I would like to propose a constructive use of ChatGPT that arose directly from a real troubleshooting experience.
I recently used ChatGPT to help troubleshoot the setup of an IKEA lighting system. The product ultimately worked, but the setup process was unnecessarily difficult. An unexplained symbol was used for pairing, there was no obvious indication that the remote itself was functioning before pairing, reset and pairing states were difficult to distinguish, and even basic operation such as how to dim the lights was not apparent from the controls.
ChatGPT eventually helped me solve the problem. During that conversation, however, I raised a much more interesting question:
How many other people have asked ChatGPT essentially the same question about the same product?
Individually, these conversations are simply troubleshooting. Aggregated and anonymised, however, they could become an extraordinarily valuable source of real-world information about how consumer products actually perform in the hands of ordinary people.
ChatGPT could potentially identify recurring patterns such as:
- products generating unusually high numbers of troubleshooting conversations;
- common setup and commissioning failure points;
- confusing controls, symbols and terminology;
- users being unable to determine whether a component is actually functioning;
- repeated reset, pairing or configuration failures;
- recurring deficiencies in manufacturer instructions;
- significant amounts of consumer time being wasted; and
- products apparently being returned or replaced when they may not actually be faulty.
This has implications beyond inconvenience. Poor product usability can generate unnecessary technical support, product returns, freight, packaging, replacement products and ultimately electronic waste.
I am not suggesting that individual ChatGPT conversations or personal information should be provided to manufacturers or other organisations. The valuable information is the aggregate pattern. Privacy, anonymisation and appropriate thresholds would need to be fundamental to such a system.
OpenAI could potentially develop a privacy-preserving real-world usability signal from troubleshooting interactions. Significant recurring problems could be identified and, where appropriate, anonymised aggregate findings could be made available to manufacturers, standards organisations, consumer-protection bodies and environmental regulators.
There is also a broader regulatory principle worth considering:
A consumer product intended for ordinary consumer use should be capable of initial setup, basic operation, reset and basic fault identification without specialist product knowledge or reliance on external information. Non-standard symbols should be accompanied by intelligible labels or instructions, and products should provide sufficient feedback for a user to determine whether major components are functioning.
This need not be subjective. Products could actually be tested using representative first-time users to determine whether ordinary consumers can perform these basic tasks successfully.
There could also be a useful ChatGPT feature arising from this. After resolving a recurring consumer problem, ChatGPT might offer the user an option such as:
“Submit this as systemic product feedback.”
ChatGPT could then prepare a structured report, show the user exactly what information would be contributed, remove personal information, obtain explicit consent and add the resulting information to an aggregated dataset.
More broadly, I think this illustrates a positive application of AI that receives relatively little attention. AI should not merely help the millionth person work around the same poor product design. If privacy can be protected, it should also be capable of recognising that large numbers of people are encountering the same problem and turning that collective experience into useful knowledge.
There is an interesting aspect to this particular proposal: the proposal itself emerged through my conversation with ChatGPT.
I identified the broader implications while we were troubleshooting the product, and ChatGPT helped me examine the idea, develop it and turn it into the structured proposal presented here. The process therefore demonstrates part of the opportunity I am proposing: an ordinary support interaction developed into recognition of a potentially systemic problem and then into a possible solution.
I am submitting this as my proposal, developed with the assistance of ChatGPT, and would be very interested to know whether OpenAI has considered using aggregated troubleshooting interactions in this way, or whether the concept could be passed to an appropriate product or research team for consideration.
Eric Nielsen
Australia
ChatGPT Plus user