Want to Augment a Product’s Aesthetic Design? Use Machine Learning

This content relates to : DIGITAL TRANSFORMATION

Based on the research seminar at the Snyder Innovation Management Center by:

John R. Hauser, MIT

John Hauser gave a seminar on April 14, 2021, describing joint work with Alex Burnap (Yale) and Artem Timoshenko (Northwestern University). In the seminar, Prof. Hauser noted that aesthetics is consistently ranked as one of the most important factors in product choice. This is particularly true in the automotive industry where aesthetics can explain up to 60% of the uncertainty in purchase decisions for certain segments. Not surprisingly, automotive firms invest heavily in design and redesign – $1.25 billion on average with up to $ 5.7 billion in critical designs. He stated that traditionally, aesthetic design has been driven by human judgment in at least two ways. First, the aesthetic designs are typically generated and screened by design teams that have an “eye” for visual design. Second, human judgement influences aesthetic designs via consumer evaluations. In the automotive industry, consumers evaluate and rate aesthetic designs on established benchmarks such as sporty, innovative, and luxurious at “theme clinics.” These theme clinics are expensive; automotive firms typically invest as much as $100,000 per theme clinic for a single new vehicle design. With multiple designs per vehicle and several vehicles in their product mix, automotive manufacturers can spend millions of dollars on theme clinics.

Against this background, he and his co-authors propose a methodology to improve the process of aesthetic product design and testing. Specifically, they show that machine learning can augment both the generation and testing aspects of aesthetic design in the automotive industry. For generation, they developed a generative model that creates new product designs that are predicted to be aesthetically appealing. This gives the design team a controllable tool that they can use to explore consumer aesthetic perceptions. In one proof of concept, the authors demonstrate that a model trained using model year 2010-2014 data can generate automotive designs introduced in the 2020 model year.

In further tests, the authors predict how consumers would rate aesthetic appeal directly from visual images. First, they used an encoding model to represent visual designs using 512 abstract features; next, they trained a predictive model that predicted aesthetic ratings based on these features. The predictive model helped screen proposed aesthetic designs to ensure that only the highest-potential designs were tested in the theme clinics. They conclude that machine learning can augment human judgement and lead to better aesthetic designs.

Author: 

To learn more, read: 

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3421771