Machine Learning, Artificial Intelligence, Deep Learning — I’m constantly being told that one (or all) of these will soon replace me in my job.
Before this time comes however, I’m still responsible for doing this job, which among other things, includes demonstrating to our clients the big difference machine intelligence can make for their business. Case studies showing such implementations are really helpful in driving these conversations.
Unfortunately, most popular case studies in the area of machine intelligence come from the Googles and Apples of the world. While they are very exciting and eye-opening, they are also done on a large scale. Very large scale. This often makes people feel that such endeavours require massive budgets and unlimited timelines.
That’s why I’m very excited about the feature that Slack announced a few weeks ago — an extension to their search. In a nutshell, whenever you search for anything, apart from showing standard search results, the tool will also show you channels and people that are known to talk about that topic. It’s not only an example of a very practical use of machine intelligence, but it also has some attributes that make it a perfect case study to show to our clients.
Small big data
If you do a quick google for “applications of machine learning” you’ll get a list of typical examples of spam classifiers and text recognition. On the other hand, if you read the news you’ll hear stories of driverless cars, on-the-fly translations and other big ideas. One doesn’t really spark inspiration, and the other one seems too out of reach — both, however, require pretty big data sets to be useful. The feature built by Slack uses a smaller set of data, based on a single company’s activity. Obviously, it can’t scale down indefinitely, that’s why Slack requires a certain amount of users in a team for this to work. Either way, it’s still a proof that you can do a viable recommendation system on a dataset way smaller than “all the images in the world”.
At a time when everyone is building chat bots and conversational UIs, seeing that an “intelligent” feature can be built right into the search results is another nice touch. The best machine intelligence is the one that doesn’t scream “Look, I’m an intelligent feature”, but just does what you expect, silently.
Extending existing functionality
Could the team at Slack have tried to create a completely new search? Of course. Did they? Well, no. By creating an extension to existing functionality, they’re saving themselves from the situation where there is not enough data. Or when the results are different than what user expected. All in all, it’s just a cherry on the top of your standard search results.
“This section will show team members who frequently talk about your search topic, as well as the public channels where they’ve discussed it”. This is basically it. It’s easy to understand what it does. It’s easy to understand how it works. It’s easy to understand how it brings value to the users. Especially because features like that usually take longer to develop, being able to easily explain the value they bring is so important.
This is a great start for Slack’s Search, Learning, & Intelligence group. I’m very excited for what their future work will bring to Slack and how it will make creating “inspiration decks” easier for me.
Until I get replaced by machines that is.
Featured gif from giphy.
Things don’t always go to plan on projects. It comes with the territory when working in an adaptive, lean and nimble way.