Can Machines Learn Taste?

Taste has become a buzzword in Silicon Valley, appearing in pitch decks, investor theses, product manifestos, and job descriptions. Companies like xAI, Meta, and Palantir have started incorporating "taste" into their branding and hiring practices. This shift in focus has led to venture capitalists showing interest in companies that can objectively evaluate and improve the quality of AI-generated content.

Anu Atluru, a tech essayist, has noted that "Taste is Eating Silicon Valley," and Greg Brockman, OpenAI's president, has declared it a "new core skill."

Paul Graham, co-founder of Y Combinator, believes that in the AI age, taste will become even more crucial as it becomes the differentiator for successful startups. Thais Castello Branco, the founder of Taste Labs, aims to establish a standard of quality for AI-generated content by pooling a network of expert tastemakers and creating data and tools that can be used by AI models and applications.

Her definition of taste is "the bar of quality in the absence of correctness," which is a compressed judgment learned from pattern recognition and opinionated decision-making.

While taste is often considered a subjective preference, it can be trained and developed, as both Paul Graham and Thais Castello Branco argue. The philosopher David Hume, who explored the concept of taste in the 18th century, recognized that beauty is a subjective sentiment but still believed that a consensus of the most qualified critics could serve as the true standard of taste and beauty.

In the context of AI, Hume's definition of taste and its standard can be applied to create a collective human judgment that serves as the benchmark for evaluating the quality of AI-generated content.