The Reason Psychologists Are the New Gods Of AI-Driven HR

It's not just the data: it's the ability to interpret it accurately. JL

David Creelman reports in PeopleTalkOnline:

Many AI professionals like to believe that if you just give them the data, the power of machine learning will enable them to provide useful predictions. It's not that easy. The lesson from IO psychologists is that human behavior is not that predicable or controllable. You can use the best assessments in the world and still make hiring mistakes. We can analyze a team and recommend steps to improve its dynamics, but there is no guarantee those steps will boost performance. AI will make assessments and predictions cheaper and faster than human experts. It remains to be seen if they can make significant strides in the accuracy of predictions of human performance.
I’ve spoken to several HR technology vendors specializing in AI and they talk in hushed tones about an ancient group: the IO psychologists.
Industrial organizational (IO) psychologists have been with us a long time and they tend to fly under the radar because their work is slow, it’s cautious and it avoids eye-catching fads. Now, perhaps much to their surprise, their decades of slow, cautious work has suddenly come to be appreciated by AI specialists.
Why is that?

AI Struggles with Complexities of Human Data

Many AI professionals like to believe that if you just give them the data, then the incredible power of machine learning will enable them to provide useful predictions.
For example, they might say, “Send me all the data on your employees and their performance and then I’ll be able to predict future high performers.”
Unfortunately, it’s not that easy.

The Problems with People Data

There are many problems with people data that make AI applications tricky. The sort of problems that can arise include:

• Small Data Sets: It’s rare for HR to genuinely have big data; data scientists may be forced to work with data sets that are too small for their methodologies to be reliable.
• Inaccurate Data: The data may not be accurate. For example, many companies don’t have accurate date on how long it takes to fill jobs since the information was never captured correctly.
• Comparing Apples to Mangoes: Even an analysis of something seemingly well-defined such as “productivity of programmers” may flounder because there are many different types of programmers.
• Restriction of Range: If differences in the variable you are trying to predict are small, then it can be hard to build an algorithm. For example, if most employees have a performance rating of either 3 or 4, then that does give an AI much to work with.
• Overfitting the Data: It can be tempting for data scientists to take a data set and build an algorithm that predicts the output based on the input. The problem is that you may develop an algorithm that works beautifully with that one data set but doesn’t work anywhere else.
• Bias: HR professionals are alert to the fact that pre-existing bias can easily end up inside an algorithm.

The great thing about IO psychologists is that since they have been grappling with these problems for decades, they know what problems to look out for and how to get the best insights based on the data they have.

What HR Technology Vendors are Doing

Many HR technology vendors are relying on the work of academics. For example, Recepiviti’s culture assessments are grounded in the work of social psychologist Prof. James W. Pennebaker (author of The Secret Life of Pronouns); HireVue has a chief IO psychologist (Nathan Mondragon); and Knockri’s assessment tool is informed by the work Robert Gibby, a Ph.D. in IO Psych who is the chief talent scientist, talent acquisition at IBM.
The insights from the academic literature is being built into HR technology so that it can be widely and cheaply make available to HR.

The Hardest Lesson of All

While I’m a great believer in the value of AI tools for HR, perhaps the hardest lesson from IO psychologists is that human behaviour is simply not that predicable or controllable. You can use all the best assessments in the world and still make hiring mistakes. We can analyze a team and recommend steps to improve its dynamics, but there is no guarantee that those steps will lead to a big boost in performance.
IO psychologists have learned what doesn’t work and what tends to work. That’s the best we can do. AI will certainly make assessments and predictions much cheaper and faster than human experts—that will have a huge impact. However, it remains to be seen if they can make significant strides in the accuracy of their predictions of human performance.