Opinions: AI

As all good data people know, the results that you get from data are only as good as the quality of the data that you begin with.

This week, I read a question on Reddit asking if Chat-GPT is getting dumber, and while it was a subjective conversation that ensued, the fact is, yeah, maybe.

Because the data that it is picking up is ‘poor quality’. Don’t come at me tech bros. I mean that strategically, and humanly, not from a QA perspective.

It’s poor quality because we as a society are human and it’s learning from us; we have biases, we make mistakes, and we have foundations which are not always fit for purpose for the future we’re building.

No need to get defensive.

A friend of mine who is gay took an online bias test once and found that she is biased towards gay people. Our biases are not built on our personal beliefs, they’re unconsciously picked up from the society we live in.

But it’s not just about diversity biases. I think by now we’re all starkly aware of the diversity issues facing our industry, and I hope a good strong percentage of us are open to listening and driving change, so we look for and interrogate our datasets accordingly.

Data biases can be so much broader, and so much simpler.

I was reminded recently of a project to review a Welcome CRM program, where the team were told the first email in the series performed brilliantly because the click rate and subsequent web dwell time were both high.

Well, we looked into it, and it turned out that the high click rate was focused on the “why am I receiving this email?” button.

(I guarantee that every one of you who is snorting at that example has something similar happening somewhere with your own data in your own business.)

For me, this moment in time is an opportunity for analysts to learn the strategic skills to think contextually about the data, above just seeing it as values on a page.

And vice versa, strategists need to learn about the basic operational side of data to contribute to the conversation and how analysts interrogate data to gain real insights.

If everybody leans a little into each other’s discipline, we can collaborate to review the data and challenge each other, so that we don’t just allow AI to create ‘high performing’ work which is history repeating itself and actually just confusing consumers.

Like all new and exciting tools that come along, AI gives us an opportunity to do things differently, so let’s all try to use it in a way that doesn’t mean everything stays exactly the same.