Will humanoid robots ever work alongside humans?
Different types of social robots, humanoid robot purposes, then the danger of logistic and manufacturing bots just working on their own
Warning - this is a bit of a longer article in which I try to bring some sense to what robots are currently being built as well as how they will be used alongside humans. My thinking on this is work in progress, and I hope it may help you form your view on humanoid robots too.
So, let’s dive in.
There is a bit of a revolution going on in the creation of humanoid robots, with seemingly every week a new well-funded startup coming out of the woodworks (hello, Mentee Robotics), and predictions of 100m robots produced annually putting it on-par with the entire automotive industry. While some of this is hype and hyperbole, the sheer number of different robots being built by different companies has me convinced that it is just a matter of time until the first really useful humanoid robot will walk amongst us in large numbers. In fact, I would give it a 50% chance that that could be one of the robots currently under development.
Which makes this an interesting moment to reflect on the humanoid robot landscape. There’s understandably a bit of confusion of what these robots actually are, and more importantly, what they are for. For purposes of this article, I want to focus on robots that by design are supposed to interact with people; either because that’s their primary purpose, or because they have to in their line of duty. Let’s call these robots social robots. The term ‘Social Assistive Robot’ would apply to social robots who’s primary function is to assist humans in a social capacity, like the 3CPO protocol droid from the movies, or the Mirokai from Enchanted Tools.
Robots come in many shapes, sizes, and prices. Some are shaped like cute animals meant to be held by people and left on the couch when not used, some are table-top humanoids, and some are fully anthropomorphic and nearly human-sized with the ability to walk about just like you and I do. Let’s call all robots that are anthropomorphic (look like humans), and are at the same scale as us ‘humanoids’. Combining form and function, we can then define ‘social humanoid robots’
For those wondering why anyone would build a robot to look like a human if they don’t need to interact with them, the answer is quite simple: at the moment the world is designed for humans: if you build a robot to be anthropomorphic it can use everything we can: stairs, doors, handles, buttons, and tools designed for human shapes and sizes.
The shape and size of the robot is a critical design choice that is primarily determined by their purpose. That is the tasks that they are supposed to be able to perform, the ways in which they are expected to be of benefit, add value to people’s lives, and thereby allowing companies to charge for their use. Making a robot look and act like a human means that they are actually incredibly general purpose. In my view, the fact that so many of the new robots coming to market have a humanoid form is a strong indicator that the companies designing them don't have a clear idea exactly what their purpose is, and they're hedging their bets by maximising the number of applications possible. That design choice is ultimately a costly one as it prevents optimisation of the robotic platform.
Here are some use cases that human social robotics companies have indicated they are considering:
Logistics - bringing things from one place to another
Infotainment - providing information and entertainment to the user
Manufacturing - actually creating physical things
Security - observing and reporting. Let’s hope not interfering with a security breach.
Quality control - going round a site to check for process adherence and product quality
Health checks - taking health measurements using on-board instruments as well as interacting with the user
Health treatment delivery - delivering e.g. therapy or escorting a patient to their treatment
Companionship - providing a level of companionship to lonely or isolated people
I have written in the past about this curse of creating a robot platform without a clear purpose - it passes the buck to someone else to figure out how to monetise the robots, and I’m afraid we’re again at risk of going down that rabbit hole, only this time the stakes are higher and more money could be lost.
It is clear from the above that while humanoid robots look like humans, not all functions require them to interact with humans. While socially assistive robots obviously do need to, what about robots that have primarily a logistics or manufacturing role? Surely they don't actually need to work shoulder to shoulder with humans, in which case they’re often called co-bots? Proponents of the scenario in which robots work in perfect isolation from humans have recently defended this dystopian view as follows:
Making robots work alongside humans is difficult and thus expensive. It requires sensors, algorithms, and computing power to enable human-robot interaction. It requires not just speech recognition, but full-on language understanding. It requires serious safety features, which are both costly to implement and limit what you can equip the robot with (no huge circular saws for hands, please). It means speeds of action and the amount of information provided have to be limited to that of us humans with our slow bio bodies and brains.
In the absence of humans, you don’t have to design around them. You can be as fast and efficient as the limits of technology will allow you to. You can optimise robots for their task only, introducing specialisation of form and function, and robot to robot interaction. You can even let go of the main driver of humanoid form, the need to design robots for their environment: just design the environment for the robot! Let’s call this dystopia where we have got rid of that pesky human nuisance the bot2bot world.
There’s a real risk here that our current strict safety and AI regulatory constraints actively make this bot2bot dystopia a reality (I'm looking at you here EU AI Act with your abysmal emotion recognition system clause). It might just be easier to make large hangars full of robots and with exactly 0 humans inside. It would be cheaper to do. It would be easier to achieve. And once done, who’s going to reverse that? It’d just be optimised to hell.
This would be a grave mistake in my view. We would throw away what is in my opinion the optimal solution, making the best of the very different capabilities of humans and machines. Robots will never be humans, and what human downsides they offer to improve on, there will always be robot downsides that humans can improve on. We need to collaborate, deliver true team-work. And to achieve that, the human-robot interaction problem needs to be solved, properly.
BLUESKEYE AI will play its role in this, creating the very best human behaviour analysis SDKs ready for integration in any social robot. Humanoid robot producers should play theirs - calculate the benefits of a synergestic relation between humans and machines and then make human-robot interaction a priority in your development. Design for it early on. Get the right sensors and actuators and allocate some compute resource dedicated to human robot interaction. The only real cost is a bit of energy and a bit of hardware - perhaps a couple of hundred dollars extra per robot. The value of the increased output is vastly bigger than that.
Regulators should play their part too: don’t force humanoid producers into the Bot2Bot trap, one that they’ll never get out of and which will cost countless jobs. Make sure safety regulations are fit for purpose. Don’t over-regulate AI just because you don’t understand parts of it. Regulate sector by sector rather than have one AI regulation catch-all.
I will be at the forefront of the NotBot2Bot movement (a movement of one, today). If you want to join me, contact me on LinkedIN and let me show you the art of the possible in human robot interaction today. Live long and prosper, humans and robots together!