A photo of the Divisumma 24 with its cover removed, revealing complex internals. The complex Olivetti Divisumma 24. Photo credit: Hannes Grobe via Wikimedia Commons (CC BY-SA 2.5)

In “Simplicity is An Advantage but Sadly Complexity Sells Better”, Eugene Yan argues that “complexity sells”, even when simplicity would do just as well (or better). He points out a few reasons for this dynamic:

  1. Complexity signals effort
  2. Complexity signals mastery
  3. Complexity signals innovation
  4. Complexity signals more features

These dynamics are absolutely true for contemporary machine learning research and technology. Funders (and reviewers) like to see complex, “cutting-edge” approaches.

I agree with Eugene Yan:

The objective should be to solve complex problems with as simple a solution as possible.

I think this dynamic reflects the natural push and pull between the fields of machine learning and human-computer interaction. As Jonathan Grudin argued all the way back in 2009, “When AI was ascendant, HCI languished; during AI ‘winters,’ HCI thrived.”

We’re currently in an AI summer, where the appealing aspects of complexity are apparent. When progress slows and expectations are brought in line with capabilities, the attention will return to a focus on design: solving well-scoped problems with just the needed complexity.