Trust, Then Verify: How Square Let 99 Percent of Merchants In

Lex Fridman introduces Jack Dorsey as co-founder and CEO of Twitter and founder and CEO of Square, then immediately narrows the brief. Given limited time and turmoil around Twitter leadership, they would talk Square, cryptocurrency, and some philosophy, and save an in-depth conversation about engineering Twitter for a second appearance. The recording itself was made before the pandemic. In the intro, Fridman notes that Dorsey later moved a billion dollars of Square equity, 20 percent of his wealth, into a COVID-19 relief effort. Andrew Yang called it a spectacular commitment. The organization posts every donation to a single Google Doc.

If Square has one word for its mission, Dorsey says, it is access. The company thought it was building a hardware dongle that plugged into a phone so anyone could swipe a card. Talking to people who had actually tried to accept credit cards, they heard a different story. Many were not even allowed to process payments. Banks and merchant acquirers waited with a credit check and a FICO score. A lot of small businesses, brand-new entrepreneurs taking personal and financial risk, did not have good credit or a credit history. For the job of taking money from customers, they had to get their own credit checked. It felt ridiculous.

Digging further, Dorsey decided the financial industry was not trying to be cruel. A credit score was simply the only tool it had to judge authenticity, intent, and the predictability of future behavior. Square built the hardware, but the real work was risk modeling. In the first year the company stood up a serious data-science discipline, because it already knew the business was not hardware. It was letting more people into the system.

Fridman wants to talk about engineering at the scale of trillions of interactions a day. Dorsey started in C. He refuses the title. "I'm a hacker. I'm not really an engineer, not a legit software engineer." A hacker, in his definition, does whatever it takes to make the thing work so he can see it and feel it, then learns what has to come next. What comes next is usually bringing it to more people, which is scale. You hire people you can learn from, people with experience or people who learn extremely fast. Square, he says, has benefited enormously from open source: mistakes, successes, and problems laid bare. Open source moves slowly, but it is deliberate, and because of the pace you can see what it actually takes to build something meaningful.

Is Square's reliability a hardware problem, a software problem, or a social problem of large engineering teams? Dorsey will not pick. You have to focus on all of them, and the ability to do that comes down to how you face problems and whether you can break them into parts. The usual failures are trying to address too many things at once, not going deep enough with the questions, not being critical of the answers, or not taking the time to form a hypothesis you can test. If he could improve one skill every day, it would be critical thinking and problem solving. That is what produces learning. He tries to resist the categories. Hardware solutions hide inside software problems and the other way around. Underlying systems only show up if you can ask critical questions and have the patience to go deep.

The financial industry's default, as Dorsey describes it, was distrust: constantly looking for reasons people should not get in. Square flipped it. Trust, then verify, verify, verify, verify, verify. They stopped watching at the merchant level and started watching at the transaction level. Come in, run some payments. As long as the behavior looks high-integrity and not suspicious, the company keeps serving you. If something interesting shows up, it bubbles to a human who looks for something nefarious. That is when they might ask you to leave. Fridman calls the shift beautiful: assume people are fundamentally good, then catch the ones who are not, instead of assuming everybody is bad.

The number is the proof. When Square entered, only about 30 to 40 percent of people who applied to accept cards got through. They took that to 99 percent. Machine learning was there from the start, just under other names. First it was data science. Then machine learning became the culture. Then deep learning. It will be called something else later. All the same vein. They built the discipline in year one for another reason: they had to partner with a bank, with Visa, with Mastercard, and show that bringing more people in would not compromise those systems.

How does an upstart with some cool tricks convince those incumbents? They staked out in tiers. A bucket of 500 people, then results, then a thousand, then 10,000, then 50,000, and then the constraint was lifted. Get something tangible out there. Show what you can do rather than talk about it. That pressure, Dorsey says, created a culture of accountability and a little more transparency, and it incentivized the early team in the right way. External constraint, in other words, was not the enemy of culture. It was how the culture got made.

Asked what the future of access looks like, Dorsey goes to currency. He believes the internet deserves and requires a native currency, which is why he is a huge believer in Bitcoin. Square's biggest problem as a company, right now, is that it cannot act like an internet company. Open a new market and you need a local bank and a different regulatory onboarding environment. A digital currency like Bitcoin takes a bunch of that away. In theory you could launch a product in every market because they are all using the same currency, with a more consistent understanding of regulation. Internet access is still number one. Currency is number two. He wants to see it and feel it in his lifetime.

