Back in October I was invited by IBM to attend their World of Watson event in Las Vegas – I wrote a little about it at the time.
Now that I have had some time following the event, I’ve been able to percolate and put my thoughts to paper, as it were. In the interest of you, dear reader, I’m splitting these thoughts into three different posts; Technology, Business and Philosophy. Note that there are no bright lines here: I’m sure to touch on each of the three topics in all three posts.
(I predict #1 and #2 will far outshine #3 in terms of traffic – such is life, friends!)
This post is the second, talking a little bit about Business. You can find my first post, discussing the Technology and my experience of it, here.
Context: SAO, A8C and WOW
(That’s “Simon Anthony Ouderkirk,” “Automattic,” and “World of Watson”, respectively)
I started the interview process at Automattic in the Spring of 2013 – that means that I’m bumping up against this being the longest I’ve been with one company in my whole life. Automattic is also the largest company that I’ve worked for, as an adult.
(When I say, ‘as an adult,’ I mean that yes, being a barista at Starbucks in grad school does mean I worked for a huge corporation, but I think of that differently. I didn’t really have a stake in the company – there was mutual tolerance but general disinterest in one another’s success. I’m glad they’re still around, and I am sure no one at Starbucks thought about me even once after I left their employ. This is fine, this is work.)
Let’s go with some personal context to paint a broader picture of how I experienced this event. Here come some numbers and comparisons.
I grew up in a small dairy farming town in Upstate New York. The population estimate for Westernville today ranges from 288 to the 340s. That means that the entire town could fit on a 747 and still have room left over.
So, that is to say, Automattic at just over 500 employees, is either already, or will soon, be twice the size of my home town.
If all of Automattic attended World of Watson, we would represent about two and a half percent of the total attendees. It would be possible that our entire company could represent a rounding error for the organizers.
Scope and Scale: Let the World Be Big
There was a moment when, walking from my hotel room to the conference floor, I looked out one of those huge floor to ceiling windows that are so rare in Las Vegas, and noticed with some surprise that the expansive pool and forest of pool chairs at Mandalay Bay were entirely empty. It took me a moment to realize it was because all of Mandalay Bay’s guests were attending the conference. All of them.
As one last note of comparison, IBM has over 350,000 employees. I had known this intellectually, but it wasn’t until being faced with the expansive actuality of this reality that it really became clear to me. Departments on departments, folks working on all sorts of projects and programs; not to mention the immense amount of spending on outside vendors for all sorts of operations.
I remember talking with someone at the conference, and discussing the leadership workshops that I organize within Automattic. Their response; “Oh, yeah, we do something like that, but we hire an outside firm to put it together and run the sessions.”
This is obvious, right? There’s no good reason it should surprise me – but it did! That there exists this thing, this thing that I do at work in my semi-unstructured time, one of the smaller hats that I wear, that this can be scaled up to create an entire corporate education and training company – it really did take me by surprise.
All of this is to say, to paint a picture, to help illustrate that the way that IBM and other enterprise level companies do business, this was fairly alien to me.
(Past tense here is inappropriate: this is still alien to me – but I’m learning!)
The scale at which these businesses operate is enormous: why this is interesting is personal to me and my experience but might help you to think about you and your experience.
During one of our free-wheeling conversations, my friend Bill once told me, “The world is big, don’t make it small.”
This idea was ringing out to me throughout my time at Watson – I had, intentionally or otherwise, applied blinders to myself, limited my perspective and my ambitions and even my professional circle to the small (and weird!) software and startup world.
When you’re inside that echo chamber, it can really feel like the whole world. Spending three days in Las Vegas in an entirely different, and entirely more enormous echo chamber did worlds to divest me of that notion. The world is big. Let it be big!
Enough about me – let’s talk about IBM, about the World of Watson conference.
One of the big threads running through the conference, and through many of IBM’s Watson-branded products, was their move to make data and data analysis more accessible, to lower the barrier of entry when it comes to working with a company or organization’s warehouse of existing information.
You can see this intent shine through in products like Watson Analytics – once your database is imported to the platform, you can work to better understand the data by asking regular-voice questions in plain language – “How does the month column impact the total sales column?” – leveraging the existing natural language processing power of the Watson infrastructure to interpret plain-language questions into more sophisticated queries.
The question then results in what Watson believes is the best visualization for the information at hand. When it works perfectly, it’s a very slick interface.
