About half of Americans, when asked, report that they don’t read the user manual for new technical devices they acquire. Although I suspect that many people are like me – I read them sometimes, and then only partly. If there is a “quick user guide” I will often look at that. These are helpful quick summaries with just the critical bits of information you need. But often I give it a go by myself and then only consult the guide for troubleshooting. I do this partly because I want to see how intuitive the device or app is.

But of course there are situations where this approach is not adequate, especially when assembling something complex or using an entirely unfamiliar and complicated piece of technology. In these situations I find, most of the time (with a few happy exceptions) that the instructions are terrible. Sometimes they were clearly written by someone for whom English is a second language. Or the instructions are entirely pictoral (I guess to be language neutral). Other times it seems they were written by engineers who lack the ability to communicate effectively their arcane craft to the general public. I think this is why many people try to bypass the instructions – they are often terrible and frustrating.

In industry, however, this option may not exist. Further, the instructions may need to be highly technical, which is great if you are already an expert but may be challenging for most users.

This is where artificial intelligence enters the picture. Large language model AIs, like Chat GPT, can “read” material and then answer questions about that material, or even give a summary. I have used Chat GPT to analyze a scientific study, then asked it to find specific information within the study or explain certain findings, and it does pretty well. The idea, therefore, is to feed an entire technical user manual into an LLM and then ask it specific questions or have it give a summary or perhaps step-by-step guide.

There are also at least two companies who have made AIs specifically designed for this task (rather than using an all-purpose LLM). One is Aveva who has made an LLM that can read a technical manual and then act as a resource for a user. It can work you through a process, suggest troubleshooting options, or answer specific questions. The idea is that this will be much faster than looking through a manual that might be hundreds of pages long. It will also be more convenient than having to access a more experienced engineer for help.

Another product, CreatorPro by Dozuki, creates the user manual itself. An engineer can make a video of themselves walking through a process, such as assembling a piece of equipment, and then software will make a user manual from the video.

I have not used either of these products so I cannot review how they actually function. Since they are dedicated to this specific task I would assume that they are better at it than Chat GPT or other all-purpose LLMs, which are already pretty good. This is exactly where I hoped the AI community would be going – using the basic technology to develop specific applications. For example, it would be great to have an AI app designed for medical use, actually many medical uses. There are many already is preliminary use or in the works, but I can imagine a future where AI would read most diagnostic studies and present their findings to the physician – EKGs, MRI and CT scans, EEGs, Ultrasounds, etc. AI is good at being thorough, while humans have a tendency to miss stuff.

Also, reading some studies can be tedious. As a fellow one of my tasks was to listen to 24 hour EEG recordings (at fast speed) to listen for any seizure activity. I had to pay attention for hours to go through the recordings for the week. Now we already have software that can automatically detect probable seizures, but the newer AIs can likely do more thorough analysis. This would then open up the possibilities for diagnostic tests that generate lots of data – too much to be practical for humans to go through, but AIs can.

Getting back to user manuals, I would love to see a near future in which I never have to crack open a user manual again. A new purchase will simply come with a URL or QR code leading to an AI app that will be an interactive user manual. An AI expert assistant could walk you through any task, answer questions, and show you helpful images that it can also explain as necessary. Further, it would likely be possible for such an AI to be trained by every interaction with a user, so it gets better at anticipating mistakes and misunderstandings.

While this would be great, I do think it would add to the general trend of humans outsources our thinking to software. This is not necessarily a bad thing, but it is interesting to think about. Prior to ubiquitous computers, I had to actually remember phone numbers, and know how to navigate a map. Now I just do whatever the GPS tells me to do.  Make not mistake, I would not want to go back to a pre-GPS world. But I do think we need to pay attention to and be thoughtful about unintended consequences from overreliance on AI assistants. This theme has been explored in science fiction, depicting an end-stage of completely infantile and dependent humanity utterly helpless without their AI/robotic caretakers. Sometimes it is a good experience to go through the frustration of having to figure something out on your own.