The ML/AI Reality Check
2020-04-14 | tags: risk AI

Has COVID-19 put your ML/AI efforts under the microscope?

I've worked in the data space since the term "Big Data" was just starting to get its legs. Even then, as much as I enjoyed the work, I saw some things that just didn't add up.

In professional circumstances, I have cautioned people to think through and plan their company's ML/AI efforts in order to maximize their chances of success. In private settings, I've talked about the distance between hype and reality. A delta that large is bound to lead to problems. The data party's still going for now, but I have a hunch the music's about to stop and the house lights are about to switch on.

I've seen this movie before

I'd seen this kind of party fizzle before. It was when the 1990s dot-com boom led to the 1990s dot-bomb bust. Plenty of startups learned what happens when the hype dust settles and the company needs to generate real revenue: business eventually takes off its party garb and gets back to work.

Being appropriately educated ("scarred?") by the past, I've expected that the data world would eventually experience some watershed event that would bring us to reality. It would convince companies to be more strategic, practical, and business-like in how they collect and analyze data. Would it be a large data breach? Maybe the fallout from a company caught doing something underhanded with data?

I just wasn't thinking big enough. I was so focused on the problems inside the ML/AI field that I didn't consider how broad external factors could break up the party.

The reality check

And then COVID-19 hit. Its rapid spread and lack of treatment have led to a sudden and massive shift in how companies operate. This has, in turn, fueled fears of longer-lasting economic impact. Given all this uncertainy, companies are cutting services, scaling back on hiring, and outright downsizing.

Mind you, the virus isn't the root cause of the problem. It's just the stress that revealed the cracks in the system, forcing us to acknowledge the problem. That problem is, quite simply, the delta I mentioned earlier:

The data business isn't living up to the hype.

We've always known this. So many companies hurriedly adopted data (from Big Data, to data science, to machine learning, to AI) but hand-waved through the planning and the hiring. They wanted so badly to say that they had nails, and vendors were more than happy to sell them hammers they couldn't really use. Both ignored the elephant in the room (or, really, the gray rhino in the room) in the hopes that some nails would magically show up before time ran out.

They ignored the voices that said this was all a bad idea.

And this worked. Until it didn't.

Decisions, decisions

So, there we have it: a human virus has led to growing economic uncertainty, which has led companies to proactively tighten their belts to weather the storm. In tense times like these, companies develop this awful habit of sussing out which departments and services do not contribute to revenue.

All of this may have put your data department under the microscope. Possibly even on the chopping block.

What do you do?

Well, that depends on your role. If you're in charge of the company, your job is easy: you ask the person responsible for data -- the CDO, chief data scientist, whomever -- to explain why you've been shelling out cash for their operation. And then you wait.

If you're on the receiving end of that question, you can look to your data strategy for answers. Companies that took the time to plan their data efforts are better prepared to handle the economic crunch, because they already know why they're using ML/AI and how it's helping the company meet its goals.

If you don't have a data strategy, you'll need to quickly sort out:

You may just find that everything is in working order. You may also have to tell your CEO that the company under-planned and over-invested in data science. You'll then get to face some tough leadership decisions, in terms of restructuring, downsizing, or even shutting down.

Your company's data party might be over, but take comfort in knowing that you were able to dance while the music was still playing. Now, it's time for data to take off the party garb and get to work.

Why don't we talk more about risk in AI?

The AI world is sitting on all kinds of risk ... but no one wants to talk about it.

Identifying and Handling Risks in AI Businesses

Business models involving ML/AI are sitting on unrecognized risk.