BIOTECHNOLOGY
Dual-use technology where the researchers themselves tried to pause
| DIMENSION | BIOTECHNOLOGY · 1975 — PRESENT | AI · TODAY |
|---|---|---|
| THE ASILOMAR MODEL |
THEN
In 1975, leading biologists voluntarily paused recombinant DNA research and convened the Asilomar Conference to establish safety guidelines BEFORE accidents happened. It worked — the guidelines prevented catastrophes for decades. |
NOW
The FLI Pause Letter (2023) called for a 6-month pause on frontier AI training. No lab paused. The Asilomar model failed because AI development is driven by corporate competition, not academic research. There's no grant funding to cut off. CI FACTOR: Corporate Race
|
| DUAL-USE DILEMMA |
THEN
Same techniques that cure diseases can create bioweapons. The Biological Weapons Convention (1972) bans bioweapons but can't prevent dual-use research. |
NOW
Same models that write code can write malware. Same models that assist drug discovery can design pathogens. No AI Weapons Convention exists. The dual-use problem is inherent to general-purpose AI. CI FACTOR: Incident Frequency
|
| GAIN-OF-FUNCTION DEBATE |
THEN
Should researchers deliberately enhance pathogen capability to study defenses? Years of debate, moratoriums, and eventual regulated resumption. |
NOW
Should labs deliberately push capability frontiers to study safety? The "capability research is safety research" argument mirrors "gain-of-function is defensive." Same tension, same rationalizations. CI FACTOR: Model Capability
|
Key lesson
Biotech shows the best-case scenario: scientists who recognized danger, voluntarily paused, and established governance. The AI community tried this and failed. The difference: in 1975, there was no billion-dollar commercial pressure to keep going.