Calcified defaults
How many chatbots does it take to fix a coffee machine?
Of all the things I’ve turned to ChatGPT for, this was the strangest thus far: how to fix a coffee machine. Before you judge me, there were mitigating circumstances. First, it was the night before Christmas and nothing would be open the next day. Second, it was my in-laws’ 10-year-old coffee machine. Practically family. Third, I need coffee to function.
Amazingly, the bot recognized the brand and the make from a single picture, and walked me through a bunch of steps to diagnose what may be wrong. But ultimately, it threw its virtual hands in the air, and suggested I throw it out and get a new one. At ten years old, it was that machine’s time to go.
Two hours of poking around revealed a simpler and cheaper explanation, though. One of the valves had clogged because of calcification from hard water. We cleaned up the blocked artery, gave it a useless lecture on the need for a healthy diet, and were good to go once more.
The most likely fate of this machine would have been the landfill had we attempted to return it. Thankfully, the fix did not involve buying a new part or set of parts, because that’s a whole different challenge, because finding that part may have been impossible by design. This, after all, is why the right to repair movement has been slowly gaining ground, allowing owners and independent professionals access to parts to repair electronics, mechanical equipment and even agricultural equipment.
Why, though, did ChatGPT, so quickly suggest that we trade in the machine for a newer, shinier version? And what influence would it have on consumer expectations and behavior from now on?
Within its corpus of training knowledge, there probably are many more bits on trading in or returning a non-functional product than diagnosing this specific brand of coffee machine. In fact, it’s often easier for consumer brands to offer to ship a new product entirely than accept a return.
And therein lie two broader questions that have been with me since:
First, are we going to see a calcification of consumer expectations driven by LLMs? I confess that I did consider giving up on the machine and that was, in part, driven by my interaction with ChatGPT. As consumers converse with chatbots while they make purchase decisions, how much influence will such a bot have on consumer behavior and expectations? More precisely — and provocatively — will consumer behavior be influenced, if not determined, by preferences and tastes that are frozen in time?
Second, the flip side: are companies going to adopt at scale an application of LLMs called “synthetic users” or “digital twins” for market research? Although such user studies may be cheaper, faster and infinitely more scalable, how would one know if they are representative of the underlying market?
Within customer research, knowing who you’re talking to is just as important as what they say. Similarly, if I had Googled around for my coffee machine, perhaps I would have stumbled upon a blog or Reddit thread written by a helpful soul who specialized in repairing such a machine.
LLMs are a giant amorphous blob of a person. That’s the texture that’s invaluable in consumer behavior research, and may be hard to replicate with AI.