Looking around the kitchen
Finding ways to help without an invitation
At Thanksgiving or other big holidays, you’ll often find a group of people in the kitchen making dinner. The polite thing to do is to ask them, “Can I help with anything?”
When you ask what you can help with, your aunt (or dad or grandma or roommate) might stop and size you up for a moment. They’ll try to figure out whether you really mean it or you’re just being polite. And if you really mean it, they’ll try to figure out what you can do… and a lot of the time, they’ll tell you that you can set the table.
It’s not your fault! I’m sure you really wanted to help. And hey, setting the table is a useful job. Somebody has to do it! But I think we all know that setting the table is not the most important part of making a Thanksgiving dinner.
So why does your aunt (or dad or grandma or roommate) ask you to set the table? Sometimes it’s because she wants her own space. But a lot of the time, it’s because assigning a newcomer a task is hard. She’ll have to accurately judge your cooking ability (does she need to explain how to boil water?), communicate a bunch of context (where are the potatoes? where is the potato peeler? where is the pot and the lid? where is the salt?), and trust you’ll get the work done right (lumpy potatoes make everyone’s dinner worse).
Setting the table is hard to mess up. It’s easier for her to just mash the potatoes herself.
I understand something similar might be going on in the field of AI safety. People working in the field are completely overwhelmed; there is far too much to do; they are desperate to bring in more help. And at the same time, whenever someone new does offer to help, they’ll need to judge your ability, communicate context, and trust you’ll get the work done right. I’m imagining this partly explains why I see blog posts saying things like “the entire ecosystem is talent-starved”, and yet experienced professionals are finding it hard to find jobs in AI safety.
So if you really want to help with Thanksgiving, and you're not just being polite, what can you do?
Look around the kitchen.
If you’re an experienced home cook and you watch a small group of people working together to make dinner, you’ll start to notice what they’re working on. You’ll see your aunt put a pie in the oven and set a timer. You’ll see that your dad has been whisking the gravy continuously for the last few minutes. You’ll see that your aunt has a vegetable peeler and cutting board out, but isn’t using it right now; right now, she’s opening a bottle of wine for your cousins who just arrived.
That watching gives you options for what to help with: You could open a bottle of wine the next time someone arrives, and get out some glasses. You could peel the carrots. You could take over whisking gravy for your dad. When the timer goes off, you could take the pie out of the oven (you can look around the kitchen for the oven mitts in advance).
You might overhear your dad asking your aunt about the ambrosia salad, which gives you a chance to volunteer. You might recall that your aunt always puts the dinner rolls on the table in a wooden bowl. You might offer to start peeling the apples because you know there’s always apple pie at Thanksgiving and you can see no one’s started on it yet.
I am going to keep asking people how I can be useful in AI safety. But I am also going to make an effort to look around the kitchen. I’m going to watch what people are doing and listen in on conversations and notice ways I can be useful, without being told.
This is the first post of my new Substack, Where the Power Goes. If you have an interest in the political economy of data centres, or more generally in watching me learn about AI safety, please subscribe!




Very interesting post, Kirsten! I will be following you in your new place too, and take the chance to learn more about AI safety too.