How to solve ‘black box’ problems.

Today’s post is also available in podcast form on Soundcloud and iTunes:

New skills to pay new bills.

One of my favorite online thinkers is Penelope Trunk, entrepreneur and home-schooling mother.* She wrote “Search is the most important academic subject today,” and explained:

“So instead of wasting years teaching kids to memorize answers, why not move on to teaching kids to ask better questions? Because that’s what searching is, ultimately:  learning to phrase a smart question.”

This idea – teach skills not facts – has bubbled up elsewhere too. Cal Newport, Tyler Cowen, and  Peter Thiel all advocate for this approach in their respective books. Being able to use your brains and a computer’s brawn will be a good skill to have.

One way to do that is Fermi-izing.

 

Before we start.

Keep these 3 questions in mind:

1. How much data does Google have?
2. When will I die?
3. Will the Cleveland Cavaliers win more than 57.5 games this year?

We’ll use these questions to apply Fermi-thinking.

 

Not Furby, Fermi.

Introduced to me in Superforecasting, Philip Tetlock explains Fermi-izing as technique to solve hard (black box) type problems. “What information,” Tetlock writes, “would allow me to answer the question?”

Questions about Google, death, and the Cleveland Cavaliers are all “black box” types of questions. The answers aren’t easily knowable. You can’t Google answers about Google’s data. That doesn’t mean we can’t answer these questions. We just need to work a little harder.

We need to take the lead bullet approach.

In his book, The Hard Thing About Hard Things, Ben Horowitz writes about facing the challenge of his company’s web server being too slow. It wasn’t like he could go to the Apple store and upgrade (a silver bullet/ non-black box type of problem.

Rather, Horowitz had to take a lead bullet approach and solve one small problem after another. He broke his big ugly problem into smaller ones.

Fermi-izing requires a similar approach. Another way to understand it is to apply Warren Buffett’s approach to investing, it’s simple but not easy.

It’s the same for breaking down black box problems, simple but not easy. Lots of lead bullets rather than silver ones.

 

Before we start, get in the right mindset.  

How much data Google has is a complicated problem. I keep 16GB of data with Google. But what about you? And schools? Plus Google Photos, plus data centers Ugh, this is getting complicated already. How do we figure this out?  We think of it like a puzzle.

A lot of really successful people do this. Chris Dixon says it’s like a maze. Ray Dalio imagines gems waiting for problems he solves. Griffin use the puzzle analogy. In Surely You’re Joking Mr. Feynman , Richard Feynman said he couldn’t “do” good physics. But when he decided to “play with physics,” “it was like uncorking a bottle.”

Think of black box problems like riddles to unravel. Rather than obstacles, think of them as opportunities.

 

How to start.

Tetlock writes “What (Enrico) Fermi understood is that by breaking down the question, we can better separate the knowable and the unknowable,” and “that’s a huge advantage.”

If something isn’t Googleable, then we need to figure out what parts of it are. The breaking down the question part is what will allow us to figure out what we can figure out.  

 

How much data does Google have?

Luckily we don’t have to do all the work, just search for someone else who did (remember that search is the most important academic subject).

In his 2014 TED Talk, Randall Munroe (creator of ‘what if?’ comics)* tried to figure out this question; how much data does Google have?

Munroe broke the problem down into knowable parts. Here’s part of his talk, emphasis is mine.

“There are a few things that I looked at here. I started with money. Google has to reveal how much they spend, in general, and that lets you put some caps on how many data centers could they be building, because a big data center costs a certain amount of money. And you can also then put a cap on how much of the world hard drive market are they taking up, which turns out, it’s pretty sizable. I read a calculation at one point, I think Google has a drive failure about every minute or two, and they just throw out the hard drive and swap in a new one. So they go through a huge number of them. And so by looking at money, you can get an idea of how many of these centers they have. You can also look at power.”

Munroe goes on to figure out how much square footage Google has, how many server racks fit into a square foot, and so on. He came up with an estimate of 15 exabytes of data. More he says, than the NSA.

We get an answer to a black box question once we break it down.

 

When will I die?

An impossible question. Maybe.

“In 1982, at age forty, I was diagnosed with abdominal mesothelioma, a rare and ‘invariably fatal’ form of cancer,” writes Stephen Jay Gould in Full House. As an academic Gould jumped into learning more

“All the literature contained the same brutal message: mesothelioma is incurable, with a median mortality of eight months following diagnosis.”

Gould had eight months to live, maybe.

Gould’s book precedes Munroe’s talk by a decade but he applies the same break-this-problem-down-into-smaller-parts.

Is 8 months a hard diagnosis, like the departure time for a flight, or something else? Gould thought it was something else. Emphasis again mine.

“I realized that all factors favored a potential location on the right tail – I was young, rarin’ to fight the bastard, located in a city offering the best possible medical treatments, blessed with a supportive family, and lucky that my disease had been discovered relatively early in its course.”

In the same way that Munroe collected variables like “money spent, drive market size, and power consumed,” Gould collected variables like age, location, and timing to update his answer. For someone with those variables the mortality timetable expanded. Gould lived for twenty more years.

 

How many games will the Cleveland Cavaliers win?

Before the 2015-2016 NBA season the betting number was 57.5 games. On his podcast Bill Simmons said it was a “lock” that they would win less than that. Why? Emphasis again, is mine.

“It scares me a little bit when guys in the NBA have bad shoulder injuries,” Simmons says about Cavalier Kevin Love. “I went under because of the injuries, because Lebron is now in year 13… what’s it going to take to get the #1 seed in the east, 52 wins?” “There were two straight years of his teams in the low to mid 50’s (wins) with Lebron playing like Lebron.”

Simmons breaks down the complicated problem to more manageable ones; will the team stay injury free, what does the competition look like, and what does the historical data say?

 

Don’t make predictions, find facts.

Each of the examples above, Munroe, Gould, and Simmons follow the same process.

First, they break black box problems down into smaller questions and they answer those questions with facts.

• Munroe does computations with the dimensions of a server rack.
• Gould reads medical journals about mortality rates.
• Simmons finds past season win totals.

Charlie Munger said, “[Projections] are put together by people who have an interest in a particular outcome, have a subconscious bias, and its apparent precision makes it fallacios. They remind me of Mark Twain’s saying, “A mine is a hole in the ground owned by a liar.”

Tren Griffin writes in his book about Munger, “Effective (Benjamin) Graham value investors are like great detectives. They are constantly looking for bottom-up clues about what has happened in the past, and more importantly, what is happening now. Graham value investors like Munger stay away from making predictions…What Munger looks for is a business that has a significant track record.”

Mellody Hobson said that when her company bought Madison Square Garden some people undervalued it because the basketball team playing inside wasn’t good. That’s not finding facts. “There may be a perspective,” Hobson said,  “of ‘they’re winning or not winning’ but we can look at the data like season ticket holders buying every year.”

 

Summary: 3 steps summary to fermi-ize well

1- Don’t be overwhelmed by complexity, just take things one bite at a time. Imagine your problem is just a giant Lego castle and the pieces need sorted.

2- Filter out what is knowable and what is unknowable and focus on the former. Separate the two piles.

3- Find factual answers for questions in your knowable piles.

–

Thanks for reading, I’m @mikedariano on Twitter. If you liked this post you can donate a few bucks here.

–

* I hate giving titles like “entrepreneur and homeschooling mother” or “billionaire” but I don’t know a better way to share that a person is awesome.