Outsourced Intuition Makes Us Easier to Manipulate

In the last post, I argued that one of the greatest risks of mass adoption of AI is not simply that it weakens our ability to think for ourselves.

However I think there’s an even worse consequence of our sliding down the slippery slope of AI mass adoption. As we rely on AI more and more to answer all of our questions and to recommend what we read, buy, and watch, the more AI effectively becomes the default place where truth and knowledge live for all of us.

Once that shift happens, AI becomes more than a tool; It becomes the filter through which we experience information.

For example, if I ask the AI what are the best books to read about the negative effects of capitalism, I am essentially asking the AI to give me its statistical prediction of which books on the subject would be most worth reading. Implicitly, and for most of us, without knowing, I am accepting the biases of the humans that chose which sources to include in the model training, the built-in biases of the sources themselves used to train the model, and the preferences of the humans that provided the reinforcement learning to train the model.

This is especially dangerous when researching controversial subjects where the model was specifically trained on sources that favored one side of the argument, and the model was actively disincentivied to discredit the other side of the argument.

This hidden bias is especially insidious because manipulation rarely announces itself. It does not usually look like an obvious lie. More often, it looks like a helpful summary, a confident answer, a ranked list, a suggested next step, or a framing of an answer that feels neutral enough to accept without resistance.

Perhaps even more dangerously, AI sets the guardrails of what is acceptable to ask. It subtly frames the range of possible answers. Things we might have thought to ask are never even considered. The machine offers us the illusion of bountiful choice while narrowing the scope of our imaginations.

This is where the problem becomes larger than accuracy. A perfectly accurate answer can still be manipulative if it narrows the question. A useful summary can still distort the world if it removes the tension, conflict, or uncertainty that made the subject worth thinking about in the first place.

This implicit acceptance of model bias coupled with the collective weaking of our thinking muscles means that we become recipients of information curated by someone we do not know and, quite likely, do not trust.

And as long as we, the public, are not the ones defining what sources and biases go into training the models, the more vulnerable we become to corporate manipulation and government propaganda.

And on the note of government propagande, if you don’t think governments are already coming up with ways to poison models in their favor and against the favor of their enemies, you are naive.

So when we are using AI, we should keep in mind the following questions: Who benefits from this framing? What assumptions are hidden inside the answer? What source is being privileged? What alternative explanation is missing? What product is being pushed on me? Who paid to tailor the answer to their own agenda?

If AI becomes the front door to knowledge, then whoever shapes the model, the retrieval system, the ranking logic, the policy boundary, or the interface has enormous influence over what we see and what we never think to look for. That influence does not have to be malicious to matter. Incentives are enough. Defaults are enough. Convenience is enough.

The answer is not to reject AI. The answer is to refuse to become intellectually passive in its presence.

Use AI to challenge your thinking, not to replace it. Ask it for counterarguments. Ask what is missing. Ask for sources. Ask who might disagree. Ask what assumptions the answer depends on. Then step outside the machine and read, listen, compare, and decide what you want to think.