AI Slop Will Eat Itself

ChatGPT said it was probably fine, so off he drove... The next day, the garage confirmed it was a brake fluid leak.

A friend of mine recently had an issue with his car. There was a mysterious leak dripping underneath.

To my surprise, he took a video of the drip and sent it to ChatGPT for diagnosis.

ChatGPT said it was probably fine, so off he drove.

The next day, the garage confirmed it was a brake fluid leak. I’m not a car expert but I think that’s quite serious – I’ve heard braking is an important part of the whole driving experience.

You can see what I’m driving at here – blindly following AI output without thinking for yourself is a lazy and dangerous practice.

At the IIEX conference last month, I was surprised by the number of companies pushing shiny new omniscient technologies that promise to spare you the pain of rolling-up your sleeves and actually thinking for yourself.

In the clamor for faster and cheaper, the industry risks morphing into a slop factory that, metaphorically at least, could find itself upside down in a ditch wondering what the hell happened.

We’re talking here, of course, about tools offering synthetic data, digital twins, agentic AI, AI generated analysis, and so on.

There’s nothing inherently wrong with these AI technologies of course – at codeit, we’ve been at the forefront of applying AI to Market Research for the last 10 years. However, as the industry races to get these tools to market, we seem to be sacrificing the ability to keep people – real Market Research experts – involved in the process.

Building a wrapper around an LLM is easy, the hard bit is building a rock solid UX that enables people to stay in control and work intelligently with that AI. Trust me, we’ve spent years at this particular coalface – slop is easy, UX is hard.

How quickly as an industry, we have moved from “AI is coming” via “Human in the loop” to “Press a button and ta-da! there’s your insights”.

The problem here is that outputs are not insights. True insights come from real people getting involved in the data, thinking about it from all angles, applying their expert context knowledge and crafting a considered interpretation.

There is a lot of evidence emerging that AI done badly seems to speed up one process, but makes downstream processes slower as other people spend time unpicking low-quality work.

For Market Research, this smells of a race to the bottom where our work becomes less “Wow” and more “Meh”.

Over time, my hope is that research technology iterates and we get the human back in the loop and in charge again. This will take time and money, but it’s worth the effort because it’s the only way to harness AI without throwing the baby out with the bathwater.

This has always been our ethos at codeit. AI-enabled software should help people go much faster, but it has to be human-led.

Our unbeatable user interface, perfected over 10 years, enables people to drive cutting-edge AI, but at all times to have them guiding the process, keeping the AI in check and getting to useful results they understand and fully trust.

Find out how to do AI correctly with a free trial of codeit