Chris Lengerich October 31, 2023[^1] A close analogy for broad AI licensing is probably something like chemo to target cancer cells. Yes, chemo might be able to stop a threat model of rapidly proliferating, resource-seeking, adversarial threats. But it also can kill you. Chemo kills *all* fast-dividing cells. It suppresses the immune system which needs to be able to recognize threats (via experimentation) and scale up against them (via funding). That suppression allows for other opportunistic threats to invade. On chemo, you can die of a common cold. The better medicine is to upgrade your existing immune system against the specific cancer type that you're facing by adding additional context to your therapy. This is what we do with T-cell immunotherapy, designed by scientists with higher-context episodic memories than our immune system. With that context expansion, T-cells can hunt down and destroy much more specific cancer types without resorting to carpet bombing anything that moves. In the Internet immune system, this medicine is equivalent to building and testing long-context personal AIs to verify content. It also means improving the web-of-trust protocols that underpin search and saying no to chemo-like broad AI regulations that suppress freedom of speech, and offer the chance for an opportunistic takeover by a rogue political cell. NB: Metaphors aren't perfect (be suspicious by default of simple analogies, although some are good). Some axes of variation re: chemo: Delivery mechanisms: Delivery of untargeted chemo in the body is relatively easy (via the circulatory system) but in a distributed economy is hard. The body works much more closely together than most economies do right now, for better or worse. For AI licensing, therefore, there’s an additional challenge of compliance. Evidence: We haven’t tested broad AI licensing for general-purpose language tech. Jumping directly to licensing would be like applying an untested, unregulated chemo drug, useful only in the case of rapidly approaching death when other treatments have failed. While we should have empathy for the initial emotional reaction of wanting to reach for a risky chemo treatment, the more rational response is building precision medicine which can treat actual problems - both known and emergent - without suppressing innovation or opening the opportunity to hostile takeovers. That’s what many of us who helped kick off this LLM wave are in fact working on now - an immune system for the Internet, powered by precise and open-source AI that works for you. [^1]: Post originally appeared @ https://twitter.com/chrislengerich/status/1719479090321793298