The big news this week was STAR, a robot that outperformed surgeons. Initially, the robots were designed to look and work like humans. This was the image from popular comics when I was growing up. However, now that robots have become mainstream, most robots are very unlike humans, and that is their strength. Let us explore this thought.
Are robots mainstream? Well that depends on what you see as a machine with artificial intelligence. Every one of you reading this blog has used an ATM machine to withdraw cash. And, this is how robots differ from the comic book concept. The best robots do one thing rather well. They are single purpose machines like the ubiquitous robotic arms in factories.
Industrial robots have diffused rapidly in countries facing declining working age population and young people who do not wish for a factory career. In 2013, China became the leading buyer of industrial robots. By 2016, China will have the largest installed base of industrial robots in the world. Yet, China has only 36 robots per 10,000 manufacturing employees compared to Germany with 292, Japan with 314, and South Korea with 487 robots per 10,000 manufacturing workers. These industrial robots of course look nothing like human beings.
Artificial intelligence has now moved the focus from industrial robots working on preprogrammed repetitive tasks in enclosed spaces to robots that can adapt to changing conditions (i.e., learn) and interact with humans.
Just as industrial robots don’t look like humans, robots with artificial intelligence don’t think like humans. The original effort to build a machine that could beat a chess player began by interviewing grandmasters in the hope of having them articulate how they play and then fusing that logic into the computer. Well, as we are well aware in marketing, consumers are clueless when it comes to telling you why they chose a certain brand. Similarly, experts were useless when asked to describe how they played chess so well. You got the same vacuous answers that you get from managers on what makes them successful – intuition and experience!
The breakthrough victory of Deep Blue, IBM’s supercomputer over the then world champion, Garry Kasparov, was based on sheer processing power combined with massive data storage capability. Similarly, Google’s AlphaGo recently beat the champion Lee Sedol in four out of five games of Go without mimicking the ways champions play. Instead, these victories were achieved by approaching the task differently than humans through what is now called deep learning. These are machine learning algorithms that rely on endless trial and error method to improve their performance on a task.
While Japan and Germany dominated the industrial robotic industry, China and USA are racing to dominate the deep learning robotic space. China does lead in the patents related to robots with 35% of all the patents filed in 2015. This is double that of the next highest country, Japan. Clearly, China has hardware capabilities second to none. However, USA because of its software expertise is the real leader as robots become increasingly more software than hardware in character driven by deep learning.
The victories of Deep Blue and AlphaGo, as impressive as they are, were still on tasks with defined rules requiring processing power. The challenge that remained was to succeed on tasks that require a combination of deep expertise and manual dexterity. Surgery seems to be just such a problem. Which is why STAR is getting so much attention. When tested on piglets (pigs apparently have an internal system which is close to humans), with minimal human intervention, the robots were found to have done more evenly spaced stitches and resulted in less leaky sutured gut. The researchers now claim to have managed to make STAR work autonomously.
The advantages of robots over humans are easy to see. It reduces variance in the performance of the task (greater precision as no mood swings or distractions), ability to work in hazardous conditions, and most of all, without making the many demands that humans do such as wages and breaks. Still, as robots increasingly interact with humans, think of three situations:
- You are flying on an airplane with no pilot, essentially a drone
- You are in an autonomous vehicle, where there are no controls to manipulate
- You are under surgery and being operated by a robot
In these situations, will you trust robots? And, if something goes wrong, who is accountable?
It brings to the forefront the special domain still considered human – creativity, feelings, empathy, and the ability to exercise judgment. People would not feel comfortable, for example, a programmed car to make the decision on whether it should let the impeding vehicle go unharmed since the latter has four occupants, while allowing the death of the lone rider in the autonomous car (assuming it is one or the other). Keeping humans in the loop makes robots more socially acceptable. Which is why we would insist on having a pilot on the plane, a surgeon in the operating theatre, and the ability for the occupant to over-ride the autonomous car.
Speaking of feelings, people can fall in love with robots. When asked if it is better than falling in love with a person, an insightful researcher replied no, but what if you don’t have that option. This leads me to end with the story from 1963 when Luther Simjian filed a patent for an ATM machine that allowed deposits. After failing the pilot test in New York and being discontinued, he remarked:
The only people using the machine were prostitutes and gamblers who didn’t want to deal with tellers face to face.