The intent here is a good one: bringing understanding of complex information to more folks within an organization is one toward which I am deeply sympathetic. The more competence around these things, the better off we all are, and the better decisions and conversations that we have as a result will surely bring more value to the company.
I also believe that as statistical competence grows within an organization, we can start to set aside some sacred cows, and start to dissolve the notion that Data has all the answers, or is something to be feared. “Statistically significant” does not mean “the right thing to do.”
(I won’t go into a rabbit hole around statistics and good-faith reporting, but if this rings your bell you should check out this blog post from Cathy O’Neil, and the post from Gelman she’s referencing.)
The other side of this argument is that creating a product like Watson Analytics, that is intended takes the hard parts out of data analysis, means we don’t ever actually arrive at that statistical competence. The tool bypasses the previously required underlying knowledge – how sampling works, how to think about statistical significance versus effect size, where that regression curve comes from, why removing those error fields is not a good idea – and so forth.
Increased exposure to and competence around information, statistics and data more broadly is a very good thing, because it increases transparency and allows us to have better, more valuable, more clear conversations.
Creating tooling that allows folks to report on and disseminate information without understanding how that reporting is generated or what might be squishy (or plain wrong!), is dangerous, because it removes transparency and makes our conversations less clear and more difficult to arrive at confidence.
If you’ve worked on a product or project that requires some intake of information, processing that information, and then outputting an actionable take-away, you know how important tidying up your data is, and doing your due diligence on the quality and accuracy of that data.
I think we’d be hard-pressed to find someone working in statistics or data analysis today who doesn’t have a story about a time they had to reverse their recommendation because they discovered too late that some part of their underlying data was incorrect.
A tool that lets us skip past these competencies means that folks won’t always know to double-check their information, or know how to check it even if they’d like to.
Put another way: Unknown unknowns can do damage, invisibly.
I can understand that this is not a problem for IBM; the product today is not the finished, final product. Once these Watson products are truly complete, the platform itself will watch out for bad data, for inconsistencies or avoidable errors, allowing folks who have never stepped foot in Stats 101 to create beautiful, meaningful visualizations to add value to their companies and inform their own work safely and confidently.
That sounds great to me, actually! That is a future I can get on board with.
One question that remains – once the platform is able to check off all of those boxes, why does it need an operator at all?
(We’ll dive more into that question in part three!)
One last point here; analysis is not the end of the simplication / democratization game from the Watson family of products:
They have big plans – I would describe my outlook here as skeptically hopeful.
Here are two stories that were shared by presenters at the conference.
The car manufacturer, Honda, estimates that in the past they have received but were not able to act on something like 500,000 customer complaints per year. The issue was not interest but bandwidth; organizing the reading of these complaints, then tabulating that reading in a useful way, then creating a (possibly!) actionable report, is a huge task, monumental in size but only possibly valuable – and you can’t know if it’s going to yield value until after you’ve already invested in the process.
This is a problem many companies face: they know that they have this information, but due to format or time constraints, they aren’t able to even investigate whether or not there’s valuable stuff in there.
Honda leveraged Watson’s language processing product to turn the unstructured, conversational mountain of complaints into an actionable outcome – more specifically, the cupholder was in the wrong place.
This touches some on the last section; could Honda have hired some Python devs with Natural Language Toolkit experience to do this same analysis, and get the same outcome?
For sure, more than likely that’s the case. But they already leverage IBM, and this product allows folks access to these tools without a developer’s experience – and, presumably, without a developer’s salary.
There was also a presentation from folks at IBM who had worked with an unnamed NFL team – discussing the best way to approach their draft picks. The NFL draft process represents over $1 billion in total salary and bonus payouts, which is mind-boggling to me – but it also means that getting those choices right, or as right as they can be, is a Big Deal.
Especially interesting, to me, is the promise that data-driven player selection in the draft offers moving forward; especially as IBM promises more and more machine learning baked into these products, a sports team will be able to track a player through their entire career, and then feed that back into their draft-picking algorithm – all without a Machine Learning expert on staff. Crazy stuff!
I’m Simon Ouderkirk, I write about small data, remote work, and leadership at s12k.com. If you liked this second part of my World of Watson recap, please do follow my blog via your favorite RSS reader – Part III, my reflections on Philosophy at World of Watson, will be published soon! I’m also on LinkedIn & Twitter
IBM has paid for me to attend World of Watson and provide unbiased coverage of the event. They have not provided content for me to publish, but ask that I do publish regarding the event on blogs and social media in exchange for free admission and travel expenses. My thanks to the Watson Analytics team for inviting me.