I recently joined Slack as Head of Machine Learning because I believe this is the best place in the world to be doing applied ML. That’s a bold statement; I’ll explain.
When I was choosing my next role, I spent a lot of time thinking about what it would look like to be set up for success with machine learning, both as an individual and as an organization. This was my list of ingredients:
- Great product
- Unique data
- High-caliber people
- Scope of impact
- Kind of impact
- Social responsibility
Let’s dive into each of these in more detail.
- Great product. Delivering value for customers is the goal. Delivering that value is substantially easier when the product is already beloved by a loyal and rapidly growing user base that enthusiastically engages with the product every day. It’s fine to be motivated by a love of novel algorithms, clever statistics, or applying the latest-and-greatest from the research community, but it’s usually prudent to first try the simplest approach that’s likely to address the problem at hand. Compelling machine learning happens at the intersection of data and customer value.
- Unique data. A sufficient quantity of sufficiently high-quality proprietary data is necessary for building solutions with machine learning, and being the place “where work happens” means we have a unique, data-driven perspective on how to help organizations work better. What is “sufficient”? Who knows! But Slack has lots of data (more on this below). Why “unique”? Because Systems of Intelligence are the new moats, and unique data is among the most strategically defensible ways a modern company can deliver customer value.
It may seem surprising that those are the only two technical items on the list. Certainly, there are many, many other factors that affect the merit of the resulting solution, but the two items above are the technical factors that I’ve seen materially affect the feasibility of doing great ML. The items below pertain more to the company and the product, and are perhaps even more important.
- High-caliber people. Great machine learning doesn’t happen in a vacuum. To deliver product value, most machine learning applications rely on other teams to build the customer-facing last-mile. That means collaborating with those product teams, with project and product management, and with teams that generate or manage the data that we rely on. It’s not enough for the ML engineering team to be smart and talented; they must sit within an organization that is aligned, supportive, and similarly high-caliber.
- Scope of impact. Most of the hard work and daunting-but-tractable technical problems lie ahead of us. For some people, working on last-2% projects is inspiring. A sense of accomplishment can come from squeezing a tiny bit more performance or quality out of a system that has been optimized by the numerous, brilliant engineers who preceded you. But for others, it can be stifling. You’re bounded from above by how much improvement is possible, which also limits your potential impact; a system that’s getting 98% accuracy can only get 2% better. Most of the broad-scope, high-impact work to develop infrastructure and tooling, and to translate that into customer value, has already been done. At Slack, opportunities to build new capabilities, or make double-digit strides in the quality of existing capabilities, abound.
- Kind of impact. Impact at Slack means making peoples’ working lives simpler, happier, and more productive, not serving them ads or feeding their addictions. That’s not to knock those who do such work; I thoroughly enjoy some of my addictions. Personally, however, I would rather my work go toward making someone more productive by surfacing relevant information when they need it, automating rote tasks that distract from more important work, and so on.
- Social responsibility. Slack aims to protect user privacy and security, to prevent or mitigate biases (algorithmic or otherwise), and to generally behave ethically with respect to customer data and how we use it to improve the product. Any responsible company will implement policies and enforcement mechanisms, of course, but Slack’s standards for what constitutes safe, responsible, non-creepy use of data permeate our culture as well. Laws, policies, and enforcement mechanisms are necessary but not sufficient; you also need a pervasive attitude of social responsibility.
It won’t surprise you to learn that Slack has all of those ingredients. That makes Slack a rare bird, and it’s why I joined.
That’s not to say it will be easy. There will be plenty of obstacles, false starts, and stumbles. Slack is a rapidly growing company in a rapidly evolving competitive landscape. But we’ll be facing the right kinds of challenges: ones that exist for the right reasons and that are worth overcoming. That makes me happy, and hopeful, and I feel fortunate to be here.
Would it surprise you to learn that we’re hiring? ☺️
Bio: Adam Oliner recently joined Slack as Head of Machine Learning. He was previously Director of Engineering at Splunk, where he helped lead and grow the machine learning team for more than four years. Adam co-founded Kuro Labs, which commercialized battery analytics technology. He was a postdoctoral scholar in the EECS Department at UC Berkeley, working in the AMP Lab, which specialized in cloud computing and Big Data. He earned a PhD in computer science from Stanford University and a MEng in EECS from MIT, where he also earned degrees in computer science and mathematics.