Machine learning and deep learning AI have gone from the niche realm of PhDs to tools that will be used throughout all types of companies. That equates to a big skills gap, says Gil Arditi, product lead for Lyft’s Machine Learning Platform.
“From very early on, Lyft was a data-driven company and a machine learning company,” Arditi told attendees today at VB Summit 2017 in Berkeley, California.
From car routing and fraud protection (two critical functions for the ride-hailing service) to self-driving cars, Arditi is working to extend machine learning across the company. That requires lowering the technical barrier to bring in a lot of staffers, from traditional software engineers to product managers, who are brand-new to AI.
“I see [machine learning] as something in the primordial soup phase,” said Arditi, in a conversation with Chris Messina, a veteran of Google and Lyft rival Uber. “It’s similar to, let’s say, database in the early ’80s or late ’70s. You really had to be a world’s expert to get these things to work.”
Today, of course, any engineer with a modicum of experience can spin up databases on user-friendly cloud services. That’s the path that AI processes have to travel, he says. Luckily, machine learning is making AI more accessible to newbies without a PhD in statistics, mathematics, or computer science.
“Part of the promise of machine learning in general but deep learning in particular … is that there actually is not a lot of statistical modeling,” said Arditi. “Instead of giving to the machines exact formulas that will address the problem, you just give it the tools and treat it like a black box.”
It’s not so important to find people with years of experience, he says — which is good, because very few people have that experience. Expanding machine learning just requires finding more people who are willing to learn. Messina asked Arditi if there was culture shock for engineers, who are used to clearly defined projects and bug fixes, to get used to the trial-and-error process of gradually refining machine learning models until they get it right.
“The most important part is to define the rules of the game … and then define the target function as actively as possible,” said Arditi. “Then people do try different ideas, and a lot of them fail, but there’s kind of this magic of evolutionary progress towards their objective, that bullseye point.”
And no company should be in it alone, said Arditi, who praised the value of sharing datasets developed by tech companies like Facebook, Microsoft, and Google (as well as universities like Stanford). Lyft is also committed to sharing its knowledge, making its AI model simulation and testing tool available to academia. Sharing data also requires thinking through how to protect user privacy. Lyft is discussing how it can expose some of the data that it has to help the academic community improve on fundamental algorithms used in machine learning.
That’s not a purely altruistic pursuit, though — Arditi said that doing so would benefit Lyft by letting the company leverage those ideas to improve its own systems.