Arthur Samuel, Failed Startups, and the Democratization of Large-Scale Computational Power
Arthur Samuel is widely considered by many as a pioneer in the field of artificial intelligence and machine learning. Many recognize him as a creator of computer checkers—that’s right, the game. This may not be a big deal in 2015, but in 1959 it was!
Arthur’s groundbreaking research took place almost 70 years ago. In fact, he was one of the first people to introduce the idea of “teaching” computers to play games. But why checkers? The reason is two-fold: it’s a relatively simple game to follow, yet it has ‘depth’—that is, it requires thinking and strategizing. A game of chess (like Deep Blue) would have required an equivalent of 6,000 times the computational power available at the time. Arthur applied alpha-beta pruning to save the very limited amount of memory to play with. He also wrote algorithms to make his program better.
As I read about Arthur and similar experimental efforts to explore computational possibilities, three things of significance stood out: in his time, computers were restricted to the research fraternity (e.g., governments, universities, funded research studies typically backed by corporations); had multi-year research cycles (and thus outcomes); and they were ‘elitist’ and accessible only by a privileged few.
I was speaking with one of my close friends, a post-graduate research associate from the National University of Singapore. His team’s research concentrated on machine learning with applications for detecting credit card fraud. While they were energized by the research results, they were disappointed not to see this extensive effort find its way into the ‘real world’. They wanted to launch a startup, but weren’t sure whether it would find footing beyond the thriving and encouraging University campus. After much deliberation and bouncing the idea off a number of people (including me), they eventually decided to take the plunge.
They started their journey on March 15, 2008. It was a bumpy ride they hadn’t exactly planned for. They started with a $100,000 investment to get them off the ground. Almost 90% went into buying/assembling the technology infrastructure (computing, storing, programming, etc.) and saved on human capital since all of the team members were PhDs and domain experts. The first six months were anything but picture perfect. They couldn’t attract the right sales/operations leader—all worthy candidates didn’t quite understand the data play or how it would help solve the growing identity and financial theft problem.
The upfront investment of $100,000 wasn’t nearly enough to support the computing power they needed to show results, and customers weren’t willing to bet on an unproven team and technology. To top it off, Lehman Brothers filed for bankruptcy on Sept 15, 2008.
The startup shutdown in June 2009. The silver lining: they were able to sell their proprietary IP in order to return invested money to friends and family.
By some estimates, the world has over half a million startups. Similar studies approximate that about 40,000 new startups emerge every year about the same number fold. Equally encouraging is the rise of what can be called “Deep Information Technology” companies—artificial intelligence, computer vision, Natural Language Processing, predictive analytics, robotics, etc. We also see a sudden surge in business viability of such efforts given the data explosion from the last 20-25 years (and the high noise-to-signal ratio).
This made me wonder: what’s really changed in the last 6-7 years? Well, almost everything!
Investors increasingly see data science (as a discipline) coming of age with more ways to solve real-world problems.
With the advent of the cloud, ‘Cost to Compute’ is close to zero (at least for startups)—most serious, large-scale cloud vendors offer virtually unlimited cloud sources with zero down payment. Investors increasingly see data science (as a discipline) coming of age with more ways to solve real-world problems. Thus, it can be monetized and can access diverse talent. More and more investors are now willing to pursue such opportunities, to influence and transform some of the most foundational industries with the help of data science. With super-efficient algorithms, exploding open-source libraries, and large corporations contributing generously, the possibilities for starting up are better than ever before. In fact, average valuations of (and investments in) data science-focused startups, on average, are 15-20% higher than their non-data science-centered peers.
Oh, and one more thing—remember my friend who couldn't quite make it work in 2009? He’s back in the game. He and his team worked their way through more traditional career paths as data scientists, researchers, and data practitioners, saving up their pennies and waiting for the right moment to relaunch what they believed in. They started again last year and what a start it was. Cloud computing costs for the first 2 years: $0. Infrastructure is supplied by a large-scale cloud service provider. Cost to create advanced algorithms: $0. Several of the team’s open-source projects received contributions. Still, they had to find the right operational leader, and that meant taking money out of the piggy bank. But my friend didn’t even have to post a position on LinkedIn or other job portals—word of mouth brought him over 50 candidates for a couple of operational/sales roles. He finally had all the pieces of the puzzle!
The company is doubling customers every quarter. And did I mention they secured a couple of million dollars in seed funding, fueling his company’s expansion plans? Though my friend views this as a cash reserve “just in case”—they are significantly cash flow positive.
I met up with him recently. He was speaking to a group of emerging data scientists (a mentor now just as he was 20 years back) and told them there is no better time to start a company, even better if it is focused on data science.
What a difference a decade can make. Just ask my friend!