The Consensus Machine

The current explosion of artificial intelligence and machine learning uses has a lot of people worried.

AI can replace human workers, write college papers, create very realistic internet deep fakes and so on – those are the worries that stand out now.

There is an even bigger problem: AI as a consensus machine.

Where does it [AI] get the information...? In theory, everywhere on the web.

In theory.

That’s because the programs are written by people, therefore the program is guided by the ideas, thoughts, and biases of those people. Writing AI does not magically "blank slate" a coder.

In other words, the machine learning systems do not learn at random or willy-nilly. They learn what they are programmed to learn.

. . . the parameters [for AI] are not, it seems, just about checking factualness. They clearly include very specific political and social ideas and constructs, and it is this problem that should be the most worrisome.

Because what AI is and will do with greater and greater frequency (and eventually looping in and narrowing itself) is to create [a false] consensus on an issue. The pandemic response, climate change, homelessness, diversity, equity, etc. are all complicated issues, but the AI programs can make them all seem much simpler by giving rote woke responses.

It is claimed that 97% of climate scientists believe in anthropogenic global warming. That’s because those surveys are self-selecting for people who believe it already. That and the fact that governments and foundations are dumping money into the madness so if you say don’t believe it you’re off the gravy train.

AI poses a similar issue as it could also self-select based on the whims of the programmers. And then everything – slowly or maybe not – gets a right and a wrong way to think. By its very nature, AI will . . . decide what is the putative "consensus" of the moment and repeat it ad nauseum to reinforce it.

Case in point, Google Gemini > Learn more: Google's Gemini Debacle