Transcript
I asked you guys to figure out Goodhart's Law from three examples. You had to be accurate and succinct. There were a lot of good responses and way too many cheaters who obviously either knew it or looked it up, based on the specifics of the language that they used. But here are the top three examples from Instagram and TikTok. Number three. From Instagram. When a metric chosen as a proxy becomes the desired outcome. Cleanly succinctly stated. But it doesn't quite encapsulate how this can create a functional gap between the new desired outcome and the original. Number three from TikTok. Attaching scores or ratings to a particular rather than a whole system, causes the part that is being treated to be prioritized, leading to the illusion of progress or growth while the system as a whole degrades or declines. It's wordy, but I actually really like how it captures a key part, which is the illusion of progress. Thoughtful. Number two from Instagram. From Chris Hope, what you measure becomes the goal, regardless of alignments to the true objective. Nice, but still doesn't talk about the true objective being detracted from. And number two from TikTok, Benedict Duda. If you incentivize a certain behavior and reward the fulfillment, you create a system that is based around, fulfillment rather than the original intended behavior. Solid deduction encapsulates the gap. A little bit wordy. And number one from Instagram, Union Thug. When measures of success become goals in and of themselves, corrupting the systems that were being measured. Well put, succinctly covering both parts. And number one from TikTok, check-ch-rum. When you reward a simple metric for a complex system, people optimize for the metric and you get worse results. And honorable mentions. This is well-reasoned and clearly not cheating, but a little bit verbose. This is succinct, but a little bit of extreme and might be a cheater. And this nails it, but also enough to be a bit suss. But well done if not cheating. Good job, everyone.
Additional notes
Replying to @Union Thug🇦🇺🦘🐨 These are the top ones from the time at which I started making this video, but there have been a lot of really great ones since then! (and it's possible I may have missed a good one– sometimes it’s hard to judge) Also, when going to post this i realized I messed up, and completely got my platform wires crossed. Every time I said TT it should have been IG and vice versa 🤦♂️ ❓How does this differ from Perverse Incentives? “Perverse Incentive” is the big umbrella. It's any incentive structure that produces an outcome opposite to what was intended. It doesn't require a metric, a target, or a measurement system — just a reward that backfires. Example: paying firefighters per fire responded to could incentivize arson. The term is purely about unintended consequences of a reward structure. The Cobra Effect is a specific type of perverse incentive where the incentive directly increases the problem it was designed to solve Goodhart's Law is about measurement corruption. It requires a metric that starts as a useful indicator, then gets adopted as a target, which causes people to optimize for the metric itself rather than the underlying thing it was supposed to represent. It doesn't even require a formal incentive- just the act of watching a number closely enough can distort behavior around it. The law is more epistemic: it's also about losing informational value, not just getting bad outcomes. #Thinking #science #economics #tiktoklearningcampaign @Avisha =🧬Mildly Mad Scientist
References
- Goodhart's Law, perverse incentives, and the Cobra Effect discussed in caption/transcript; source links not listed in workbook.