simpler than a full human form. An electric pencil sharpener sure does a better job than a two-armed robot would trying to use a knife to whittle the wood.
For the cases where high dexterity is required, we have to think about whether there’s another form factor that can do the task better than a 7 DOF (Degree of Freedom) human arm and hand. If you need to maximize for flexibility and dexterity, maybe try an octopus inspired design instead.
I just don’t buy into the argument that we need human-shaped robots using our tools and coming with all of the compromises that our bodies have. We’re amazing creatures, but when it comes to the future of automation, we’re better off designing the right tool for the job and giving it the smarts to run by itself.