Paul English Shares His Approach to Effective Giving

Fair enough. I guess I'll engage with this silliness once. Let's start with the basics. There's this vague, handwaving notion of "intelligence." In the community pontificating about this it seems to be operationalized as a combination of manipulating abstractions and extracting new, higher level entities from lower levels of abstraction. For the latter, the only AI work to date that I am aware of on the latter is Drescher's 'Made-up Minds'. Multi-layered neural nets end up with interior points that map to some sort of features, but there is a big jump from that to a system being able to develop an abstraction and then operate in terms of it. For the former, computers are very good at it because humans built them as tools to extend the human ability to do it, just like we built hammers because our hands are kind of a lousy tool to hit things with. There is an illusion of sudden, dramatic progress on this kind of intelligence because curves with lots of parameters (neural nets, SVMs, etc.) can fit all kinds of things if you can get the fit to converge. The advances in machine learning have been tricks to get fits to converge and ways of mapping data into spaces where we know how to get fits to converge. This isn't a denigration of the field. It's given us enormous capabilities that we didn't have before, but we need to be clear about what it is before we start extrapolating: curve fitting. Machine learning today lets us fit curves in order to engineer task specific algorithms. And even that is already going to get strongly restricted over the next few years as legal requirements get piled on to not discriminate against protected classes, provide an appealable argument that can be subjected to a reasonable person standard in a court of law, and a pile of other stuff.

Plus all of this ignores the wide range of forms of intelligence that psychologists have identified. The sociological reason why that's so are interesting in their own right. The stories the practitioners tell about themselves are about brilliant individuals grappling with hard symbolic problems and conquering themselves, and that is the view they project on intelligence rather than a more nuanced, realistic one.

Then there is the link from some form of symbolic intelligence to reshaping the world. Let's take a quote from http://www.openphilanthropy.org/research/cause-reports/ai-ri... which I think is representative:

> There was therefore relatively little time for evolutionary pressure to lead to improvements in human intelligence relative to the intelligence of our hominid ancestors, suggesting that the increases in intelligence may be small on some absolute scale. Yet it seems that these increases in intelligence have meant the difference between mammals with a limited impact on the biosphere and a species that has had massive impact.

This makes very strong assumptions about what drove the shaping of the environment by Homo sapiens. Chimpanzees engage in coordinated problem solving behavior. So do dolphins (and they have engaged humans in their solutions). For most of the history of Homo sapiens we operated at levels not very different. So what led to the accumulation of discoveries that culminated in our modern world? The ability to transmit behavior abstractly was a big one. The ability to engage larger social groups by using language for grooming instead of picking lice off. We know that humans have differences in their mental structure that let us retain vastly larger vocabularies than gorillas or chimpanzees, but that's important for transmission of behaviors rather than social grooming. The accumulation of behaviors that led to our modern world is much more akin to natural selection than it is the model of a lone genius thinking up something.

I have yet to see any indication that there is some superhuman level of various kinds of intelligence in the offing or a link between that and major effect on the world.