My (recurrent) meeting with Goodhart
How to greet him when he’s at our front door in the third edition of Data and Decision
A person standing in front of a door photo – Free Person Image on Unsplash
Musings - Meeting with Goodhart’s Idea
For those who are wondering, yes, Charles Goodhart is still alive. He might be reading a book in his house right now while I type this sentence. And no, I wasn’t meeting him in person. I’ve met with his “law”.
You might know Goodhart from a story about Cobra or Streissand’s effect. It’s a story that you read to pass your time - wondering what kind of knowledge you can get by reading that. But Goodhart might sit closer than you think. Thousands of people who call themselves “decision makers” have faced this situation at least once.
Boss: Hey, what’s up with the analysis?
Co-worker: You know what, boss, we saw that a user who clicks this red button 5 times in their 1st week is more likely to be retained next month.
Boss: This is great!
You: Yeah, but we need to be careful using it. We need to-
Boss: Alright! We have the new sales team’s KPI for next month, then. Also, what do you think if we give personalized discounts to everyone who hits the red button more than 3 times?A few weeks later……
Co-worker: Hey, boss. Ehm, so you know about the red button thing?
Boss: Yeah, what about it?
Co-worker: So, we followed your decision. But 3 days ago, the fraud team found something about that…
You: (Visibly angry noises)
Annoying, isn’t it? Watching something bad happen is never good, especially when you predicted it. You know what the result will be if we just use insight like that willy-nilly. This type of incident so frequently happens that we have a term for Goodhart’s Law.
Goodhart’s Law is defined precisely as “ any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes.” Or we could also cite simple explanations usually found in the wild: “ When a measure becomes a target, it ceases to be a good measure.”
So, why is this important to you?
You can meet Goodhart’s in every venue of your life. Even something as “profound” as the AI alignment problem is eerily similar to the problem described by Goodhart’s Law.
“Oh wow, an intelligent agent tasked to do something is doing everything it can (including finding questionable shortcuts) to achieve it? Where do I see this before?”
Every time anyone sets a target without deeper thinking and reflection, Goodhart will find them.
Every time a decision is made without sparing some thought on how it could go wrong, Goodhart will come to their house and (often) give the opposite of what they want to achieve.
And who are the groups who usually talk (and recommend) about metrics and targets? Yep, analytics folk (including me).
As an analytics team, we are tasked not only to provide action-oriented recommendations but also to predict where the recommendations could go wrong. It’s both for practical matters (you might save millions of dollars of the company’s money by preventing big problems) or for ethical matters (as a professional, you must disclose every trade-off for every recommendation you propose).
We must face this Thanos of ours with every weapon we have. It also shows that we take our work as analytics professionals seriously.
2 Findings - 2 Banger Line
Apple Vision by Ben Thompson
We’re really fortunate to witness the development of 2 products that will define this decade: ChatGPT and Apple Vision Pro. There are apparent differences between them. One has already been released, and the other, we only had the demo. But you can’t deny it; most people saw these two felt experiences that iPhone users had in its first release.
So it’s a real treat finding one of my favorite writers, Ben Thompson, write this post about Vision Pro. He talks about the products, Apple’s aspirations, and his critique of Vision Pro. One passage stuck in my mind: Ben highlights how Apple and Meta (who already had VR products) have two opposing visions. Apple is personal and more solitary, while Meta focuses on more social.
One banger line from the post
“In other words, there is actually a reason to hope that Meta might win: it seems like we could all do with more connectedness, and less isolation with incredible immersive experiences to dull the pain of loneliness. One wonders, though, if Meta is in fact fighting Apple not just on hardware, but on the overall trend of society; to put it another way, bullishness about the Vision Pro may in fact be a function of being bearish about our capability to meaningfully connect.”
To be fully honest, I already have a horse on this debate. I’m a remote worker from a small city in Indonesia. I have long dreamed of the moment I can talk with other peers in a virtual world - outside the confine of the Zoom container. My dream is for Meta to win!
Why You Should Read More Thomas Sowell
This post argues why Sowell is underrated as an economist and why you should read more about him. For those who don’t know, Sowell is on the team of President Reagan’s Council of Economic Advisers. He’s also a Milton Friedman disciple - and received a “genius” label from the Nobel Laureate. But most people who know him only read his social critique and political commentary.
He’s much more than that.
Some of his work is quite important for analytics fellows like us. He’s the first economist that defines “knowledge” quite precisely (sometimes everyone uses knowledge and information interchangeably, missing the nuance that is quite critical) and elaborates on the role of knowledge in our society.
I enjoyed Sowell so much that I sometimes spend the entire weekend only reading his books (and I’m still able to get new insights every time I do this).
One banger line from the post - (Quoted in Substack)
“Moreover, not all knowledge is the same. I often think of Sowell’s framing of different meanings of “knowing”:
To say that a farm boy knows how to milk a cow is to say that we can send him out to the barn with an empty pail and expect him to return with milk. To say that a criminologist understands crime is not to say that we can send him out with a grant or a law and expect him to return with a lower crime rate. He is more likely to return with a report on why he has not succeeded yet, and including the inevitable need for more money, a larger staff, more sweeping powers, etc.”
Thank you, everyone! Have a lovely weekend.