Congratulations to @TheEconomist and @tcross81 on a fantastic survey on #AI and its limitations this week: https://www.economist.com/technology-quarterly/2020/06/11/an-understanding-of-ais-limitations-is-starting-to-sink-in. There is a fundamental issue/irrationality about the whole AI way of thinking about some very good underlying technologies (such as deep nets)

that is holding back the industry, creating unnecessary paranoia and all sorts of harms and injustices. Because the folks who tend to be enthusiasts for this technology have an extreme rationalistic/economistic way of seeing things, let me here eschew my usual complaints

and try to put this in a very economistic mode. AI is a dumb, inefficient and economically irrational direction of thinking about the future of technology. It is based around stories like Alpha Go and the Turing Test that are at their core the OPPOSITE of how technology

should be designed. As AI enthusiast @halvarian once put it (referring to humans), you always want to be scarce and complementary to something abundant. Human intelligence is abundant, and AI people have been predicting for years that it will be made even more so

by being "replaced" by AI. Therefore any smart technologist actually interested in improving human life effectually should be trying to design tech that a) does something people want and b) complements rather than substitutes for what human intelligence is good at.

Consider Alpha Go. Who, exactly, wants a super Go player? And how, exactly, does that system complement human intelligence? And the Turing Test: the whole idea here is to be indistinguishable from a human. That is an incredibly unhelpful goal, given that a) it is insanely

hard and b) WE ALREADY HAVE 9 BILLION HUMANS ON THIS PLANET. The last thing we need is to make machines that are indistinguishable from those humans. Why not build machines that can do things humans cannot do and make humans more productive?

For example, rather than trying to substitute for human recruiters and in the process replicating or exacerbating human racial bias, why not build systems that are good at catching human racial biases and redirecting a file to someone less likely to have that bias?

Humans are extremely good at some things and extremely poor at others. If rather than trying to imitate humans we tried to complement their weaknesses, reduce their conflict, etc. we would have genuinely intelligent systems and humans with rising living standards.

It seems to me that AI is the ultimate expression of the hyper masculine and especially white European desire to dominate, beat, outcompete, etc. taken to its logical extreme of dominating all other humans. Maybe if we had more diversity in the tech industry we would

have technology that actually improved people's lives and made them more productive, rather than trying to constantly prove it is better than people at the things they already do well.

Sure, you might say: well we don't have that many models of intelligence other than humans so we are just trying to imitate that. But that is an admission of a profound lack of creativity, empathy and analytic capacity, at best a crutch and certainly not something

that should be made the pinnacle goal of a whole industry. We need not to "do AI ethically" but to get beyond a discourse based on autonomous systems to imitate and replace humans to one that seeks to complement and empower humans and is focused on its interdependence.