How to Think About A.I. Like a VC
Automating intelligence changes jobs. That’s different from ending work.
Let’s start with the obvious: You don’t need to feel sorry for venture capitalists. At the same time, with a little empathy for what they’re going through you can gets some clues about where a tech-driven economy is going.
VCs are, after all, people who listen all day to pitches from other people who typically want to put the latest and coolest technology against the most challenging and rewarding problems. Even on a good day, this can mean listening to a lot of the same thing.
These days, that means hearing nonstop pitches about Artificial Intelligence. While A.I. is certainly a very big deal — it’s at the center of self-driving cars, home assistants, and most of the activities of my employer, Google — that doesn’t mean it’s fun hearing how every startup is now A.I. compliant or an A.I innovator.
“We’ve gone through three big trends in the last 10 years of pitches,” said Frank Chen, head of deals and research at Andreessen Horowitz, one of Silicon Valley’s larger VCs. “Everyone was ‘mobile first.’ Then they were ‘cloud native.’ Now it’s ‘A.I. inside.’”
Mr. Chen was speaking at — where else? — an A.I. competition, sponsored by Google Cloud. First announced at the Google Cloud’s annual Next event last March, the competition attracted hundreds of startups, vying for investments of $500,000 each from two venture firms, and up to $1 million in cloud credits (including lots of A.I. tools) at Google. There were 10 finalists, and Mr. Chen was one of eight judges, all VCs.
Tech trends, like tech bubbles, are funny things. Most of them tend to become overblown because there is at the technology’s heart something important, and it hasn’t yet found its place in the world.
For individual companies, think of finding the right place as the product-market fit. For trends and bubbles, think of learning how to choose and price. The nineties Internet bubble, for example, was right: The Web really was going to change everything, it just wasn’t clear how. The telecom bubble at first rewarded companies digging trenches and laying fiber, but the bigger economic winner were the companies that used the pipes to remake global supply chains.
Contests like the one Google Cloud sponsored are useful, therefore, for clues about how VCs are see how a tech will embed in the world. They prosper not by how amazing a tech is, but how it makes them money.
That’s why it was notable that six of the 10 finalists were from medical fields, including one in veterinary health.
A.I., is about finding previously unseen patterns, then predicting outcomes from those patterns. In the currently popular form of A.I. known as Machine Learning, the software learns from the patterns it’s finding, figuring out what to look for next.
The most valuable pattern-finding tends to be from data that’s hard to secure, particularly in industries rife with expensive existing methods.
Health care has both: Medical data is historically tough to obtain in sizes large enough for an A.I.-type study. In some cases, as with two different startups (Brainspec and ReviveMed) looking at molecular physiology, that’s because it’s hard to find. In others, because of regulatory and paperwork issues, or because the data is new on the market, or even that it hasn’t been well-collected before.
That is interesting to a V.C., because specialist knowledge or new and proprietary collection and analysis methods raise the bar to competitors. The other part, the costs in the existing market, also make healthcare an attractive target: The early money in a new trend often comes from arbitraging past practices against new methods.
“A.I. for the sake of A.I. is investing in science,” said Matt Ocko, managing partner at Data Collective, which along with Emergence Capital would invest into one of the two winners. “We’re looking for what we call ‘golden triangles’: The proprietary nature of the data, informed by a great algorithm, used against an existing industry in a way that strengthens and feeds the data set.”
Data Collective invested in Brainspec, he said, because the science of spotting metabolites in the brain (the core data) involved a new way of using magnetic resonance imaging (or MRI) equipment. The A.I. analysis seems like a promising way to spot brain tumors, an expensive problem. The business is clearly attractive to MRI makers, since it helps their business.
“This business already has golden triangle feedback,” Mr. Ocko said.
Much like the medical companies, the other four companies tended to have people with a deep experience inside the industries they hoped to dominate. Emergence Capital made its investment in LiftIgnighter, a company formed by former Google Machine Learning specialists that says it can offer any company Netflix-, Amazon-, or Google-style personalization to their Web pages.
Though the company sifts through some 50 billion data “signals,” “they have a strong lack of jargon,” said Santi Subotovsky, general partner at Emergence. “The founders aren’t losing sight of the problems they’re trying to solve, for the sake of the tech.”
LiftIgnighter also won $500,000 in Google Cloud credits. The winner of the $1 million GCP Credits prize was PicnicHealth, which captures digital data from faxed records for precision medicine (healthcare is still remarkably dependent on paper-based records.) The finalists that didn’t get investments or the big Google Cloud prizes each still received $200,000 of credits. They included a “fitbit for machines” that judges machine health by its pattern of vibrations, a warning system for customers about to discontinue online services, and a retail trend spotting service.
There are big worries that A.I. will take away jobs. To the extent that automation has, for over two centuries, remade work, that is true. Historically, though, that automation has led to better focus on so-called “comparative advantage” — doing the thing you are best at, while others do the things they are best at.
Based on the competition, and lots more early venture investments, A.I. is doing the same thing, simply in the latest way. The comparative advantage is the domain knowledge. We don’t know how many things there are to learn about, tag, count, and learn to predict, but it’s clearly a big number, and likely to grow bigger as we delve further into a well-tagged world.
Very soon, the question facing workers may be, What kind of data tagger do you want to be?