by Skeptical Scalpel
An attempt to teach common sense to computers is underway at the Allen Institute for Artificial Intelligence [AI2]. Microsoft billionaire Paul Allen has donated $125 million to finance the effort. “Despite the recent AI successes, common sense—which is trivially easy for people—is remarkably difficult for AI,” said Oren Etzioni, CEO of AI2.
Oren, you’ve been working with computers too long. Common sense is not universally “trivially easy for people.” In fact for some, it is difficult to impossible.
The other day I asked Twitter if common sense could be taught. The results of that admittedly unscientific poll are as follows:
So almost two thirds of Twitter users who voted think common sense cannot be taught. I’m with them.
What is common sense anyway? Here are some definitions. Take your pick.
– The ability to use good judgment in making decisions and to live in a reasonable and safe way
– Sound judgment derived from experience rather than study
– Sound and prudent judgment based on a simple perception of the situation or facts
– Natural ability to make good judgments and to behave in a practical and sensible way
– Sound practical judgment that is independent of specialized knowledge, training, or the like; normal native intelligence
– The wisdom common to all human beings
My Twitter friend @michelaccad says, “Some may be born with a defect in common sense, the same way some people may be born with a defective eye or a defective limb.”
Although I’ve never seen it happen, I believe a few of those born without it may be able to acquire it based on experience.
The US Defense Department is also interested in artificial intelligence and common sense. According to a recent article in Financial Times, a Defense Advanced Research Projects Agency [DARPA] manager said they want systems that “don’t drive off a cliff and have the sense to come in out of the rain” and can adapt to unexpected events.
DARPA hopes to achieve its goal by using more available data, which I’m not sure is the answer, and by crowdsourcing from Internet users. Let’s hope they don’t tap into the “comments” sections of posts or Twitter flame wars.
The idea of imbuing computers with common sense is not new. Stanford computer scientist John McCarthy, considered the father of artificial intelligence, published a paper about this in 1959. He suggested a program which he called advice taker. He wanted it to learn from experience and improve its behavior as effectively as humans do “by making statements to it, telling it about its symbolic environment and what is wanted from it.”
We haven’t progressed much. Some 59 years later, another AI researcher, Yejin Choi, had this to say about deep learning systems, “They don’t generalize well across different topics and aren’t robust in unforeseen situations.”
Maybe our common sense will help us hang on to our jobs for a few more years.
Skeptical Scalpel is a retired surgeon and was a surgical department chairman and residency program director for many years. He is board-certified in general surgery and a surgical sub-specialty and has re-certified in both several times. For the last 8 years, he has been blogging at SkepticalScalpel.blogspot.com and tweeting as @SkepticScalpel. His blog has had more than 3,000,000 page views, and he has over 18,000 followers on Twitter.