liveBook · Manning

Complexity bias

As explained in Machine Learning System Design

A tendency to give undue credit to or prefer complex ideas and systems over simpler alternatives.

Complexity bias is the tendency to give undue credit to complex ideas and systems instead of simpler alternatives. Complexity can signal effort, mastery, innovation, or more features, even when it adds little practical value. section 8.1

A baseline counters this bias by keeping the initial system simple and agile. It provides a reference point for judging later changes without assuming that sophistication automatically improves outcomes. section 8.1

Treat complexity as a decision to justify, not a default signal of quality:

  • Compare complex proposals with a credible baseline before investing in additional features or modeling sophistication. section 8.1
  • Skip a baseline when accuracy is life-critical, requirements are already certain, or an existing system has been proven in production. section 8.1
  • Optimize an already valuable system only after recognizing that its working version already serves as the baseline. section 8.1

How Machine Learning System Design covers it

How deep it goes
Defines the bias, identifies its social signals, and gives baseline-related exceptions without formal measurement methods.
Examples
Uses cancer detection, autonomous cars, manual control, and search-engine model replacement as contrasting cases.
What you can do after
Evaluate complex proposals against simple baselines and recognize when safety, certainty, or an existing system changes that rule.
How it connects
Connects complexity bias with baselines, manual control, deep semantic similarity models, and Transformer-based models.
Cover of Machine Learning System Design
liveBookmanning.com
Read Machine Learning System Design on liveBook Get the book at manning.com

This explainer was generated by Manning's knowledge graph from the text of Machine Learning System Design. The section links point to the parts of the book it draws on.