Fridman had just talked with Vitalik Buterin and asks what Dorsey finds appealing. The most beautiful thing, Dorsey says, is that no one person sets the direction and no one person on the other side can stop it. It is organic, and principled in its original design. He thinks the Bitcoin white paper is one of the most seminal works of computer science in the last 20 or 30 years. "It's poetry." Releasing it under a pseudonym was a powerful statement. The timing was an activist move, noble and honorable, and it let everyone be part of the story. Ten or 20 years out, the amazing thing is that no one knows. Every developer and every user can change its direction. That is what the internet showed was possible, even if large companies now own too much of it and equal access is not here yet. Those problems are visible, which for him means they are discussable, which means they are solvable.

Would digital currency become the world's main money and push control away from national fiat? Definitely, he says. The bigger ramification is how it changes how society works. Money is a foundational layer that enables much more than money. In Ethiopia he met a woman building the Lyft or Uber of that country. Riders could not easily pay the company; she could not easily pay drivers. Growth stalled because she had to think about payments instead of the rider and driver experience. He spent a month on the continent. The majority of companies he met were payments-oriented. Moving money across African borders, corruption in existing systems, and the lack of an easy way to pay people make starting anything hard. A more durable, resilient, global standard, he thinks, would let innovation show up in health, sustainability, and transportation, not just in payments shops.

Fridman hears "anonymous launch" and infers that Satoshi Nakamoto walked away from credit. Dorsey corrects the word. It was not an anonymous identity. It was a pseudonym. Anonymity is random: drop something off and leave, with no intention of building an identity around it. A pseudonym is a constructed identity. Nakamoto might be one person or several. The window was short, but the identity was meant to stick around and be known, and the community still honors it. Doing it as a pseudonym rather than a real name, and rather than pure anonymity, was smart. It builds tangibility and a little empathy. You can imagine a human, or a set of humans, behind it. The cost is a sacrifice of ego.

Would Dorsey do the same things if his name were not attached? Is his ego tied into the decisions? He hopes not. He believes he would attempt the work without the name. Inside a corporation, legally, that is hard. On an open-source project, absolutely: he does not need his real identity on it. The appreciation that comes from doing something good, seeing it, and seeing people use it is pretty overwhelming, he says, more so than seeing your name in the headlines.

On AI, Dorsey does not start with philosophy. For now, and for at least years out, models can bubble up interestingness very quickly and pair it with human judgment about severity, depth, nuance, and meaning. The chasm to general intelligence, for him, is the ability to explain why, and the meaning behind a decision. Explainability in natural language is essential. He thinks one of the biggest risks going forward is that we are building black boxes that cannot say why they decided, or what criteria they used, and we are trusting them more: lending, content recommendation, driving, health. A watch that tells you to stand up is a simple case. The systems get much more complex. He adds a fair note. Sometimes we cannot explain our own decisions. We are being a little unfair to AI, because we are not very good at this either.

Could we build something like the movie *Her* that a person could fall in love with? Hasn't that already happened, Dorsey asks, hasn't someone in Japan married an AI? Fridman means a larger scale of deep connection, not a single outlier. Dorsey moves the camera onto the human. Whether a deep connection forms is less a function of the AI than of how and where that person finds meaning. Can humans find meaning in technology this way? A hundred percent, and he does not necessarily think it is a negative. Meaning is subjective. There will not be a magic algorithm that makes everyone love the machine.

Will we still need to tell a bot from a human in ten or twenty years? In some problem domains, yes; in others, maybe not. What is interesting is the speed gap. The technology to create is moving much faster than the technology to detect. Adversarial machine learning is full of systems whose job is to fool other systems. Detection has to be two or ten steps ahead. For Square that is not abstract. Risk models are built around identity. Payments come down to identity. Deepfakes pointed at driver's licenses, passports, and state IDs are a way to walk through Square or Cash App and start taking cards. Those creation tools are moving very quickly. Detection, he thinks, is probably lagging. With more focus you can catch up. There will never be a perfect detector, same as security. What you can do is raise the speed of iteration, take signals of correctness and error as fast as possible, and feed them into newer or self-learning models.

Asked about existential risk and Elon Musk's worries, Dorsey talks about Yuval Noah Harari's *21 Lessons for the 21st Century*. The last chapters look like they are about meditation. Harari's actual claim is that children being born today will have Google with a stronger sense of their preferences than they do. Dorsey can easily imagine that Google already knows his preferences better than his mother does. For someone who grew up only knowing the internet, only knowing what Google or Facebook or Twitter or Square can do, self-awareness is being offloaded to other systems, especially these algorithms. Harari's fear is that we lose it because it now lives outside us and does a better job of directing decisions: should I stand, which doctor, whom should I date.

