Stay informed with free updates
Just enter the Artificial intelligence myFT Digest — delivered straight to your inbox.
Pepper the humanoid robot was born in 2014. visit the Financial Times to meet the editor. “This is a robot that behaves autonomously, driven by love,” said Masayoshi Son, head of its main sponsor, SoftBank. Alibaba and Foxconn have invested hundreds of millions in an effort to make robots a part of everyday life. But it was not to be. You still find Pepper from time to time in a public library in Japan, stripped, head bowed, like a four-foot Pinocchio who dreamed of being a real boy but never was. Production ceased in 2021 and only 27,000 units were ever made.
Yet the idea of humanoid robots—machines like us that can do all the work we don't want to do—is too attractive to leave for long. Recent, surprising advances in artificial intelligence have inspired a new wave interest in robotics. “The next wave of AI is physical AI. AI that understands the laws of physics, AI that can work among us,” said Jensen Huang, chief executive of chip designer Nvidia, earlier this year. Nvidia is riding the boom in training AI models to become the second largest company in the world. world by making money in the market.
Billions of dollars in venture capital are pouring into robotics startups. They intend to use the same kind of training methods that allow computers to predict how a protein will fold or generate a surprisingly realistic text. They aim, firstly, to allow robots to understand what they see in the virtual world, and secondly, to interact with it naturally, to solve large-scale programming tasks consisting of simple actions such as picking up and manipulating an object.
Such is the dream. The latest round of investors and entrepreneurs, however, may end up as disappointed as those who backed Pepper. It's not because AI isn't helpful. Rather, it's because the barriers to making an economical robot that can cook dinner and clean toilets are a matter of hardware, not just software, and AI doesn't solve them, not solve them.
These physical challenges are many and difficult. For example, a human arm or leg is moved by muscles, while a robot limb must be moved by motors. Each axis of motion along which the legs must move requires multiple motors. All of this is possible, as the robotic arms in factories show, but the high-performance motors, gears and transmissions involved create bulk, cost, energy requirements and many parts that can break.
After creating the desired movement, there is a challenge to hear and respond. If you pick up a piece of fruit, for example, the human nerves in your hand will tell you how soft it is and how hard it is to squeeze. You can taste that the food is cooked and smell that it is burning. None of those sensors are easy to provide for a robot, and to the extent possible, they add more cost. Machine vision and AI can pay off, by checking if the fruit is smooth or the food in the pan has gone the right color, but they are imperfect substitutes.
Then there is the issue of power. Any independent machine needs its own power source. Industrial robot arms are connected to main pipes. They can't walk around. A humanoid robot can use a battery, but then there are trade-offs in mass, power, energy, flexibility, uptime, useful life and cost. These are just some of the problems. Many intelligent people try to solve themselves and make progress. But the point is that these are physical challenges, long lasting and difficult. Even the revolution in AI doesn't make them go away.
What, then, makes AI possible in the physical world? Instead of thinking about how technology will enable new machines, it's more practical to think about how existing machines will change after AI is applied to them.
An obvious example is self-driving cars. In this case, the machine does not need to change at all: the movement of the car in the virtual world and its power source will work as they always are, while the feeling involved in driving the car is almost completely invisible. With the new vogue of AI, the hype cycle for autonomous vehicles is over. In fact it should be the opposite: self-driving is a big market and a real-world challenge that AI can easily tackle, a point that anyone who is tempted to invest in other applications in robotics should consider.
It is also reasonable to think about how existing robots – from industrial robotic arms to vacuum cleaning robots – will evolve. AI-powered machine vision will subtly expand the range of tasks a robotic arm can perform and make it safer for them to work alongside humans. Lightweight, single-purpose tools such as robotic cleaners will be more useful. In Chinese hotels, for example, it has become common for a robot to bring you to your room. That kind of limited and controlled autonomy is the most easily delivered.
In this way, AI will slowly bring us closer to androids. As for a robot like Pepper that can clean the toilet – sadly it's too easy to make someone write bad poetry, and that's unlikely to change anytime soon.