Three thoughts from this past week:
AI and Innovation
AI and Understanding
UX Design and Data
AI and Innovation
There is a line from the excellent show about the early days of the personal computer, Halt and Catch Fire, where one of the main characters says:
"Computers aren't the thing. They're the thing that gets us to the thing."
I think this line applies to AI as well.
In this excellent talk that has stuck with me since 2013, the founder of Asana, Justin Rosenstein, talked about their project management software not being something that changes the world. But that it is something that helps the people that do, do so more efficiently.
What excites me most about AI is that it is allowing humanity as a whole a faster pace of innovation. At that great scale it may be barely perceptible now. However, the thing about innovation is that it builds on itself.
AI doesn't solve every problem for everybody today. And, as long as humans desire the smile of another, I don't anticipate that it ever will. But it does help people solve problems faster today. That allows them to move onto the next problem faster. And so forth. Those small steps put together is what unlocks discovery.
AI isn't the end goal. It is a tool for us to use to achieve that--whatever it may be.
AI and Understanding
With AI there seem to be two modes: teacher mode and student mode. The AI assuming the role of the teacher when I am chatting with it and asking it questions. And it taking the role of student or trainee when I am training it to complete a task.
It is said that one needs to be able to understand a problem completely in order to communicate it simply. This is limitation on my own use of LLMs where the more I understand the problem, and what direction the solution lies in, the better I am able to communicate that and get the result I want. When I don't (understand), I don't (get results). This is no different than with people and I am not sure why some commentators expect LLMs to read minds in the way that we don't expect each other to.1
Understanding to me seems to be holding a data model of a domain in your head and interacting with that model with logic. The data model is refined as the logic fails. Once the logic is no longer failing, the domain is understood.2
My conversations with AI today don't move naturally between teacher and student mode as well as they may with colleagues. In fact, right now if I have a question while in teacher mode, it is usually better to launch a new conversation and ask that question separately rather than pollute the context of the current teaching.
This is in the minor inconvenience category.
UX Design and Data
A huge part of the data integrity battle is getting users to input data. And the more friction there is the more likely they will require help to input it, will input it wrong, or won't input it at all.
Thus I found this thought experiment interesting: No tech support
If you were never allowed to provide tech support, in any form, what would have to change?
How would on-boarding need to be improved, to the point where customers would self-serve and be happy doing it? This might be the hardest step, because the customer is least-familiar with your product, and least-motivated to power through barriers to their success. And setup stuff often has to be done only once, so you might not learn how to improve those things from existing customers.
Where would the product provide the user with more control, since they can’t ask Support to do it for them? What information requires better visibility, since they can’t ask Support to look it up for them? What actions would need to be become more intuitive, because they can’t ask Support how to use functionality that they know “is in here somewhere?”
If you're fighting a data integrity battle (and who isn't these days?) then taking a hard look at your UX should be part of your battle plans.
I have been working with humans for thirty years and AI only for a couple. I am giving my opinions on AI a little more time to mature.
At a given level. Mastery being a level of understanding.