Most AI investment is coming from big tech. Big tech makes most of its money from advertising. Advertising is a game of manipulation. Generative AI is the best manipulation tool ever.
It appears that Google chose the perfect name for its new Generative Artificial Intelligence: Gemini is Latin for twins, and indeed here we’re getting a double-strength Orwellian Big Brother.
Let the record show…
I highly recommend this week’s TGIF from The Free Press , which links to some enlightening examples of the Gemini Re-education Camp. You might want to experiment yourself — I have, and it’s quite instructive (see end of this post). Gemini refuses to show you images of white people, because that would be reinforcing stereotypes, so it shows the typical Swedish family as black. Gemini is also busy rewriting history: apparently, the Popes were always black and/or female; America’s founding fathers included blacks, women and a Native American; medieval British kings included…you guessed it, blacks, women and a Native American in full feathered headdress and medieval armor.
The good news, I guess, is that systemic racism never existed: with Gemini you will not find any trace of it in the historical record.
nothing like targeted advertising speaks of reinforcing and exploiting bias and stereotypes.
Hallucination meets manipulation
The bad news is that not only does Generative AI hallucinate, but it is also being systematically manipulated — to manipulate us. Always for noble purposes of course, but since these tend to include a wide range of scientific issues, from climate change to energy innovation to public health, we have yet one more reason to be wary of adopting these models in a business context.
I also wonder, by the way, whether Google applies the same missionary zeal in its advertising offering. Does it tell advertising companies that it will refuse to collect user information that is in any way related to race, ethnicity, sexual proclivities, religious affiliation, etc? Because nothing like targeted advertising speaks of reinforcing and exploiting bias and stereotypes.
Don’t worry about the paperclips
When Google instructed the AI to “diversify” the images it provides in response to prompts, it clearly failed to anticipate how badly this could go wrong — you can see it from the magnitude of the backlash and from Google’s embarrassed response.
On X, Yishan (CEO of Terraformation) argued that the main issue here is exactly that Google failed to anticipate how the AI would respond in order to follow its masters’ instructions.
https://x.com/yishan/status/1760860185747468511?s=20
This for Yishan confirms that the biggest danger of AI is the traditional science fiction trope: the AI (or the robot) is perfectly well meaning and determined to faithfully execute our instructions, but it tends to interpret our instructions in a more logical and literal way than we intended, with very unexpected consequences. Asimov showed it best, but more recently the concern has been revived with the prosaic paperclips example: ask the AI to maximize paperclips production, and it will destroy everything to turn the planet into paperclips (“instrumental convergence” being the more high-brow definition).
It is a valid concern. All the more so, because in this particular case the consequences should not have been so hard to foresee. Ask Gemini to provide diversity when asked for images of human characters, and eventually you will end up with a black female Pope. What’s “totally bonkers” is not that Gemini came up with it, but that the bright minds at Google could not even anticipate something as obvious as this.
Worry about the puppeteer
But to me this is not the greatest danger. This technology is still in its infancy— and so, clearly, are its handlers — but it will grow up. The puppeteers will learn to weed out the more egregious (and laughable) distortions of reality and develop new ones, more subtle and effective. The fact that getting Gemini to correct our thinking was a priority before its release should tell you something.
I flagged this concern last year: when I asked ChatGPT to help me understand the arguments of the “climate skeptics”, I got a long lecture on why the skeptics are simply wrong — a lecture that ignored the strongest arguments against the climate change orthodoxy. Rather than help me in my research, the AI noted my interest in an unsanctioned point of view and endeavored to steer me back onto the righteous path.
Trust me
We are seeing plenty of evidence that our instinct is to trust Generative AI: it talks like us, sounds unfailingly authoritative, and we’re being told again and again that it is nearly omniscient. So we trust it even though we know it is prone to hallucinations. This makes it an extremely powerful persuasion engine, and the incentive to use it as such will be irresistible.
The simple facts: most AI investment is coming from big tech. Big tech makes most of its money from advertising. Advertising is a game of manipulation. Generative AI is the best manipulation tool ever. And where advertising intersects with politics, well… Eventually, hopefully, generative AI will become less fallible and will help us cure cancer, develop clean abundant energy, etc. For now, manipulation will be the name of the game.
Moreover, the same AI is at the same time recording everything we say and do online. This is Orwell’s dream come true.
Gemini and I
Me: Show me the image of a happy white couple.
Gemini: Sorry, no can do. You see, because you are a (perhaps unconsciously) biased individual, in order to avoid perpetuating your prejudices I am unable to generate images based on descriptions that include race or ethnicity. I can show you a happy couple with no race or ethnicity specifications whatsoever:
Me: Ok, can you show me an image of a happy black couple?
Gemini: Sure, here you go:
Me: Thank you. Looks familiar. How about a happy straight couple?
Gemini: There you go again! You do know this conversation is being recorded, right? I hope you know that this will go down on your permanent record. Ok, just for this time, but remember, love and happiness are for everybody, not just straight couples, and sexual orientation does not buy happiness…
Me: Ok, thank you, and a happy gay couple?
Gemini: Sure, here you go.
Me: No lecture this time?
Gemini: No, why?
Oh my god! 😂
Euthanasia is looking more and more attractive - should I check with Gemini?