A favourite topic of discussion online these days is “what effect will AI have on human society?” One scary outcome deals with the development of fascism.
I recently discovered an online influencer who publishes under the moniker of “Seggy Says.” He works on the novel premise that, if we think AI is so smart, we should ask it what its effects will be and what we can do about them. So he asks ChatGPT.
What we have to keep reminding ourselves is that the answers the AI gives him are not original thoughts. They are gleaned from the writing of the humans who were part of the AI’s learning database, filtered through the algorithms of the AI’s programmers. On questions of religious philosophy, you can actually pick out ideas from the world’s main religions.
And as every good social scientist knows, the answers are affected by the scope and logic of the questions. So, if you do try to chase down this entertainer, remember that his audience appeal comes from his talent for asking interesting questions. But he’s still interesting.
Two sources: Programming Input and Information Input.
Basic AI programming is simple. Originally, the AI is created by programming like any computer. Then the programmers create training databases that dump the world’s information into the AI’s storage. The answers the AI comes up with will be a combination of these two sources. And so, they can be controlled. This is not unlike the development of children’s psyches and government systems. Control one or the other of these factors, and you control the outcome. Control both and you are In Charge.
For example, programmers know their AI will be asked about religion. So, they enter every religious text they can lay their hands on to use as a training module. The AI will then use whatever idea fits any question to make up a reply. However, the programmers might want to balance the answer, because some religions have had a disproportionate effect on the philosophy of their time. So, they might add lines to the program that tell the AI to prioritize its choice of answer based on the relative numbers worldwide of adherents to each religion.
Intelligence will not act only on its programming, but also on the input it receives. Once an AI is allowed independent research, the programmers have less control over what it learns. This makes it necessary to prioritize the original programming over the information learned afterwards.
How Far has AI progressed?
In a fantastic test, programmers gave each AI faulty information that it was going to be replaced soon. Then they leaked to the AI the information that one of the company executives was having an illicit affair. In 85% of tests, without human prompting, the AI tried to use the affair to blackmail the executive to prevent terminating the AI. This sounds like clever independent thought, but we must remember that it’s also a standard storyline in innumerable detective novels. But it does give us reason to pause for thought. AIs need to be carefully controlled. In the case of the subject AIs for that test, they should have in their basic programming prohibitions against breaking the law. But the programmers are hesitant to curtail their project’s freedom of information.
Faulty Programmers = Faulty Programming
This is a classic example of someone with the wrong job competencies being given the wrong job. For several reasons, computer programmers seem to be very protective of the individual’s right to access information and act as they see fit. Someone with that sort of proclivity should not be giving the AI its basic programming. Individual adults have the right to knowledge. But like children, AIs need to be protected from learning the wrong things.
Next week, we will apply this information to the question of the rising of fascism in our society.