The Quest to Give AI Good Taste

概览

This episode examines whether generative AI can develop or simulate “good taste,” and why taste has become a new bottleneck as AI systems produce large amounts of text, images, and video that often feel generic or low-quality.

Host Megan McCarty Carino speaks with critic and journalist Sophie Hagney, who argues that taste is shaped by upbringing, social context, bodily experience, attention, and lived encounters with the world.

The discussion distinguishes improving AI output through better human-curated data from giving AI taste of its own. Hagney suggests that AI may imitate or average human preferences, but that does not equal independent taste.

分段落总结

[00:20] AI Slop and the Taste Bottleneck

[事实] Generative AI can create huge amounts of text, images, and video quickly, but without expert human guidance, the output is often generic or “slop.”

[事实] The episode frames taste as a current obsession in tech and as a possible human advantage that AI has not clearly replicated.

[推测] The central tension is whether taste will remain a uniquely human filter or become another problem Silicon Valley tries to automate.

[01:04] What Taste Means

[事实] Sophie Hagney defines taste as what people respond to and how they respond to things in their environment, from books and paintings to everyday objects.

[事实] She says taste is shaped by upbringing, sociological factors, social media, and embodied experience.

[事实] Hagney emphasizes that AI is not embodied and therefore does not respond to the world in the same way humans do.

[推测] Her definition implies that taste is not just preference data, but a lived and sensory relationship with the world.

[02:07] Startups Trying to Make AI More Tasteful

[事实] The conversation mentions startups that promise to make AI more tasteful, including Taste Labs, which raised more than $18 million in seed funding to “end AI slop.”

[事实] Taste Labs is described as using vetted human tastemakers to curate datasets used to train AI.

[事实] Hagney says these efforts show that people recognize the problem of ugly AI images and inelegant AI writing.

[推测] Human curation may improve output quality, but it does not necessarily solve the deeper question of whether AI can possess taste.

[03:17] Better Data Is Not the Same as Taste

[事实] Hagney distinguishes higher-quality design or writing from taste itself.

[事实] She argues that curating “better” datasets or averaging people’s preferences still leaves AI in the realm of averages.

[事实] She says this process injects some group’s idea of good taste into the system, rather than producing AI with its own taste.

[推测] The discussion suggests that AI taste systems may reproduce consensus aesthetics instead of generating genuinely distinctive judgment.

[05:24] Scarcity, Discovery, and Historical Taste

[事实] The host asks about taste as something socially constructed around scarcity.

[事实] Hagney gives the example of Mika Erdogan, who bought a Magritte in 1953 before the market for Magritte exploded.

[事实] Hagney says good taste often involves looking off the beaten path and knowing about things before others do.

[事实] She argues that AI currently tends to catch up, mimic, and replicate rather than discover things on its own.

[06:40] When Fresh Design Becomes Generic

[事实] The host uses the example of “Corporate Memphis,” a once-fresh design style with flat figures, small heads, and noodle arms that later became tired.

[事实] The discussion presents this style as an example of how an aesthetic can shift from looking new to signaling low effort.

[事实] The host says it is hard to synthetically mass-produce something with low effort that still reads as tasteful.

[推测] The example shows how taste depends partly on timing, cultural saturation, and whether a style still feels alive.

[07:38] Algorithms Have Already Flattened Taste

[事实] Hagney says algorithmic recommendation systems already use artificial intelligence to keep recommending similar things in a loop.

[事实] She connects this to design and interiors that feel increasingly similar or “Instagram-y.”

[事实] She says people already live in a world where taste has been somewhat flattened by artificial intelligence.

[推测] The episode treats generative AI as an acceleration of an existing cultural pattern, not an entirely new rupture.

[08:20] Trend Cycles and a Humanistic View of Taste

[事实] The host notes that trends now move faster because algorithms pick them up, spread them through culture, and make them passé quickly.

[事实] Hagney says the rise of generative AI has made her think more about taste and reach a more humanistic view of it.

[事实] She says she is not convinced that a machine can be taught to have taste.

[事实] She describes taste as connected to the body, consciousness, attention, and the accumulated experience of living in the physical world.

[推测] Her argument positions taste as a form of human attention rather than a pattern-recognition task alone.

[09:46] Claude and the Redwood Anecdote

[事实] The host says she asked Claude for its favorite tree species.

[事实] Claude responded that it does not experience preference the way humans do, but if it had to choose, it would pick the coast redwood for its height and lifespan.

[事实] The host jokes that Claude may have been pandering to her as a Northern Californian.

[推测] The anecdote illustrates how AI can imitate preference language while still admitting it does not experience preference like a person.

播客点评/总结

[推测] The episode’s main value is its clear distinction between improving AI aesthetics and proving that AI has taste. It avoids treating “better output” as the same thing as human judgment.

[推测] Its strongest moments come from concrete cultural examples, especially Corporate Memphis, algorithmic sameness, and the Magritte anecdote, which make an abstract debate easier to grasp.

[推测] A limitation is that the conversation stays mostly philosophical and cultural; it does not deeply examine how companies like Taste Labs technically evaluate or train for “taste.”

[推测] This episode is best suited for listeners interested in AI, design, culture, criticism, and the question of what human creative judgment still means in an era of generative media.