How Unlearning Helps To Build $7B Data & AI Companies
This is the Three Data Point Thursday, making your business smarter with data & AI.
Let’s dive in!
“New information triggers a new understanding of the world [... ] With technology advancing at the near-exponential rate [...] we are constantly bombarded with new information. [...] We can turn to the Cycle of Unlearning to help us embrace new approaches that will bring us success in the future.“ (Barry O`Reilly - Unlearn)
“Do not be a lazy thinker. Be an active thinker.” (Alexandr Wang, CEO and Founder of Scale AI)
Nowadays, a lot of Silicon Valley folks pride themselves on “thinking from first principles.” However, I’ve seen a different but related skill to be more valuable than just that - the continuous practice of unlearning.
In 219 B.C., the great Roman empire declared on Carthage a trouble-making city-state inside Spain. But before sending their armies to Spain, they got the news that a 28-year-old youngster called Hannibal was coming to them - to Rome. He’d already crossed the Alps, a direction Rome never thought anyone would march (after all, it is the Alps!), so no one was stationed there to warn them early.
Hanibal's army was 1/30th of that of the Roman empire, but over the years of his campaign, the Romans would become so frightened of him as to “avoid battle with him like the plague.” according to Robert Greene’s “The 33 Strategies of War.”
Greene crowns Hannibal as the ancient master of unorthodox warfare, utilizing hideouts, archers on elephants that so scared the Roman fighters as to send thousands of them drowning, a mysterious travel route no one could figure out, and much more.
Hannibal used a continuous practice of unlearning throughout years of battle against an enemy 30 times his size.
It is a continuous practice, never-ending.
It questions only the big problems he is facing (how do I get to Rome unnoticed?...)
It throws away all previous ideas (you can’t cross the Alps)
It allows you to discover what is truly right at this moment (using first principle thinking) - you can cross the Alps if you’re smart about it
It allows you, like Hannibal, to be the first to learn something new
And to crush your competition in each battle
As Barry O’Reilly alludes to in his book “Unlearning,” the data space is a space that necessitates this skill because of its exponentially growing nature. So it is in this space, the data space, that this practice will be one of the key leadership skills.
Key idea: If you’re a data leader and you don’t have a continuous practice of unlearning, you’re living in the past. If you constantly unlearn, you’re able to move 10x faster than your competitors.
A data leader that impersonates this skill is Alexandr Wang, founder and CEO of the data unicorn Scale AI, currently valued at 7 billion USD. Scale AI provides high-quality training data for AI applications, something that sounds as fun as going to the dentist.
Let’s see how and why you should build a practice of unlearning yourself.
Unlearning lets you be the first.
While many people can think from first principles and discover new business models, unlearning is the weird combination of accepting everything that works and only questioning the key hindrances you have right now.
If you asked Alexandr Wang 4 years ago what the biggest misconceptions about machine learning were, he would’ve told you this.
Alexandr is very clear on the potential of ML (and has always been) but also about the huge challenges.
He never bought into the hype, not into the ML hype, not into the big data hype, and not even into the generative AI hype. He’s deeply focused on the big opportunity of data & AI while focusing his business on getting the big problems most people do not see and ignore out of the way.
But you only get to this point by focusing on what’s the hindrance, the bottleneck, and unlearning what everyone thinks is true.
Unlearning is connected to first principles.
Once Wang has identified one big opportunity (AI), he starts to unlearn, to question the basic assumptions.
He uses first-principle thinking, going to very basic concepts like flows and bottlenecks.
“I think that's a tried and true method. In any new industry, or any even old school problems, there's always going to be bottlenecks that can be solved with technology or a new approach” Alexandr Wang in an interview with businessofbusiness.com
It’s how he realized data labeling is a problem, long before titans like Andrew Ng realized what was wrong with ML. Andrew Ng has been developing large-scale AI systems across the globe for over a decade, teaching machine learning to tens of thousands, and yet, it took him until recently to found the data-centric AI movement.
Not Wang, he founded Scale AI in 2016 and turned this data-centric approach into a 7 billion dollar business. Looking around the industry today, you might notice that Scale AI is pretty much alone in the field. The data-centric AI movement hasn’t really picked up; this approach, worth billions of dollars, is still discarded by most, like Hannibal’s tactics.
Even though Hannibal’s use of war elephants brought him a significant advantage, the Romans never really got the hang of it, never able to unlearn their focus on infantry and discipline.
Unlearning is saying no again
In 2016, Opendoor’s CEO Eric Wu was trying to convince Wang to join one of those hot start-ups over choosing one of the bigger mid-level tech companies.
In a classic unlearning moment, Wang realized he needed to question the basic assumptions in the first place.
“But overall, it persuaded me against accepting any offers at all.
