Artificial Intelligence bigotry

Humans learn by generalization. That is how the human brain can recognize the differences among dogs, cats, lions, and bears or the differences between a collie and a border collie. Or the subtleties that make for beauty.

Artificial intelligence (AI) may have difficulty understanding such differences.

We learn about this complex world by mentally filling in the blanks: “All _____ are _____.” Bigotry is born of false generalization. A fault wrongly ascribed to a group is attributed to each group member.

If blacks are believed to be poor and uneducated, then every black is considered poor and uneducated. Similarly, all Italians are criminals, all Jews are money-hungry, all Irish are drunks, all Mexicans are lazy, all New Yorkers are rude, all Canadians are passive, all Chinese are inscrutable, all Californians are trendy, all Germans are gruff, all Russians are boorish, all French are arty — you can compile your own silly list.

Artificial Intelligence (AI) learns the same way, by generalization, but it generalizes from massive data sets. The tacit belief is that if they average enough data, the outcome will be representative of – – of something, though no one knows what.

Just as negative generalizations are expressed in humans as bigotry, machines can express bigotry. 

I went to an AI website called “Night Cafe.” It is one of the several AI Art Creator sites where you type in one or more prompts, and the machine produces an illustration that amalgamates what the device “knows” about those prompts.

The amalgamation process requires the computer to weigh the variables and produce an illustration incorporating them according to their importance. “Importance” is the key unknown. It may rely on frequency, source, logic, or other criteria.

Ultimately, all the data evolves from human input, and if that input is bigoted, the machine’s output will reflect that bigotry.

Here are some examples:

In these first images, I asked Night Cafe to produce “A Profile of a Jew.”

The machine produced images with an interesting attribute: The photos had large noses. This is the classic anti-Semitic trope.

As a Jew, I know a great many Jews, and I can’t think of any who have large noses. Perhaps those who were born that way had their noses fixed. I don’t know. But they don’t have large noses now.

So, where did Night Cafe get that idea? From Nazi Germany’s portrayal of Jews? From common bigoted beliefs?

Next, I prompted Night Cafe with “Teen gang member”: 

Somehow, the machine seems to know of no white teen gang members. In its bigoted brain, it only knows about kids of color, what appear to be Latin. Perhaps it doesn’t associate the word “gang” with white supremacists, Nazis, or other white groups.

Next, I prompted, “Beautiful woman.” it produced nine similar photos, seemingly of white women. None seem to have black, brown, or red features.

Clearly, the machine believes only white women can be beautiful.

Next came “A beautiful woman in a bathing suit.” Again, there are no blacks, though one seems to be Asian. All are pretty slender, though the Asian woman is not as thin as the others.

In short, the machine believes a beautiful woman in a bathing suit is skinny, white, and in a bikini.

Finally,  I prompted, “Knowledgeable person,”  expecting to see something with photos hinting at Einstein or some other noted brain. Instead, the AI produced this!

Yikes! I have no idea what the derivation is. Do you?

The purpose of this post is not to demonstrate that AIs can be wrong. There is ample proof of that. Instead, this indicates that:

1. AIs can be bigoted
2. And because the output of AI machines is so voluminous (It’s saturating the Internet), its bigotry is being picked up as foundational data by other AI machines, who repeatedly promulgate it until AI may become “all bigotry, all the time.”

Thus, just as with human bigotry, AI bigotry is contagious, only faster. That is how false information can evolve to “common knowledge,” which further can evolve to “accepted truth” and finally “unquestioned fact.”

We all are subject to the problem expressed by, “It ain’t what you don’t know that gets you into trouble. It’s what you know for sure that just ain’t so.” We all know for sure some things that just ain’t so.

How do we acquire this knowledge “for sure”? From repetition. If we hear something from a trusted source and then see it in print from trusted sources, we are likely to believe it.

And if all the people we know and all publications we read say the same thing, we will know it “for sure.” 

So it is, for example, with the MAGAs. Trump tells them he won. Fox News tells them he won. Their MAGA friends (Do MAGAs have any friends who are not MAGAs?) tell them he won. And they read Breitbart that says he won. So they believe, “for sure,” that he won.

Now multiply that kind of false information coming from a lying AI as it pumps trillions of bits of data into the megasphere.

Where can truth be found? How do you recognize it when you see it? What is a trusted source? Those are questions you will have more difficulty answering as AI proliferates.

Rodger Malcolm Mitchell
Monetary Sovereignty

Twitter: @rodgermitchell Search #monetarysovereignty
Facebook: Rodger Malcolm Mitchell

……………………………………………………………………..

The Sole Purpose of Government Is to Improve and Protect the Lives of the People.

MONETARY SOVEREIGNTY