Artificial Intelligence

Tim's AI Background

Tim holds a graduate degree in AI from 1989, making him (by his own admission) 34 years behind current developments. However, he notes that neural networks existed back then — what's changed is sheer scale and training data quality. He views modern LLMs through the same lens he viewed neural nets in the '90s: powerful tools with a fundamental opacity problem. You feed them input, get output, but tracing how they reach their answer through billions of connections is humanly impossible.

The Force Multiplier Philosophy

Tim's former producer Eric DeMille used the term "force multiplier," and Tim applies it directly to AI. The core idea: AI shouldn't replace humans, it should amplify them. One artist making more art, more detailed art, or rapid concept art before the creative direction is even clear. The human still needs to understand the intent — an LLM trained on adjacent material won't produce originality, just variations on whatever themes it was fed.

Specific Applications in Game Development

Art and Visual Assets

• Concept art is the ideal early use case — quickly exploring styles ("What if Art Deco? What if '50s sci-fi novel covers? What if Atari VCS painterly covers?") before committing to a direction
• 3D model generation for initial assets, or AI-assisted detail passes after a human creates the hero asset
• Variation grinding — every artist dreads being told "great leather armor, now make 10 more variants." AI can handle the mind-numbing color/detail permutations

Narrative and Dialogue

• Procedural narrative similar to what Tim attempted in Arcanum, but vastly more capable
• Main characters should stay human-written, but the hundreds of side NPCs who only say "What do you want to buy?" or "What are you doing in my house?" are perfect AI candidates
• Side quest generation expanding on Bethesda's radiant quest system — fetch quests, escort missions, delivery tasks generated appropriately for the player's level, build, and current map, with AI-generated dialogue for the creatures and NPCs involved

Level Design

• Procedural dungeon/cave generation trained on making small explorable spaces and instances
• Tim's vision: every game could have Skyrim-scale exploration (hundreds of locations) while being built by much smaller teams

Code Assistance

• Tim was initially skeptical but has been impressed by AI code generation helping humans with specific tasks: sorting algorithms, data structure recommendations, search optimizations
• The pattern is always human-directed: "How should I store this data for fast lookups?" — and the AI suggests appropriate structures

The Bigger Theme

The thread running through all these applications is the same: bigger games from smaller teams. Not firing people — still wanting large teams for AAA projects — but enabling indie developers and small studios to build games that currently require a Blizzard, Bungie, or Epic.

Beyond Games: Music Streaming Rant

Tim pivots to a personal frustration with music streaming AI. Despite years of listening data, his streaming service still can't figure out basic preferences: he hates live versions, remastered tracks, and censored songs. It once repeatedly played "Grandma Got Run Over by a Reindeer" despite him explicitly saying he hated it. He wants context-aware recommendations — office music vs. car music vs. bathroom music, factoring in time of day and day of week. He's especially frustrated that no service has ever surfaced an entirely new genre for him; he had to discover witch house on his own. Tim sees this as a prime area where modern AI could dramatically improve.

Concerns and Optimism

Tim's two biggest worries about AI:

1. Political misinformation — AI-generated disinformation at scale
2. Advertising saturation — noting that every human communication medium (speech, writing, TV, radio, internet) eventually gets "jammed full of advertisements," and AI will be no different

Despite these concerns, Tim describes himself as "cautiously optimistic" about AI's potential.