Imagine it’s January 2020. You get an email from LinkedIn with the following headlines:

  • You’re not as happy as you could be in your job. (This is based on your recent search history including articles read, shopping patterns, how often you update your LinkedIn Profile, your credit report, health records, the quality of your group discussion and your life satisfaction score compared to your peers.)
  • The companies you’re looking at won’t make you any happier.
  • Your career growth is being stymied by too much (something) and too little (something else). (This is based on your rate of change, your documented accomplishments, how you spend your time online and what people are saying about you and what you’re saying on social media, among others.)
  • Here’s your current individual balance sheet of strengths and weaknesses (like a career credit report) and what you must do in the next 12 months to meet your personal and professional goals.
  • Here are a few jobs your personal AI advisor suggests you check out. Exploratory phone calls have been arranged and the hiring manager is anxious to chat with you.

Artificial Intelligence is becoming more commonplace, so this type of career email is not that farfetched, especially since recruiters do much of this now manually. However, the dumb way to do it is to use some statistical engine duct-taped to some poorly designed algorithm on top of an assessment test to predict whom to see and hire. Companies are starting to build AI systems like this that, in most cases, will make things worse, not better. For the record, let me suggest a smarter AI approach.

During the holidays I watched Ex Machina describing how AI would take over the world. It seemed pretty realistic although unsettling. A few days later I read an article in Businessweek on how George Hotz developed a driverless car in about 90 days using a combination of expert system modeling and deep learning techniques. This is different and potentially superior to the bottoms up techniques Google and Tesla are using to develop driverless cars.

Collectively this caused me to check out my old copy of Tony Robbins’ 1987 classic, Unlimited Power. In the book Robbins described how he quickly developed a vastly superior training system for the U.S. Army by modeling their best shooters. This shortcut seems parallel to the approach Hotz has taken. Both seem to be examples of using expert systems based on the modeling of best practices rather than starting from scratch. This is where I believe HR should head – modeling how the best people perform, how they get hired and why they get promoted rather than continuing to hire people based on their skills, experience, personality traits and some form of gladiator-based interviewing.

As I’m finishing up this post about using AI for hiring, I get a notification from Google (I think) suggesting I also read this article in the LA Times: Venture capitalist is investing in firms that use computers to improve healthcare. It’s about combining expert systems, deep learning and behavioral patterns to predict healthcare outcomes and treatments. The article itself is about developing AI systems by modeling how the best doctors make decisions.

While this future might sound scary for some job seekers and for those doing the hiring, it actually should be welcomed. How people now go about changing jobs and how companies currently select new hires is even scarier.

Given this, here’s my recommendations on how AI for hiring should be developed:

  • Start by replacing traditional skills and competency-based job descriptions with performance profiles. These performance-based job descriptions define success as a series of 5-6 performance objectives. Google and Gallup already have proven this is the right way to maximize performance, job satisfaction and retention.
  • Assess unknown external candidates the same way known internal people are assessed and promoted since the internal decisions are extremely accurate. Note that they're based on the person’s past performance doing comparable work, not statistics or past behavior.
  • To build more cohesive and stronger teams focus on factors that predict group over individual success. For benchmarking, model how the best coaches and managers build great teams. Note also that coaches get fired for failing. Maybe hiring managers should be, too.
  • Rather than force fit a person into an ill-defined job, customize the job to better fit the person’s motivating needs. This is what Prof. Todd Rose of Harvard is proving in his new book, The End of Average, as a means to maximize individual and company performance.

Properly executed, artificial intelligence offers great promise for raising the overall talent level of a company while providing people with more intrinsically motivating and satisfying jobs and careers. However, this is an unlikely outcome if HR continues its focus on automating existing and fundamentally flawed hiring processes. Ultimately, that’s the difference between smart and dumb AI regardless of the function. For HR, the first offers great promise, the latter self-generating and unstoppable bureaucracy.

That’s scary.