AI is Starving for Real People
- nicole34197
- Jan 26
- 3 min read
AI models have a problem: data. That might sound strange, because there seems to be no shortage of data, but AI models are particularly hungry for one type of data: authentic, first-hand data from real people. And that's scarce: the amount of data we humans generate simply can't keep up with the amount of data AI models need. And this has consequences.
When AI Learns from AI
Recently, I was in conversation with various agency directors and seniors from the market research industry about the proposition 'No Artificial Intelligence without Human Intelligence'. AI expert Casper Rutjes was also present and states: AI loses quality when it only learns from AI-generated data.
"AI loses quality when it only learns from AI-generated data"
But how does that work? Imagine: you ask a thousand people their opinion about a product. 950 give a predictable answer, but those other 50 - they say something different, unexpected. AI analyzes those thousand answers and learns: this is the pattern. Then AI generates new answers based on that pattern. And here's where the bias emerges: those 50 divergent answers are seen as 'noise'. Too far from the average. AI filters them out.
Then train a new AI on those AI-generated answers? Those 50 divergent answers have already disappeared. Do it again? Then the 'somewhat divergent' opinions also vanish.
With each generation that AI learns from AI, the spread becomes narrower. The outliers disappear. This phenomenon is called 'model collapse'. What remains is the predictable, the safe, the average.
Predictable Creativity
Lucas Hulsebos (CEO DVJ Insights) illustrates another consequence of 'model collapse' using a study on creativity. He analyzed various types of advertisements for years and saw that their creativity has been declining for years - ads are looking more and more alike, and with AI in the mix that trend is only accelerating.
He also compared AI-generated campaigns with work from 100% human creative skills. Result? The AI ads scored better on average in pretests, but the absolute top performers - those campaigns people truly remember and that win awards - still came from humans.
So AI optimizes toward the safe middle, while human creativity dares to step outside it. And you don't just see that pattern in advertising. It applies to every domain where AI is deployed: everything converges toward the average.
Feigned Efficiency
Another consequence of 'model collapse' is the phenomenon 'workslop': AI-generated content that looks perfect, but lacks substance. Jan Zwang (Director CI VodafoneZiggo) indicates that reviewing AI-generated output takes him more time than when that same colleague had sent a messy but substantial first draft. With mess you immediately see what's missing. With 'AI slickness' you have to puzzle out what's not there. Which question hasn't been answered? What nuance is missing?
In short, AI seems like an efficient way to produce content, but requires more review time if you want real quality. And that review can only be performed by someone with domain knowledge and experience, who knows what good content looks like versus slick-but-empty.
Now What?
Synthetic data is the new reality and you get the best results when you deploy AI with human intelligence, according to Casper Rutjes. It's precisely the combination that's the secret sauce.
For brands this means three things:
Keep investing in primary data. Want to know what's really going on? Gather insights from real people. That's where valuable insights emerge.
AI output can be very impressive, but always ensure there are people who can verify it. The seniors in an organization have value you can deploy perfectly for AI review, but they're also the ones who need to transfer knowledge: what makes something correct, creative and quality? Give juniors room to learn and experiment.
Choose consciously where you deploy what. Not everything needs to excel. Much of your content can be average and is fast, scalable, good enough. But also make choices about where real creativity and quality matter and where you want to stand out - that's where you invest in human expertise and primary insights.
Marjolein van Ballegooij
Comments