AI WILL NOT TAKE OVER THE WORLD – UNLESS WE LET IT
In the movie Transcendence, Johnny Depp plays a University of California at Berkeley AI professor who is shot by anti-AI activists worried about what will happen if scientists succeed in creating AGI systems. I hope I have convinced you by now that AGI is not imminent and that it is about as likely to occur in our lifetimes as time travel or warp speed. There will be no evil robot overlords taking over the world.
That said, narrow AI has such a significant impact that some people would say it is taking over the world. But what they really mean is that AI is taking over just like computers have taken over the world, and the internet has taken over the world, and smartphones have taken over the world, and before that cars, televisions, electricity, and many other technologies took over the world too. So, if you want to argue that AI is one of the many technologies that have taken over the world, I will not disagree.
One difficulty in assessing the impact of AI and determining how to regulate it is that AI is an umbrella term with many different meanings.1 From 1950 to the late 1980s, when most people talked about AI, they were referring to AGI. Techniques such as linear and logistic regression, clustering, and factor analysis had yet to be subsumed by AI, and the researchers still referred to them as statistics. Neural networks had been invented but had not solved any significant engineering problems and had yet to get traction. By 1990, most of the work on AGI had stopped, because the effort required to hand engineer all the knowledge of the world posed an impossible barrier. However, interest in AGI and fear of AGI is now building again because of the success of narrow AI, even though AGI cannot evolve from narrow AI.
Not only does the umbrella term AI include both AGI and narrow AI, but narrow AI is itself an umbrella term as well. It is not simply deep learning plus natural language processing, even though perhaps it should be, since those are the key technologies underpinning all the impressive recent accomplishments of AI. Narrow AI encompasses all machine learning, which includes a great deal of plain old boring statistics. Issues of fairness and discrimination that people attribute to AI are really those issues with statistics (enhanced by computational methods) applied to the massive amounts of data that are now available because of the pervasiveness of the internet.
Because AI is such an umbrella term, it would be a mistake to treat all the different social issues attributed to AI in a unitary fashion. If governments try to boil the ocean by attempting to regulate all these different categories of issues under one set of regulations, it will hamper effective regulation, and we may end up with unnecessary regulations that stifle progress. It might even lead to a ban on AI research.
Aside from AGI being an improbable occurrence, a century’s worth of science fiction writers have shown us that AGI regulation would be a difficult, if not impossible, task. Contemporary researchers writing on this topic have many different ideas of how an AGI future would unfold. However, the common thread is that AGI robots would be able to think and reason. As a result, regulation would need to target this general capability rather than narrow AI’s specific capabilities, such as the ability to drive a car. Science fiction author Isaac Asimov exemplified this type of general regulation with his three laws of robotics.2 Fortunately, we do not need to regulate AGI, because it does not exist and most likely never will.
Creating effective regulations for narrow AI is a far greater imperative. Narrow AI issues are complex and multilayered and impact several areas of our lives. In contrast to AGI, which would require a generic set of regulations, we need to tackle AI issues with targeted regulations and policies. If we do so, we will reap the many positive benefits of narrow AI and minimize the negative impacts. Narrow AI systems will not take over the world unless we let them. AGI systems will not take over the world period. And there is no reason to shoot Johnny Depp.