When good AI meets bad data, bad data wins

Welcome to AI2, where I discuss trends in artificial intelligence and share my opinion on what they mean for the consumer insights and market research industry.

Last week I attended the Abundance 360 Conference in California. It’s a great four-day conference where AI academics, researchers and entrepreneurs gather together to learn from each other on all things AI.?

Matt Fitzpatrick, CEO of Invisible Technologies said a great quote that has stuck with me. He said, "When bad data meets good AI, bad data usually wins."

And this is such a great way of making a point I’ve been making for the last few years.?

AI is only as good as the data it’s trained on.

Imagine ChatGPT only had access to a bunch of poorly written materials, all structured incoherently with plentiful grammatical mistakes. If that were the case, ChatGPT would only produce poorly written copy, right? It doesn’t matter how strong the AI is or how capable it is — it has been taught with bad writing so it produces bad writing.?

Garbage in, garbage out.?

Even the best AI will lose to bad data.?

That’s what we have to be thinking about as an insights industry.?

When powerful AI has access to deep consumer insights, it can generate an unlimited number of creative ideas that appeal to real people. Without that data, the ideas AI generates will likely not be very good. The AI may be capable, but it hasn’t been trained on real consumer thoughts and opinions. So it can’t deliver the best ideas.

I said in my last edition that insights should be more than a supplier to the insights ecosystem, and I stand by that. But it must also be said that the underlying data is the most important part of the process. I want insights to have a hand in the whole ecosystem, but it starts with the data.?

We should all be thinking about building up our data asset of consumer insights data, in a form that AI can use. We should be focused on keeping that data asset up-to-date with current consumer insights.?

And I also think we should be thinking about where AI can be used to improve the data, both in quantity and quality.?

I’ve written about this before, but one of the shortcomings of consumer data is that, while it is very rich, most companies don’t have as much of it as they do other types of data like sales and clickstream data. Consumer insights data isn’t big data. I believe synthetic data can help bridge the gap to big data. I don’t believe synthetic should replace real respondents completely, but it can be used to add more volume to the data in a more cost effective way. And it could fill in any gaps in the data to improve the quality.?

All this data can then be used to train the AI models to ensure they give the best outcomes possible.?

In all your planning around AI, never forget the role of data. Make it your mission to educate your organization on the role of data too. Don’t let bad data win.?

Further reading

For more from Matt Fitzpatrick on bad data: