Should UX contractors fear the rapid rise of AI, specifically ChatGPT and GPT4?
Contractors have already heard that Chat CPT probably isn’t going to eat the lunch of IR35 contract reviewers. And similarly, no, ChatGPT -- or any subsequent form of AI for that matter -- is not any kind of threat to your career or professional existence in UX, writes Dr Nick Fine, a principal UX research consultant and strategist.
The dig-deep required is deep, just like the watching time
How can I say that with such confidence? Late last year, when ChatGPT was first made available to the public, I was a contractor working as a user researcher for the Cabinet Office. User research is a practice that involves a lot of hard work, heavy lifting and sustained effort. Many long hours watching and analysing data, often interview or test video footage. It can be soul-destroying and requires the (human) user researcher to dig deep -- constantly.
To give you some quick context. In a round of user research, typically we see 5-8 users per week. Each of those sessions can be up to an hour, but for this exercise we’ll say 30 minutes. Eight user sessions of 30 minutes equates to FOUR HOURS of watching time. Those 4 hours are broken up by frequent pauses, so usually has a 3-4x multiplier on it, so 4 hours easily turns into TWELVE-PLUS HOURS. And remember, that was with an efficient 30-minute session! It’s easy to see how the user researcher’s life can rapidly become overloaded, because if you don’t clear the previous week of footage, it starts to build up the following week. Before you know it you have days of watching to do, and it can be all-consuming in the worst of ways.
Promise, pressure and product managers to keep happy
Enter ChatGPT in late 2022. As a user researcher, this new AI technology held lots of promise; to be able to help me relieve some of the pressure of analysis, to keep my delivery and product managers happy and on schedule.
My first experiments involved anonymising a previously recorded Google Meet video session transcript and then chopping it in half, so that I could enter it into ChatGPT. This first experiment was genuinely mind-blowing. ChatGPT, without me even prompting, gave me an accurate summary of the conversation held in the transcript. My jaw was literally wide open. I sat for a few minutes absolutely stunned. What would have taken me not an insignificant amount of minutes took this AI tool seconds.
That first dread-filled glance...
Initially, I had all the fears that you probably are having right now - ‘Omg, this thing replaces me and is faster/better/cheaper.’ At first glance, and this is why I think LinkedIn is ablaze with ‘End of Days’ forecasts for many careers since the tool came along, ChatGPT looks to be a career killer – at first glance.
Excited by possibilities, I carried on kicking the AI tyres. User research is predominantly all about user needs and pain points, specifically mining them from observed or behavioural data. I wanted to know how ChatGPT fared when I asked it to list the user needs and pains from the conversation. ChatGPT proceeded to produce two lists for me, one listing all the user needs, the other listing all the pain points.
Holy moly, this looked (again) like ‘game over’ for me, and I figured that maybe the end really was nigh. Here’s the thing though. On deeper analysis, ChatGPT fundamentally missed the biggest user need of all. There was absolutely no mention or reference to it, it was completely overlooked or not considered.
GPT4? Invention on another level
The rest was all very accurate, and importantly, highly convincing. But to reiterate, it missed a huge nugget of insight. When GPT4 was released a month or two ago, I tried again; same methodology, same transcript. This time not only did it still miss the major user need, but it actually inserted a user need and a pain point that did not exist in the original transcript!
So GPT4 completely invented this unnecessary user need and pain point (presumably because the tool determined they were plausible and relevant), but the user in the video had not said either, implied either or hinted at anything related.
Halluciantions, help, and the Holy Grail
I’ve come to understand that these inventions are referred to as ‘hallucinations.’ Indeed, there are major source validity and hallucinatory problems with AI right now. Very convincing, compelling, engaging output that is driven by a wilful confirmation bias in users, who can see their effort being supported and pain being removed.
With such problems abound, AI output must be checked by human intelligence, rendering it pointless for the time being. Whether these are teething issues to be trained or programmed out or whether they are longer term issues, remains to be seen. One thing is for certain: AI tooling will help user researchers, and countless other roles, with all the heavy lifting that currently requires human intelligence. We will be able to outsource some of that intelligence to tools like GPT4, to support us to do our work. But crucially, you will not be replaced by AI. As the current cliche goes, ‘You’re more likely to be replaced by someone who has found an effective way to use AI reliably.’ That’s the holy grail. These are still the early days of AI, and as an immature technology it cannot be recommended for use in any meaningful work context -- at least in the UX space.