Why Market Researchers Can't Afford to Ignore Outliers

egonzalez267
Jun 8
2 min read

From the desk of CEO, Kevin Karty.

With the advent of LLMs, it seems that pop intelligentsia has turned against heroes of the last decade. Alas, Mr. Gladwell, we will miss you.

In the breathless AI-driven news cycle, when was the last time you heard pop science media expounding on the importance of understanding Outliers? Remember when Outliers were all the rage and every Keynote Address at a conference touched on them in some way?

That's not to say that anyone is challenging the empirical evidence that Outliers matter. It's just that they aren't talking about them anymore because... well, they're inconvenient. Weird. Hard to predict. Instead, we seem to be focusing on 'Inliers'. I thought that was a new term, but it's already in the lexicon. AI is GOOD at understanding Inliers. Trends, norms, central distributions, consistency, and central patterns. It's fast and cheap, and reasonably accurate. When simulating synthetic respondent data, LLMs can often do a reasonable job at this (if the category and sample are well known).

The thing is, AI can ALSO be used to find Outliers, but you have to actually go looking for them. If you are trying to simulate data, or build a digital twin persona, AI tends to ignore Outliers and focus on the 'safely central' Inliers, which are easier to predict.

What I struggle with isn't the technical issue around this, however. As a researcher and social scientist, it's the deeper impact on literally EVERYTHING around us. I'm less afraid of everyone behaving the same because they are trying to 'fit in'. I'm more afraid of everyone slowly converging to the sameness because they are being subtly nudged by an every-present AI 'invisible hand' giving us helpful advice.

Over the last few decades, we've increasingly documented the hidden benefits of different kinds of diversity: biodiversity, neurodiversity, cultural diversity, etc.

It seems like there's going to be a critical need to support some other type of diversity form the confirming incentives of AI:

Cognitive diversity?

Psychodiversity?

I'm not sure what it is, but I know we need it.