Everyone is talking about COVID-19, infection rates, and mortality rates. But is anyone thinking about it “infecting” the robots?

COVID-19 is an unprecedented situation. It is challenging every policy, playbook, and expectation related to work and how we engage in our daily work.

Not long ago, the topic of focus was automation: artificial intelligence, machine learning, RPA. The robots were coming to invade us, take our jobs and put millions of people out work.

Sound familiar?

But when the coronavirus hits and collides with the robots, I think the virus might win. At least in its battle with the HR robots.

To explain why, it’s helpful to ground ourselves on what the HR robots really are. For this post, we’re really talking about algorithms generated by machine learning. Bryan Briscoe authored a three part series on machine learning a few years ago. You may want to look back at this post, this post, and this post for a refresher.

For those who don’t want to re-read those posts and who don’t really understand how “machine learning” works, here is the really short version:

  1. Get lots of data (which is easy and cheap these days)
  2. Have a computer process all that data (which is easy and cheap these days) and find patterns that lead to good or bad outcomes
  3. Let the computer engage with the world, and when it spots the patterns of the past have it help you get the good outcomes and avoid the bad ones

If you want a video explanation, I like this one:

As data gathering and computing power have become commodities, the algorithms have stepped in and increased our ability to gain insights and guide decisions. Algorithms are just math – computers doing lots and lots of simple calculations to find what works best. Kids, this is why you need to focus on that distance learning math lesson. Trust me.

Because it’s all just math, the machine is entirely dependent on the data it is given. Artificial intelligence is very smart, but not very creative. It will beat you at Jeopardy, and can even beat you at Go, but it can’t beat my five year old at his ever customized version of Connect Four (because the rules chnge every time he plays). The robots can tell an airline that fewer people book tickets on sunny days and Tuesdays and they are ok with $35 (but not $38) for an aisle seat, but the robots can’t tell the airline when people will want to start travelling again. When the rules change or when past experience is less helpful to make future decisions, the robots will struggle.

There are two key ways in which the Coronavirus is changing the rules and might distort machine learning from an HR perspective (and more broadly).

Strange COVID-19 data may swamp the past

A lot of unusual things are occurring that will distort the data used to train HR predictive models. People are risk averse about leaving jobs (understandably). Peer and manager relationships – a common driver of attrition or staying – are challenged by social distancing. Unemployment might pay a worker more than taking a new job, so some jobs will stay open longer. Healthcare systems are scary places to visit, so fewer “normal” claims are occurring and healthcare costs are unnaturally falling.

In most normal circumstances, algorithms like abnormalities, because it can use that as a signal and prescribe actions. See those 200 people with blue star on their bellies? They tend to be happier at work, so let’s go get more blue-star-belly people! This is possible because the model can compar the blue-bellied people to those without blue-star-bellies. But in this circumstance, there is nothing to compare to. As a truly global pandemic, there is no control or comparison population.

The HR Technology ecosystem will suggest that models will figure this out – that the algorithm will see past the noise in the data. But, as our Bryan’s posts or the video above point out, the robots don’t always know how/why it is drawing the conclusions it reaches, so it will be challenging for the robot to know if it is being influenced. Be wary of any historical data that includes this period of time.

The past may not be as insightful in the “new normal”

Even if the robot figure out that the current days are “bad data” and can throw them out… all models will still assume that past data can inform he future. But it is possible that COVID-19 will permanently alter certain work norms and attitudes, so past data and patterns won’t apply as well in the future. A few exmaples:

  • I’m aware of at least one company that has quantified the retention impact of being physically present with one’s manager – having a local manager meant relatively lower attrition. But what if remote working norms and virtual collaboration change that?
  • Some companies have made conscious choices to support workers in different ways during the crisis. That may change how workers view the need for a company to balance its various stakeholders.


It is entirely possible that all of the smart people in the HR tech space will come up with an algorithm to fix the algorithm. And it’s very possible that the new normal will be very similar to the old normal. But as this plays out, it’s safer to assume that your robot might get infected – so practice some “social distancing” until you can get some test results.


Want to help us understand why people change jobs? Click here to participate in a 5 minute survey about your last job change.