Deep Learning

The Concept of Deep Learning

Tim Cain borrows the term "deep learning" (unrelated to machine learning) to describe something you've learned at such a fundamental level that you literally cannot turn it off. He first encountered the concept in 1987 at UCI, where his graduate school roommate Larry — a neuroscience PhD student — introduced him to the idea.

Larry's demonstration was simple: he wrote a word on paper and asked Tim to see it only as a collection of lines, not as letters or a word. Tim couldn't do it. Having learned to read at a young age, the pattern recognition was permanently baked in — he could not not read the word.

The Brain Age Example

The Nintendo DS game Brain Age had a perfect illustration: words for colors (like "BLUE") displayed in a different color (like yellow). Players had to say the color, not the word. It was voice-activated and timed. The deep learning of reading made it nearly impossible to ignore the word and just see the color — a phenomenon known in psychology as the Stroop effect.

Things You Didn't Learn — You Were Born With Them

Some deep responses aren't learned at all — they're evolutionary. Babies are frightened of spiders and snakes even in countries where these creatures don't exist. Evolutionary biologists explain this as a survival trait: babies who feared venomous creatures survived at higher rates, so the instinct became hardwired.

Uncanny Valley as Evolutionary Residue

Tim offers a provocative explanation for uncanny valley — the revulsion people feel toward faces that look almost but not quite human. Since Homo sapiens once coexisted with other human species (Neanderthals and others), we may have evolved an instinct to distinguish "our kind" from "not quite our kind." Even though those other species are long gone, the instinct persists, which is why near-photorealistic faces in games and AI-generated images trigger discomfort.

Manufactured Deep Learning: Marketing

Some deep learning is entirely manufactured by culture:

• Gendered colors — The idea that pink is feminine is completely artificial, not innate, and varies across cultures.
• Gendered fragrances — Floral scents are marketed as "for women" and leather as "for men," which is arbitrary. Tim mentions owning a fragrance (Molecule 01 or 02) that smells different to different people, making gendered categorization absurd.
• Gendered food sizing — Five Guys names their regular-sized burger "Little" and the oversized one "Hamburger," knowing many men can't bring themselves to order anything called "little." A clever marketing trick that makes people spend more money on more food than they actually want.

The Application to Games

This is the payoff: many gaming preferences that feel like conscious choices are actually deep learning at work. Players dismiss entire genres — turn-based combat, retro graphics, certain art styles — believing they've made a rational decision, when in reality:

• Someone may have made fun of them for liking a feature when they were younger
• Cultural messaging taught them certain types of games aren't "for them"
• Surface-level reactions (like uncanny valley in older 3D games) created lasting aversions

Tim acknowledges some reactions are innate — Obsidian had to add an arachnophobia mode to Grounded because spider phobia is genuinely hardwired. But most gaming preferences aren't innate; they're learned.

The Challenge

Tim's call to action: just like the Brain Age game forces you to look past the word and see the color, try to look past your ingrained biases about games. If you've dismissed an entire class of games because of genre, art style, or combat system, you're probably missing out. The only person suffering from that closed-mindedness is you.

References

• Tim Cain. YouTube video. https://www.youtube.com/watch?v=K8nYj-wdaHE