Where will we get our history?
By now, everyone has seen Google Gemini’s image generator that refuses to generate pictures of white people…
…and ends up rewriting history.
It turns out that the Gemini designers decided it was important that generated images reflect the diversity in the world. They achieved this by amending user prompts to specify an ethnicity and gender chosen at random from a list that explicitly excluded white men.
If you ask Gemini for a white man, it says “no”.
So far, so bad.
Maybe it’s possible to just dismiss this as a software bug. Bugs happen and they get fixed in the next release. Google have switched off the ability to generate images of people until they are able to fix this. But it’s much worse than a simple software bug.
The obvious lesson to learn is that this is diversity and inclusion gone mad. This problem did not arise by accident. Someone high up at Google was determined to build an image generator that was anti-racist and ended up building the most racist image generator that you can imagine instead. My country has an election this year. Maybe yours does too. A lot of people will change their vote because of this.
It’s worse than that though. Put images to one side for a moment and think about the future of information.
In the early 2020s, we still rely on books for information and we use the Library of Congress to look up facts about the past. Imagine if, by 2030, we have digitised all the books in the Library of Congress and when you want to look something up, you use the Official AI — “Brought to You by Google” — to do so.
Google’s AI, in the interest of diversity, will tell you that Vikings, mediaeval knights and popes were all black and they won’t allow you to access information about Galileo or Julius Caesar. How will you know whether Google got it right or not? What will you compare its answer with?
Maybe there are several AIs, each with its own political bias. The Conservapedia AI will tell you that the founders were all Christian and the animals came into the Ark two by two. You can compare that with Google’s answer but Google says the founders were all Black and there was no Ark. The other AIs refuse to answer questions that they deem political.
There’s no such thing as a politically neutral information source. When there are thousands of books about the past, we can compare what they say and squint and approximate the truth. If the only sources are Google AI, Bing AI, Open AI and Conservapedia, everyone will just pick their own information source and dismiss the others as liars. Hey! Maybe the chocolate ration did go up!
A friend of mine says that this was just a bug. Google will fix it and this will all be behind us soon. AI has got stuff related to race wrong before.
Maybe Google were just over-correcting. Maybe they didn’t notice that Gemini does not generate pictures of white people.
I don't think this is what happened. I think Google did this deliberately. When they got caught, they were embarrassed and chose to correct it. Next time they won’t get caught and Google’s version of racial politics will be the version that informs us for all time.
Comes a point where Google will have shot itself in both feet. It will become widely viewed as a risibly useless source of information - not just by those of us reading this. It is destroying itself. I hope it continues to do so.
I am glad you have picked up on this Ragged. Fo rmy part - although I have spent years trying to unpick the mess that is the West's corrupted Social Justice Religion - when I came across this Google AI story, it almost made me want to give up!