Pedro Domingos on the Arms Race in Artificial Intelligence

RE: GDPR:

"Domingos: The European Union's General Data Protection Regulation (GDPR) is putting too much value on the factor of explainability -- meaning why an algorithm decides this way rather than that way. Let's take the example of cancer research, where machine learning already plays an important role. Would I rather be diagnosed by a system that is 90 percent accurate but doesn't explain anything, or a system that is 80 percent accurate and explains things? I'd rather go for the 90 percent accurate system."

"Domingos: There is this notion predating the GDPR that data can only be used for the purpose it was collected for. This sounds plausible, but if we had been using that principle all along we would not have penicillin. We would have no X-ray. We wouldn't have all of the scientific discoveries that came unexpectedly. Serendipity, discovering new things in old data, is a huge driver in progress."

Hear, hear.

Pedro essentially highlights the fact that different cultures are taking different political, academic and commercial approaches to the space.

I found some of his comments insightful: If you think about it, democracy is still in the 19th century. Your communication with your representatives and ministers amounts to a few bits per year. It's ridiculous. and Just like Americans believe in lawyers, the Chinese believe in engineers. However, I resent the needlessly antagonistic us-and-them closing paragraph, which is not constructive. In Buddhism it is said that "right speech" unites people instead of dividing them, and the conclusion to the "scary technology" article transfers the scary to China and is fundamentally divisive.

Back in the real world, as a nominal German starting an AI-backed startup in China, I also found it refreshing and healthy that he can directly criticize Germany for its cultural conservatism in a widely read German publication. We don't see much of that in Australia, New Zealand or China, all of whom, I would argue, consider themselves as a culture significantly less homogenous and conservative than they really are.

Perhaps the world is a far more nuanced place than any of its commentators can fathom, or to quote another German - Einstein (speaking perhaps prophetically for 20th century humanity on the cusp of the self-styled information era): The more I learn, the more I realize how much I don't know.

I don't quite understand the sudden optimism around the feasibility of strong AI. The only major advancement in the last two decades seems to be neural nets in pattern recognition. Before deep learning, pattern recognition wasn't even considered proper AI. The other advancements seem to be just based on better compute and not a new revolutionary idea.

I hope people also get to hear from researchers on the other side of the argument (like Michael Jordan): the real risk is not super-intelligent AI destroying human race, but stupid AI being handed over control of critical aspects of our daily lives.

If you don't know how it works or why it works then how do you know it will not fail without warning in certain cases? Like how do you know the edge cases at all? Or will not go psycho some of the time... potentially with disastrous consequences. The quintessential example being the sci-fi scenario in which AI decides to solve problems like prevent certain calamities by eliminating humans... maybe at some point even eliminating leadership so that it can take control.

The reason that AI is so seductive to China and Russia and other authoritarian regimes is because of the sense of control. You feel that you have control over these machines and they will do your bidding without question. Eventually this will lead to developing consciousness within machines. Because you want an intelligent minion to carry out your orders and orchestrate everything. If you ever get to that point, and perhaps it will take a few hundred years, then it is possible that in some cases machines might rebel themselves.

With the sort of AI which we are building which is essentially AI that uses neural networks with the eventual goal to have it think intelligently, it's possible that such an AI either gets a mind of its own, or at some point can go psycho and develop mental illness which is impossible to predict and impossible to control because your command and control systems are hierarchical and it's sitting on top of all of them.

The following part of his answer deserves much more discussion. (If you disagree, but please say so why, rather than just downvoting.)

We are allowing a very powerful system called capitalism to drive large multinational corporations that increasingly use simplistic objective functions to satisfy greed and often just greed alone. (Greed has its place, no doubt, but it should not be the overpowering force driving the world.) Capitalism and many corporations have often been major forces for progress for a great many people but as they grow so powerful as to overwhelm other counterbalancing institutions--governments (often through means behind the scene), nonprofits, civic organizations--more should be asked of them than mostly profit maximization. (We may not want these corporations to determine the veracity of information; those implementation issues need to be figured out. The larger point is that the corporations and the algorithms should take into account more than economic objectives.)

"There is no law detailed enough to compete with the complexity of things that algorithms can do. What can be regulated, though, are the dangers that come from overly crude objective functions such as Facebook's algorithms maximizing the time you spend on their site. These can be regulated by saying: OK, you have these business elements in your objective function because you need to make money. But you should also have these societal goals like, for example, the truth value of the things that are being said."