What is your definition of ‘good?’
Happy 2025! After a longer than expected pause, the blog is back. One of my resolutions for the new year is to make regular posts. If you’re reading this and have an article idea, send it my way via the contact form on this site.
I came across an article this week that prompted some thoughts on defining what is “good” in the arts. Written for website and app designers, “What is Good Product Design in the Age of AI” discusses how tech professionals can work to create products that are grounded in “empathy and creativity.” Tech designers are working with technology that is still being trained, and as such, questions arise as to how a non-living work partner can help to produce a product that people find helpful.
Oscar Dumlao, a product designed interviewed for the article, commented on how he worked to train First Draft, an AI-driven application used to jump-start designs for all sorts of digital products:
We quickly realized that we couldn’t just hand off an all- encompassing set of design rules. Not only is it impossible to define every minutiae of what “good” is into something finite, but even if we could, token limits make it technically infeasible to feed AI a monster prompt of rules. Instead, we had to reduce the principles of good design to a small, but mighty, set of guidelines that hold generally true for any UI.
Dumlao continues his comments by drawing parallels between his work in training AI to complete certain tasks and his background in education:
This brought me back to my days teaching design. When you have to teach, you’re forced to deconstruct your own intuitive understanding into a few clear principles someone else can run with. The real challenge lies in finding rules that are clear, concrete, and instructive, while also being reliably true given any use case.
Attempts are underway to generate AI-based models that calculate subjective artistic characteristics such as musical expression. One such model, known as LlaQo, “leverages audio language modeling to provide detailed and formative assessments of music performances.” Currently, the problem with benchmarking products like LlaQo is producing an evaluation model that accounts for concepts such as melodic understanding and human expression.
Oscar’s comment on defining what is good in product design caused me to think about arts assessment in general. Using AI or not, how do we define what is good? What clear principles exist that define a good performance or artistic product? Could an AI bot ever be trained to assess performance in this way, and should it be trained to perform in this way?
Brad