Meditation, in that frame, is a tool to rebuild self-awareness and to ask why you made the decision, why you reacted, where the thought came from. Dorsey amends the destination. It leads to awareness, not necessarily control. Awareness means noticing that you are offloading a decision to an algorithm you do not fully understand and that cannot tell you why. That is not the same as saying the algorithm is bad. The best recommender systems help you on a journey of learning new ideas. They can be a great thing. The remaining question is whether you know you are doing that, whether you know you invited the influence. Fridman boils it down: the invitation is fine; drifting without noticing it is not. They detour into Pixar. Fridman ranks *WALL-E* just behind *Ratatouille*. Dorsey disagrees. *Ratatouille* is the best. It is their first disagreement of the day, a rat's entrepreneurial story versus space.

Fridman brings up Andrew Yang and job displacement. Dorsey says Yang was 100 percent right on almost every dimension. Square sees it in the business. Yang pointed at truck drivers. Dorsey is from Missouri and knows the anxiety people from there feel every day, often invisible, not talked about enough. The next big one is cashiers, which is Square's world. Point of sale is moving into the customer's phone, into apps and order-ahead. There are more kiosks, more Amazon Go-style stores. The number of retail cashiers is immense, and there are no real answers yet on how they transform those skills into something else. That will have really negative ramifications. Yang's case for universal basic income matters because the shift is coming and retraining takes time. People need a floor.

A thousand dollars a month is such a floor. It will not make you quit your job, because it is not enough. It will let you worry less about getting through the day so you can ask what to do now and what skills to acquire. The industrial-age rebuttal, that factory lines caused the same panic and everything worked out, misses the change in velocity and centralization. What makes this wave run is data and the algorithms on that data, and they are highly concentrated. The scarier second problem, for Dorsey, is who owns the data, who can operate on it, and whether we can share insights so other people can build algorithms that serve their own needs.

That is why he thinks regulation could play a strong and positive part: look at the primitives of AI and the tools used to build services that will touch every aspect of human experience, and look at where data is owned and how it is shared. As a society, as a world, we need better answers than we have. Right now it is way too centralized in a few very large companies. UBI is well past due. Martin Luther King trumpeted it, and others before him. The exact thousand-dollar number might be wrong. You still have to take a step, try it, and see what works.

Fridman and Dorsey both fast. Fridman, who is Russian and claims to enjoy suffering, says it helps him sit for hours as a programmer. Dorsey ate three meals and snacks for too long without asking why. He now eats dinner only. The first week feels like a superpower. Then you realize it is not a superpower. You realize how much the mind dictates what seems possible, and how many surrounding structures are social custom rather than bodily need. Three meals a day is one of those. He has experimented on himself since childhood, including two years as a vegan, because health is the most precious thing we have and nobody really understands it. He heard someone, famous for breath-holds, say he ate one meal a day. It sounded challenging and uncomfortable, so he did it. He learns the most when he makes himself uncomfortable. Everything comes to bear in those moments.

As a child he could not speak well, went to a speech therapist, and became extremely shy. One day he signed up for speech club: a topic on a note card, five minutes to write, then five minutes speaking without the card. It was the most uncomfortable thing he could imagine. It gave him perspective on communication, on his deficiencies, and on the fact that if he set his mind to something he would do it. Fasting sits in the same frame. It feels, occasionally, like the most boring superpower one can imagine. On death, he does not try to make sense of it. He thinks about it every day, multiple times a day, and he is not afraid. It is a transformation he does not understand, and also a tool. He has an hour: is this really what he will spend it on? He has only so many sunsets. The reminder is everywhere, in his body, a friend, a coworker, the news. Death is the ultimate clarifier of whether he is focused on what matters, work or friendship or family.

What is the meaning of life? Being aware that he is alive is already pretty meaningful. So are connections with people he just met, long friendships, and family. Seeing people use something he helped build is powerful. It comes down to a sense of connection, of being part of something bigger than himself, felt in small ways or large. Are we living in a simulation? He does not know. It would be a pretty fun one, also crazy, random, and full of problems. Would he have it another way? As a planet we have taken way too long to realize we are all in this together, connected in significant ways, and we hide that connectivity through ego and the errand of the day. That is the one thing he would want to change. If we cannot do that, how are we going to connect to the other simulations? Fridman closes with Paul Graham on Bitcoin: it has all the signs of a paradigm shift, hackers love it, yet it is described as a toy, just like microcomputers.