The alchemy of Artificial Intelligence and the problem of knowledge:
A clear example of the fallacy of socialist calculus comes mostly from Ali Rahimi, who argues that it is alchemy, not science, that is building Artificial Intelligence (AI). According to Rahimi, machines only handle simple tasks and without major skills, they process large amounts of data to see what fits. As he says, "I would like to live in a society whose systems are built on verifiable, rigorous and comprehensive knowledge, not alchemy."
This sounds reasonable enough in some contexts; after all, I have always preferred that nuclear plants be built by those who know what they are doing. However, it is less of a criticism when it comes to societies and economies; as Hayek and Mises point out, some things are too complex for us to fully understand.
In fact, in developed economies we use data processors without understanding the details in the process. We don't know how many apples will be eaten next year, so we let market processes take care of it. As Hayek insisted, this is the only computing engine capable of doing the necessary data processing to give us useful information.
"Artificial Intelligence is going to change the world" they say, "we don't know how, so we must plan! Direct! Channel those forces!"
We also see this same problem in several well-intentioned editorials. "Artificial Intelligence is going to change the world" they say, "we don't know how, therefore we must plan! Direct! Channel those forces!" But if we don't know how a technology is going to change the world, planning is simply not feasible.
As The Guardian rightly asks, who could have known that the switch from whale oil to kerosene in the 19th century would eventually lead to the development of plastics? And what plan initiated in 1880 would have given us a world with or without plastics? None, of course, because no one even knew of the possibility. The same is true when we try to figure out what effects Artificial Intelligence will have in the coming decades; in both cases, total ignorance is not a good basis for initiating a plan.
Homo Sovieticus does not exist
Another error is what I would call the Soviet New Man problem. This describes the idea that while the joys of socialism were not too well suited to real human beings, the Soviet government would eventually create an entirely new type of man who would love socialism to the fullest. Of course, homo sovieticus never materialized.
And this brings us to another common argument about Artificial Intelligence: that it should not incorporate the negative traits of real human beings.
For example, we know that many humans are racist, misogynistic, greedy and short-sighted. Artificial Intelligence, too, can pick up on those weaknesses and can definitely display what we might call "bias."
The mistake of the Soviet New Man would be to try to design AI for a world free of humans with all their disordered and illogical behavior.
To insist that they will not do so is to miss the central issue entirely. The only possible use of AIs is to provide us with knowledge about the world we live in, knowledge that we cannot derive solely from logic and what ought to be, but that must be obtained through truthful data processing.
After all, the world is full of deeply prejudiced human beings. An AI that didn't take that into account would be of little value in describing our world. That's why we should not only want, but absolutely insist that AIs incorporate our flaws.
The mistake of the Soviet New Man would be to try to design AI for a world free of humans with all their disorderly and illogical behavior. Of course, it is also an argument against the various alternatives to free market capitalism. Of course, if humans did not respond to incentives, rigid equality of outcomes might work well. In the real world, incentives matter, and any system that does not allow for a degree of inequality derived from application or effort is not going to work. The AI flaw is subtly different but relies on the same basic error.
What all this means is that we can train AIs on the basis of data and see what happens, rather than trying to understand complex interactions, which we can't do anyway. If the past century has taught us anything, it is to avoid those two key mistakes: trying to plan complex systems and assuming a world devoid of ourselves, humans, and all our messy habits. AI pioneers would do well to consider and remember these lessons.
