Thank you for the interesting response. I think we have a basic difference in viewing human strengths versus machine learning and data modeling strengths.While this is almost peripheral in the essay you responded to it is important to others things I write.
There are a very large number of challenges in developing ML systems that are 1) generally understandable by humans, 2) cleaned of human cultural relics such as racist tendencies, and 3) audit-able, again by humans. Number 1 and 3 are related but the later is far more difficult.
The whole thing is based on conclusions produced by algorithmic analysis of vast tables of data in closely related situations. People cannot do that. We can create a logical rule or set of rules that are understandable by people but making high speed adjustments to this in a feedback loop in a dynamic model means the conclusions are based on thousands if not millions of data points.
Without going into detail, ML systems are functional at this now but only in relatively simple questions. They are pretty good at suggesting additional items to purchase based on a full purchasing history. They get weird frequently in attempting to create things from data samples that people would find easy. At this point the power of ML/AI is in determining a model of simple things from a lot of examples and then using those to present similar things. That, right now, is good enough.
Obviously this is not what people think of in AI, particularly. Current systems are so far from sentient most of our pets are far smarter. But that will not last. The key here is self replicating. That is the scary thing that freaks people who know about this stuff but it will happen. And we need to happen because we cannot do this stuff and we need to.
I consider Homo sapiens as being at the edge of evolutionary failure. Of course evolutionary failure is also the trigger for evolutionary change. I agree with Ray Kurzweil on that as variations on a cyborg future. Until we can add the speed of electronic digital processing to our powerful but slow biological systems we will need to defer to more powerful AI/ML systems that are coming.
The human failure is emotional thinking and, ironically, are mental architecture that is designed for general pattern recognition from our senses. That is one of the reasons that people cannot internalize statistical models without massive training. We look for colors and shapes and movement (things that use to eat us) but not for logical are mathematical formula. So we understand statistics primarily as charts and graphs. Ahh, now I see! is a response to a graph and not a table of numbers that produced the graph.
My feeling is that most bad management by people is the result of inadequate understanding of available data. More importantly we have terrible decisions by people who refuse to look at or accept the data that says that what they think is wrong. I don’t think I need to mention the America’s current political collapse that is the result of that.
But people’s pattern sense is at its highest in watching other people. We are primates and are social by several million years of evolution. AI system can now easily recognize people but are nearly helpless at identifying emotions. They are getting there and can now produce the correct facial expressions but this take truly massive sampling and very, very careful processing. That is why it is easier to present emotion than identify it. But that will change also.
I’ll go out a little way here and say that the reason for AI for human management is the great and growing diversity and individual identity. We are a long way from ‘one size fits all’. People can’t handle that and, hence, the pseudo conservative emotional rejection of it all. We have a significant percentage of our populations that react with emotional rejection to anything but people nearly exactly like them. We don’t’ live in a world that can tolerate that.
We need rights (based in law) that are very extensive and carefully neutral for all possible identities (and, yes, this will extend one identity with characteristics for each human on the planet. Later will get into our socially connected, increasingly intelligent animals and, finally, our soon to be sentient machines.
At some point this will need to merge to some extent as equal but individually distinct. If we survive it will be very interesting. I think we need a lot of modeling and very serious help with matching statistical patterns before we start growing Intel GPUs for ourselves.