It happened again… I have an idea for a project. It’s a super cool project. I’d love for it to actually happen.
The only problem is… I don’t have time to do it myself, and I want credit. But I don’t see a way to get it — well, not enough of it, at least. Having this experience a lot is one of the reasons I started Evitable: When I have a cool idea, I want to be able to execute it in house!
As I was writing this, I realized I should share the example of the Statement on AI Risk, widely regarded as the most significant piece of public comms about AI existential risk ever, and which I initiated:
The media emphasized the AI CEOs’ signatures, misleading many into thinking of this as corporate PR, but over 100 AI professors also signed, including the two most cited scientists of all time who helped pioneer the techniques powering modern AI.
It still feels awkward to be aggressively claiming credit here, because the Center for AI Safety (CAIS) really took the idea and ran with it, shaping the vision and doing the vast majority of the execution. But I realized recently how few people know I had anything to do with it at all! This is despite me being mentioned on CAIS’s website, in the press release, and in media appearances. So I am trying to toot my own horn more these days.
If ideas are cheap, why don’t you give me yours?
This tension around how to divide credit between “ideas” and “execution” is a classic conundrum in research. A common saying is “ideas are cheap”. They say something similar things in start-up land:
There’s another complication with ideas: people can always claim, after you tell them your idea, that they’ve already had it.
And people who think ideas are cheap do often say things like: “Obviously a bunch of people have had ‘your’ same idea”. But in my experience, most people don’t actually act like they believe this. I’ve seen the same researchers preaching “ideas are cheap” jealously guarding their idea: “Well, I actually am working on it and want to get credit...” seems to be the attitude when it’s your idea. Researchers, like journalists, live in fear of being “scooped”.
Why don’t people believe ideas are valuable?
What’s going on here? I think it’s an example of people acting like the way things work as fair, even though it’s clearly not. I think this happens often because people think of fairness as more about getting what you can reasonably expect to negotiate for, rather than what you in fact deserve.
It’s hard to get value out of having an idea if you don’t execute it yourself, but that doesn’t mean it’s not valuable. When people do have a great idea and manage to execute on it, this can make them wildly successful. It’s true execution matters, but if you can do the execution yourself it seems like you clearly can get way more value out of it. We’d probably be in a much better situation if people were able to share ideas more freely without losing this value. Sharing an idea creates a positive externality, and that’s hard to capture.
I’m an “idea guy”… what should I do?
It’s also possible sometimes to get a lot of credit for becoming known for an idea publicly, and people sometimes act like (again) like this means the credit assigment system isn’t broken. But, to do this you kind of have to make it your job to get credit for the idea, or, better, your ideas in general. Like you need to build a brand for yourself. I do think it’s generally true that if you’re not in a position execute on an idea (and won’t be), often you may as well blast it publicly and get as much credit as you can that way, even if it’s “pennies on the dollar”.
In research, an approach I’ve found that works is: do a bit of work, enough to generate momentum, and then tell people “I’m doing this. It’s going to be great, and here’s why! Want to help?” I did this for a lot of my papers in grad school, which I cobbled together last minute. I’d run a proof of concept experiment, and find people who were eager to get their name on a paper in exchange for running another one of the experiments we needed. This kind of approach is how I ended up with a publication having the longest list of authors among all the papers at the International Conference on Machine Learning in 2017 (“A Closer Look at Memorization in Deep Networks”).
But I’ve also heard horror stories of well-resourced teams hearing about a junior researcher’s idea explicitly threatening to “scoop” it from them. Reflecting on this example, I’m imagining a start-up telling a tech giant “We’re doing this cool thing! Want to help?” I don’t think that ends well for the start-up, unless they’re ready to be acqui-hired. Business people: weigh in here!
Oversharing as a solution?
A friend I know has what I think is a good approach: after generating momentum, send a document about the project to a few dozen colleagues as a “heads-up”. This seems wildly risky from the “getting scooped” perspective, but in practice seems to mostly generate buzz.
Actually, I did something sort of similar with the “CAIS Statement” as it’s typically referred to: I wrote to a couple dozen people I knew and told them I was probably planning to circulate such a statement, but also “it seems like the details could matter a lot in getting good uptake, but I don't think I am the right person to figure out the right strategy... It seems more like a job for marketing / political campaign experts.”
At some point, CAIS decided unilaterally to do it. I had mixed feelings because I figured they’d execute well, and I mostly just wanted the thing to happen and didn’t feel like I could claim permanent ownership of the idea. But also, I’d lost control over it, and while they kept me in the loop, I didn’t have the final say anymore. I had wanted it to be a letter from independent academics, to avoid the “corporate PR” misconception. These days I tend to think it’s better that it has the CEOs names on it, both because it made a bigger splash, and because it really pins them down on the issue. So I guess it all worked out in the end. Although I do wonder how things might’ve been different.
Should we try to capture all the positive externalities?
There’s a general thing here, where I look at the world and see lots of examples like this, where positive externalities aren’t captured and incentivized, and think that’s a bummer. It’s generally hard to get enough credit for being helpful to other people. But it seems like a lot of people’s attitude is that those externalities aren’t actually valuable, just because they aren’t valued.
At the same time, I think there’s something important about having externalities that aren’t captured, or at least, not quantified. This leaves more possibility for human connection and human judgment. Sometimes, the right way to handle things is just norms, reputations, social things. We all know people we’d trust more or less to appreciate and reciprocate our help, to scoop us or to join our team. I think we could do a better job of capturing positive externalities, but in general, I wouldn’t like to gamify being a nice person too much. The metrics are never going to capture it perfectly, and it could be bad if they crowd out our more nuanced moral judgments, or devalue relationships of trust and care.
