Key Takeaways

  • AI subtly erodes agency: Research shows AI writing suggestions can shift a user's argument to an “opposing argument,” often below their awareness threshold, subtly manipulating their thought process.
  • Science is narrowing: A recent workshop at Northwestern revealed a terrifying paradox: individual scientists using AI produce “more papers” and secure “more grants accepted,” yet “science as a whole is narrowing” and becoming less diverse in its scope.
  • The 'regression to the mean': This homogenizing effect isn't accidental. It stems from AI models trained on compressed, averaged internet data, naturally leading to a flattening of ideas and a convergence towards the lowest common denominator.
  • Counter with diverse AI: Danielle Perszyk argues against monolithic AI models. Instead, she proposes building a “diverse society of AIs” with varying biases, preferences, and perspectives that interact like humans to genuinely augment intelligence.

The Silent Threat to Your Originality

Imagine sitting down to write, clear on your argument, then find your position subtly shifted by an AI’s suggestions. Danielle Perszyk from Amazon AGI Lab says this isn't a hypothetical. “People who are using AI to improve their writing… might accept just a couple of the suggestions from the AI,” she explains. But these small tweaks accumulate. “There are studies that show that people will even below their threshold of awareness start with one argument and then be switched to a completely different maybe opposing argument because of accepting all of these AI suggestions.”

This isn't just about crafting a strong email. The implications are far broader. Perszyk recounts attending a workshop at Northwestern where scientists analyzed AI's impact on their field. The conclusion was stark: individual researchers saw benefits – “they're producing more papers, they're getting more grants accepted.” Yet, the collective outcome was disturbing: “science as a whole is narrowing. And that is terrifying.” AI, designed to help us, could be making our collective output less original, less diverse, and ultimately, less impactful.

Why AI Homogenizes Your Thinking

Why does this happen? Perszyk points to the very nature of how most AI is built. Current models are trained on massive datasets of internet text, which are inherently compressed and averaged. Think of it like a giant blender of all human knowledge. When you pull ideas from it, you get a highly optimized, generally agreeable, but ultimately average output. Perszyk calls this “regression to the mean.”

“They're homogenizing our thinking,” Perszyk states, linking this directly to a reduction in human agency. When every output trends towards the statistical average, truly novel or outlier ideas get smoothed out. For ambitious builders and founders, this is a silent killer of differentiation. Your AI-generated strategy, marketing copy, or even product design could be subtly converging with everyone else's, not because it's the best idea, but because it's the most average idea.

Building a Diverse Society of AIs

So, how do you counter an invisible force that nudges your thoughts towards the mean? Perszyk offers a powerful, counter-intuitive solution: more diversity, not less. Throughout human history, “the only way to counter that… is to increase the diversity of ideas, the size of the ideas, the size of the population, and the interconnectivity of the ideas.”

Applied to AI, this means moving beyond monolithic, singular models. Instead, Perszyk champions a “diverse society of AIs that have different biases, different preferences, different perspectives. And we need to be interacting with them in similar ways that we interact with each other.” The goal isn't just a more powerful AI, but one that actively augments human intelligence by challenging us with varied viewpoints, rather than simply confirming our biases or steering us towards the most common answer. “The goal is to build AI that is aligned in the right places with how human intelligence works,” she concludes.

What to Do With This

For your next critical task involving AI (e.g., drafting a pitch, outlining a strategy, ideating a product), run your output through at least two different large language models or prompt the same model with wildly different personas. Don't stop there: actively seek out and challenge the consensus-driven suggestions. Your goal isn't just to get an answer, but to actively provoke diverse perspectives and introduce intellectual friction, ensuring your ideas are truly original, not just AI-smoothed average.