Instead, the conversation convinced me that I needed to try to start a company. Not only would it be a great growth opportunity, I knew I would regret it if I never took the risk to be an entrepreneur at the perfect time.” (What I Learned in 2016)
Unlearning means realizing from all the options available coming up with your own is always one as well.
Interestingly, in 2016, Alexandr already had practice in saying no. Indeed, it wasn’t the first time he took an odd turn regarding career choices. Right out of high school, he used his coding skills to start working in the tech sector immediately, postponing MIT for a while.
Avoid the success trap.
Barry O’Reilly identifies the success trap as one of the key fallacies you need to unlearn. The idea of the success trap is simple: business leaders look to other business leaders to understand their sector, to see what is successful, and then imitate that.
It’s classic human behavior; it is how we learn as children and continue to learn as we grow up. But it isn’t helpful anymore when things get turbulent as they are right now in the data world.
In 2018, Google dropped its pilot project with the Department of Defense because, apparently, tech people are really passionate about this kind of stuff. Internally, this project caused quite a few problems for Google, resulting in people leaving the company because Google took on the project in the first place (and wasn’t that public about it).
But luckily for Wang, he didn’t look to the leaders at Google for decision-making but instead picked up the contract called Project Maven at the time. Most Silicon Valley tech start-ups stay away from the public defense sector to be successful; Scale AI embraces it.
But it takes courage to say no to common wisdom, unlearn what others think success looks like, and build your own image up from the ground. Wang did that and is now one of the key contractors in this field.
Unlearning is practical
Young Hannibal led his troops from the front, hands-on. That also meant he was the first to try out new terrain, realize problems right out of the bat, and learn new things in a practical manner.
You unlearn, paradoxically, not by thinking a lot but rather by doing. It is the quickest way to realize a held belief is not true. It’s by trying out.
At Scale AI, everyone is encouraged to label data because that’s the core business Scale AI is in.
Encouraging everyone at the company to be hands-on with the most basic aspects of the
business, Wang scales unlearning into the company.
As O’Reilly explains, the importance of practice lies in breaking the circle of old behaviors that lead to old thinking.
Wang’s personal realization that data labeling and data are the key problems in ML and AI, not algorithms, came from an inherently practical experience; he was building a camera inside a fridge so it could tell when he needed to buy groceries.
The problem was that he didn’t have the data to make it work, none at all!
In the data space, learning is all about acting and trying things out. It is the only possible way to unlearn and relearn because what other people think is right is likely wrong for you.
Building a continuous practice
Perhaps, the biggest difference between first-principle thinking and unlearning is the aspect of practice. While you apply first principle thinking whenever a decision comes up, unlearning is something you constantly apply to the body of knowledge you have, something you use to judge past decisions, understand events, and move forward in your own field.
It’s not just specific to the data field in that sense, a great data leader needs to unlearn, but that practice should carry over as it did for Wang when the pandemic hit.
“The biggest challenges for myself, personally, I've just been figuring out how we — it's really boring stuff — but how we can continue growing our company and making sure that we build a great company for everybody who's a part of it, and making sure that we're working for everyone who’s in our ecosystem” Scale AI ML Startup
Maybe Wang always had it, or maybe he picked the practice up at some point; in any case, his favorite Steve Jobs quote goes well with it.
“Again, you can’t connect the dots looking forward; you can only connect them looking backwards. So you have to trust that the dots will somehow connect in your future. You have to trust in something – your gut, destiny, life, karma, whatever. This approach has never let me down, and it has made all the difference in my life.” One of Wang’s favourite Steve Jobs quotes
A True Data Leadership Skill
Andy Grove didn’t let his company or product managers figure out how to get out of the slump Intel was in back when they were still doing memory. It was him, him, and Moore that took action that changed their behavior, as Barry O’Reilly points out.
Similarly, it is Wang himself that drives change, constantly unlearns, that acts day in and day out on this practice. Unlearning is a true leadership skill; it starts with you personally; you need to act and change your behavior, and only then will you be able to change and update the practices inside your company.
Yes, you need to scale the unlearning skill into your company, but you must start with yourself.
Further resources: I do recommend the book Unlearning by Barry O’Reilley. I’m also in love with another of his books called The Lean Enterprise, co-authored with Jezz Humble. Again, a very underrated book, but that seems to be Barry's specialty.
Here are some special goodies for my readers:
👉 The Data-Heavy Product Idea Checklist - Got a new product idea for a data-heavy product? Then, use this 26-point checklist to see whether it’s good!
The Practical Alternative Data Brief - The exec. Summary on (our perspective on) alternative data - filled with exercises to play around with.
The Alt Data Inspiration List - Dozens of ideas on alternative data sources to get you inspired!