Could an algorithm do a better job of picking a startup than a human?
Sometimes, technology can produce some exquisite paradoxes bordering on the self-referential: while risk capitalists are going crazy trying to pick the best machine-learning startup, it doesn’t seem to have occurred to many people to use a machine learning algorithm to decide which one to invest in.
But an entry on the BigML blog, the company that I collaborate with as an Strategic Advisor, has announced a joint project with Telefonica Open Future with just that goal: to try to develop an algorithm able to predict the success of a startup.
By now, there will be no shortage of data: years of startups asking for financing and providing detailed financial reports from which information can be extracted, along with the subsequent metrics on the evolution of the market, as well as how successful each company has been.
The real objective here is to create a better process than currently exists, which some have called Silicon Valley’s dirty secret: in reality, factors such as how good an idea the business is built on, or the team driving it, turn out to have practically nothing to do with attracting finance, because in reality, the whole thing is based on mediocre humans trying to take decisions negatively influenced by irrelevant questions that do not affect the quality of the idea, how it fits into the market, or the strength of the team. In practice, what really influences who gets financing and who not has much more to do with who you know, who introduces you, the state of the investor’s pipeline, or how that investor’s recent projects have gone, etc.
If you think there is some kind of mystery in all this whereby the best ideas always make it, think again: in reality, the whole process is a pile of garbage; quality has nothing to do with it, and many ideas that could change the world are totally ignored, while others that are worthless and end up failing miserably. If the numbers add up for some venture capitalists, then that is just the law of averages at work. Pure chance. The proverbial “acumen” of some investors is little more than the fruit of a flawed process.
Is it possible to create an algorithm that would be the perfect investor, something that would at least provide the framework to look at each project on its merits, avoiding the inevitable human bias that should have nothing to do with a decision? Can a machine study the specific characteristics of a particular technological wave, and on the basis of the startups it engenders fight for financing, establish which of them it should access to create companies with a bigger chance of success and that will maximize the investment? The complexity of the variables involved, and the relationships between them would suggest this is possible, although once again, we could just be talking about self-referential factors: the simple fact of being chosen by an algorithm could mean other elements come into play that would distort the final result for better or worse. But in any case, some well known companies are already asking algorithms to help them shape their investment portfolios…
The question is an important one, nevertheless, if one remembers that we live in a time when more and more companies are trying to develop just about every kind of machine learning algorithm possible: it is very likely that in the not-to-distant future, your being lent money to buy a home, the price of your insurance, whether the tax office looks into your financial affairs, and many things decided by supposed experts applying their supposed acumen and the supposed value of their experience will be decided by algorithms.
At the next PAPIs, the leading international conference on predictive APIs and related apps, an algorithm will be used to choose the winner from a list of startups, which will then be funded by Telefonica Open Future. A predictive algorithm trying to chose the best startup based on predictive algorithms: now that’s what I really call self-referential :-)
(En español, aquí)
