Thinking tool: Ask if the answer changes the decision
When choosing whether to research something, consider whether the decision actually changes depending on the answer.
Medicine has a rule of thumb: don't order a test if the result won't change the management plan. For example, doctors don't image low back pain in the first six weeks unless red flags are present. This is because the plan (stay active, manage the pain, wait) is the same whatever cause the scan shows.
Examples of doing this poorly I've observed:
-
Arguing over the precise definition of a frontier AI system. There's some use in making sure that people talk about roughly the same things, and sometimes if you're considering legislation it might be important to set a particular scope. But most of the debate I have been involved in seems detached from any real purpose and is a huge time sink - we all know what we're basically talking about is the LLMs that Anthropic, OpenAI and Google put out. And it's the "what do we do about them" question that is much more important and more urgent.
In a previous job, I spent about an hour with 100+ OECD delegates arguing whether a thermostat should be considered an AI system. -
Arguing about what a self-driving car should do in the extreme case where it can choose between killing different people. This basically never comes up in practice, and the decision doesn't change either way: you try to make the car not hit people at all, and we generally don't develop self-driving cars in a way that even lends itself to considering these factors.
Related concepts
This is part of my thinking tools series. Also consider:
- Refuse the idiot plot: if people are stuck on useless debate, you need to be the person to say this obvious thing out loud