2021-01-14 HR Examiner article John Sumser User Interface Design Ethics in AI Part I stock photo img cc0 by AdobeStock 212760170 ed 544x290px.png

“The essence of traditional interfaces is a deep emphasis on clarity (or intuitiveness). One look at the interface tells you what to do. That doesn’t work with likelihoods. Intelligent output like you’d find in machine learning requires the user to think before deciding.” - John Sumser

 

The Limits Of 20th Century Design Ideas in
AI & Predictive Hiring Software in HR

 

There is a revolution in the kinds of information our machines are giving us. Intelligent tools, such as predictive analytics, can produce a kind of output that was impossible to imagine even five years ago when we were concerned about storage and processing capacity.

 

The ability to gather, process, and relentlessly analyze and reanalyze data makes it possible to generate probabilities for virtually anything. The more underlying data and the more it can be processed, the deeper the potential for understanding.

 

However, predictions are just the odds, based on data, that something may occur. A prediction that is 90% likely to be right is 10% likely to be seriously wrong. Planned events always have a distribution of success probabilities.

 

It also matters what is being predicted. A candidate who meets 90% of the qualifications is not the same as a candidate who is 90% likely to succeed.

 

The former is a quantitative assessment. The latter is the equivalent of Vegas odds. Meeting 90% of the qualifications is a fact. A 90% likelihood of success is a wager. The downside of the wager is complete failure. Even more complicated is the range of probabilities that this particular candidate is 90% qualified.

 

How to present predictive information and user interface is also changing. The essence of traditional interfaces is a deep emphasis on clarity (or intuitiveness). One look at the interface tells you what to do. That doesn’t work with likelihoods.

 

Intelligent output requires the user to think before deciding.

 

This makes the quality of the interface design a fundamental ethical question. If information from intelligent tools is not presented effectively it can result in misunderstandings that directly impact the overall organization, the health and welfare of individual employees, and the sustainability and profitability of the business.

 

When probabilities are presented in current interface forms, they are often misunderstood by the user.

 

If the user misunderstands the output of intelligent tools and makes poor hiring, performance, appraisal, compensation, or conflict resolutions as a result, the impacts on the company, its culture, its workforce and its individual employees may be profoundly negative. The quality of the interface is as important as the quality of the information presented. In intelligent tools, the medium really is the message.

 

In a casino, probabilistic information means that while the odds are in your favor, you can still lose everything. Measurement tells you where you are. Probability tells you how likely it is that you are there. The same is true for a weather report. An 80% likelihood of rain does not mean it will rain.

 

An 80% probability is not a guarantee. A speedometer reading of 70 mph is pretty much a guarantee. That’s the difference between a measurement and a probability. Measurements are exact. Probabilities always include the likelihood that they are completely wrong.

 

Coming up tomorrow in part two of our series on design ethics in AI we’ll take a look at an example of how Greenhouse evolved their user interface design to change the way recruiters and other personnel used and interpreted their predictive hiring analytics data (here’s my initial write-up of Greenhouse from 2016).

 

Read The Entire Series


 
Read previous post:
2021-01-13 HR Examiner article John Sumser HR Tech AI and Intelligent Software Implementation–Part 2 stock photo img cc0 by AdobeStock 300868656 01 ed sq part 2 200px.png
HR Tech: AI and Intelligent Software Implementation – Part II

“From data quality to regulatory compliance, there are 28 key parameters to evaluate when considering a purchase of intelligent tools...

Close