Thanks for reading The Real AI! Subscribe for free to receive new posts and support my work.
It happened again… I have an idea for a project. It’s a super cool project. I’d love for it to actually happen.
The only problem is… I don’t have time to do it myself, and I want credit. But I don’t see a way to get it — well, not enough of it, at least. Having this experience a lot is one of the reasons I started Evitable: When I have a cool idea, I want to be able to execute it in house!
As I was writing this, I realized I should share the example of the Statement on AI Risk, widely regarded as the most significant piece of public comms about AI existential risk ever, and which I initiated:
It still feels awkward to be aggressively claiming credit here, because the Center for AI Safety (CAIS) really took the idea and ran with it, shaping the vision and doing the vast majority of the execution. But I realized recently how few people know I had anything to do with it at all! This is despite me being mentioned on CAIS’s website, in the press release, and in media appearances. So I am trying to toot my own horn more these days.
If ideas are cheap, why don’t you give me yours?
This tension around how to divide credit between “ideas” and “execution” is a classic conundrum in research. A common saying is “ideas are cheap”. They say something similar things in start-up land:
There’s another complication with ideas: people can always claim, after you tell them your idea, that they’ve already had it.
And people who think ideas are cheap do often say things like: “Obviously a bunch of people have had ‘your’ same idea”. But in my experience, most people don’t actually act like they believe this. I’ve seen the same researchers preaching “ideas are cheap” jealously guarding their idea: “Well, I actually am working on it and want to get credit...” seems to be the attitude when it’s your idea. Researchers, like journalists, live in fear of being “scooped”.
Why don’t people believe ideas are valuable?
What’s going on here? I think it’s an example of people acting like the way things work as fair, even though it’s clearly not. I think this happens often because people think of fairness as more about getting what you can reasonably expect to negotiate for, rather than what you in fact deserve.
It’s hard to get value out of having an idea if you don’t execute it yourself, but that doesn’t mean it’s not valuable. When people do have a great idea and manage to execute on it, this can make them wildly successful. It’s true execution matters, but if you can do the execution yourself it seems like you clearly can get way more value out of it. We’d probably be in a much better situation if people were able to share ideas more freely without losing this value. Sharing an idea creates a positive externality, and that’s hard to capture.
I’m an “idea guy”… what should I do?
It’s also possible sometimes to get a lot of credit for becoming known for an idea publicly, and people sometimes act like (again) like this means the credit assigment system isn’t broken. But, to do this you kind of have to make it your job to get credit for the idea, or, better, your ideas in general. Like you need to build a brand for yourself. I do think it’s generally true that if you’re not in a position execute on an idea (and won’t be), often you may as well blast it publicly and get as much credit as you can that way, even if it’s “pennies on the dollar”.
In research, an approach I’ve found that works is: do a bit of work, enough to generate momentum, and then tell people “I’m doing this. It’s going to be great, and here’s why! Want to help?” I did this for a lot of my papers in grad school, which I cobbled together last minute. I’d run a proof of concept experiment, and find people who were eager to get their name on a paper in exchange for running another one of the experiments we needed. This kind of approach is how I ended up with a publication having the longest list of authors among all the papers at the International Conference on Machine Learning in 2017 (“A Closer Look at Memorization in Deep Networks”).
But I’ve also heard horror stories of well-resourced teams hearing about a junior researcher’s idea explicitly threatening to “scoop” it from them. Reflecting on this example, I’m imagining a start-up telling a tech giant “We’re doing this cool thing! Want to help?” I don’t think that ends well for the start-up, unless they’re ready to be acqui-hired. Business people: weigh in here!
Oversharing as a solution?
A friend I know has what I think is a good approach: after generating momentum, send a document about the project to a few dozen colleagues as a “heads-up”. This seems wildly risky from the “getting scooped” perspective, but in practice seems to mostly generate buzz.
Actually, I did something sort of similar with the “CAIS Statement” as it’s typically referred to: I wrote to a couple dozen people I knew and told them I was probably planning to circulate such a statement, but also “it seems like the details could matter a lot in getting good uptake, but I don't think I am the right person to figure out the right strategy... It seems more like a job for marketing / political campaign experts.”
At some point, CAIS decided unilaterally to do it. I had mixed feelings because I figured they’d execute well, and I mostly just wanted the thing to happen and didn’t feel like I could claim permanent ownership of the idea. But also, I’d lost control over it, and while they kept me in the loop, I didn’t have the final say anymore. I had wanted it to be a letter from independent academics, to avoid the “corporate PR” misconception. These days I tend to think it’s better that it has the CEOs names on it, both because it made a bigger splash, and because it really pins them down on the issue. So I guess it all worked out in the end. Although I do wonder how things might’ve been different.
Should we try to capture all the positive externalities?
There’s a general thing here, where I look at the world and see lots of examples like this, where positive externalities aren’t captured and incentivized, and think that’s a bummer. It’s generally hard to get enough credit for being helpful to other people. But it seems like a lot of people’s attitude is that those externalities aren’t actually valuable, just because they aren’t valued.
At the same time, I think there’s something important about having externalities that aren’t captured, or at least, not quantified. This leaves more possibility for human connection and human judgment. Sometimes, the right way to handle things is just norms, reputations, social things. We all know people we’d trust more or less to appreciate and reciprocate our help, to scoop us or to join our team. I think we could do a better job of capturing positive externalities, but in general, I wouldn’t like to gamify being a nice person too much. The metrics are never going to capture it perfectly, and it could be bad if they crowd out our more nuanced moral judgments, or devalue relationships of trust and care.
Thanks for reading The Real AI! Subscribe for free to receive new posts and support my work.
Share