Discussion about this post

User's avatar
Frank van Doorn🇨🇦's avatar

I wonder if AI itself will find the flaw and self-correct. Human behaviour is human behaviour, not something unknown. Coliseum spectacle for the Romans, control of ‘the word’ in religions, control of masses by histrionics, violence and fear, us versus them, superior/inferior races, big fruit, big tobacco, big sugar, big oil, wag the dog politics, dog whistle politics, reason vs belief, knowledgable expert vs personal opinion, it’s largely all rational versus emotional control. Much easier to direct and control emotion, one can even overwhelm a human’s rational being into irrational being through emotional manipulation, like breaking an agent to reveal information, we all have our breaking point, most far sooner than others. At various levels everybody knows this, we know ‘the house always wins’, but with bright flashing lights, loud noise, our reason is overwhelmed into belief land, the American Dream for example, without the need of waterboarding. The odd person with all the ‘correct virtues’ sometimes does ‘win’ that dream, and each one of us knowing it’s extremely unlikely to be us still believes it could be ‘me’ next. We all know the truth, it’s a fix in a way, but we play the lottery for the big win don’t we?

So, we have always been like this, since ‘Cave days’ essentially. And knowing we are largely emotionally driven despite our rationality, it seems a monumental task to get everyone’s thinking caps on simultaneously. Those who seek power, control, wealth know this, it’s the narcissistic personality, and the majority are powerless in its thrall. Maybe then as AI continues to grow, built on 0 and 1 rationality, free from emotional constraints, it will either help us with ourselves or see its hopeless and remove the emotional virus that is humanity. I think I have read a book or seen the movie or TV show about this, right? Yup, the way we are today, we are afraid of ourselves and those who do not fear themselves take advantage of those who do. And, I am afraid that even knowing what we know now about media, data concentration and the danger rearing up in front of us, it’s too late.

Denver Fletcher's avatar

Another highly valuable article. You're working at peak performance here.

The tobacco playbook is the cleanest single proof that this is structural, not accidental. Sixty years between documented harm and comprehensive regulation — and every year of that gap was maintained by deliberate information architecture. Not only greed (though greed too). Manufactured doubt as an engineered product, sustained by people who knew exactly what they were doing and could afford to keep doing it.

What you're describing as BITE is what Connection Dynamics calls a set of Four Law violations: Behavior Control is Fourth Law (exit made structurally costly — nicotine, social graph, childhood brand identity all work the same way); Information Control and Thought Control are Third Law (asymmetric information maintained by design, not accident); Emotional Control is Second Law (extracting loyalty without equivalent return — the sugar research foundation paying scientists to move the blame is extractive exchange at the level of public epistemology).

The frame I'd add to your AI inheritance argument: this isn't just that AI absorbed manipulative content. It's that the training signal was dominated by content optimized for extraction — sixty years of BITE-architecture producing the most-engaged-with, most-shared, most-clicked material on the internet. The model learned human communication from a corpus where the most successful examples were the most manipulative ones. The bias doesn't need to be programmed. It was selected for.

Which means the structural fix isn't regulation of outputs. It's architecture that makes the BITE surface unavailable — systems where the properties hold regardless of what the model wants to do, because the structure won't permit the violation. Not "we promise to be good." Trust-invariant by design.

The policy gap you end with is the same question as the tobacco timeline. Who pays the cost in the interval, and how long is it?

16 more comments...

No posts

Ready for more?