A.S.R. (Artificially Stupid Recruiting)

Recruiting industry analyst Rob McIntosh believes “AI recruiting” is the future of recruiting. Ah, another future of recruiting article! Another tome in the latest assembly line of predictions that the profession will be elevated to the one where Supreme Bot Beings sit atop the Totem of Talent. Alas, being a human recruiter is no longer considered to be sexy enough for the Futurists.

Actually, automated recruiting isn’t what Rob is about, efficient recruiting is. We’ll get back to this notion in a few paragraphs.

It is very easy to be snookered in by the sweet smell of technology, unicorn valuations, 30 under 30 lists – and anything that comes to the masses via Google. But allow me to offer another perspective to this future of recruiting discussion.

Artificial Stupidity

If there’s artificial intelligence, then by logic there has to be artificial stupidity – as in taking the Word of the Bot as Gospel. People, this is a very slippery slope that portends to push aside human expertise, experience, and compassion. Sure, the system can learn, the AlphaGo architecture is a unique extension of previous AI architectures (more about this soon), but humans have a funny way of turning logic on its side and playing a hunch that has a deeper, more robust meaning than any Monte Carlo tree search fueled by “…a huge amount of compute power” can make. Again, more about this later.

Just to be clear, I’m not against #BotRecruiting for the repeatable, scalable types of roles that take up so much of a Sourcer’s or Recruiter’s time, with or without pipelines and talent communities. Just the opposite, I’m very bullish on #AIrecruiting, addressing 68% of all hiring that isn’t significantly specialized (this is the plus or minus one standard deviation of all roles) based upon the range of problems to be solved when on the job.

I’m talking about large-scale hiring in, for instance, customer service, early career sales, retail, warehousing, even entry-level software development, where data is plenty, efficiency is a key goal, and humans are susceptible to injecting bias into the hiring process. I’m also bullish about an #AIrecruiting system that cultivates the digital crumbs and creates “probabilities of behavior” that can be used by human recruiters at some point in the process. We’re getting there tech-wise, but have a long way to move beyond black and white stones.

As promised earlier, one thing that Rob missed out on was describing AlphaGo’s architecture (it’s a darn good read), and why its performance is more evolved than previous attempts at building artificial intelligence solutions to games:

“…it combines a state-of-the-art tree search with two deep neural networks, each of which contains many layers with millions of neuron-like connections. One neural network, the “policy network”, predicts the next move, and is used to narrow the search to consider only the moves most likely to lead to a win. The other neural network, the “value network”, is then used to reduce the depth of the search tree — estimating the winner in each position in place of searching all the way to the end of the game.”

See what AlphaGo is doing with the parallel networks? Assessing hunches…feelings…ESP is what many recruiters call it. With experience, our hunches, fueled by many different scenarios and outcomes from the past, produce a higher probability of the likelihood of success, and with some hunches we learn that they don’t produce a desired outcome. Same with AlphaGo, as long as there’s more computing power to drive the parallel nature of the algorithm.

“Of course, all of this requires a huge amount of compute power, so we made extensive use of Google Cloud Platform, which enables researchers working on AI and Machine Learning to access elastic compute, storage and networking capacity on demand. In addition, new open source libraries for numerical computation using data flow graphs, such as TensorFlow, allow researchers to efficiently deploy the computation needed for deep learning algorithms across multiple CPUs or GPUs”

Let me put this another way: Remember the post-mortem analysis of meteorologists who had tried to predict the path of Hurricane Sandy? The picture below details all the models of possible paths based on a tremendous amount of data collected over decades. These models were created using a variety of simulations running on some of the most brutish computers on the planet. Yet we remember what happened, and the cost? Simulations and predictions are just that.


Each “Hurricane Sandy” adds more data and new sets of rules (learning) that enrich the model and change the hunches. Yet we all know that even the best model results in catastrophic damages. Sometimes the recruiting and hiring of the right person truly is a confluence of hunches, to an experienced recruiter, almost leaps of faith. This #AIrecruiting sure isn’t easy.

In 7 Trends for artificial intelligence in 2016: ‘Like 2015 on steroids’, Andrew Moore, Dean of Carnegie Mellon’s School of Computer Science notes:

“One thing I’m seeing among my own faculty is the realization that we, technologists, computer scientists, engineers who are building AI, have to appeal to someone else to create these programs. When coming up with a driverless car, for example, how does the car decide what to do when an animal comes into the road? When you write the code there’s the question: How much is an animal’s life worth next to a human’s life? Is one human life worth the lives of a billion domestic cats? A million? A thousand? I would hate to be the person writing that code.”

Consider the #AIrecruiting software developer, if they are an avowed animal lover, do they play a Death Race 2000 scenario in their head while coding? This is one of many issues with developing systems to replace or augment humans.

Open source, open stack, and APIs too often mask the fact that there are human beings on the other side of the application. Artificial intelligence, machine learning, intelligence systems, and automation so good that they’ll replace human beings, are not by themselves the seeds of success but are foods that when consumed unchecked further the divide between people and technology. The carrot that is held in front of us, that will have more time to do the things we love, isn’t necessarily reality. Just like the addictiveness of drugs, alcohol, and cigarettes, technology draw us in and not let go. Ask me how often I’m hiking on a lonesome trail only to come across people glued to their smartphones.

Article Continues Below

More data does not mean necessarily translate into better decisions when the human brain is conditioned to trust the technology rather than the brain.

Perfect marriages end in divorce: what’s going to happen when all these perfect Bot selected people begin working with other perfect Bot selected people and one of them farts in a meeting? Or selects “Reply All” with an unfortunate joke? Or votes for Trump/Bernie/Clinton and posts it for the world to see? Or has sex with a co-worker on their desk, then breaks up with them the next week leading to a barrage of social media stupidity? Will the “person” who pushed these Bot hires through be dinged for a bad hire? Or will the #AIrecruiting system be forced to take a timeout?

Back to the use of AI in recruiting: the question to ask is what do we not do especially well right now? Let’s, as a profession, talk now about the rules that govern these tasks and continue to tune them until we reach consensus on best practices. Let’s decide upon where automation in recruiting makes sense for the people who are likely to be impacted by automation.

'Yes, let’s put more human back into recruiting.' Click To Tweet

The journey from predicting the movement of black and white stones to the behavior of people replete with an unending number of human variables is a huge responsibility for our profession. Rather than get all worked up over a technology to replace people, think about how #AIrecruiting can serve to re-focus organizations on how important recruiting should be, to not only the solvency of companies but to the lives of the people we touch. It is the ethical thing for us to do.

Let’s end this tome with the last line of Aldous Huxley’s Brave New World – a fitting be careful what you wish for rejoinder to what happens when you implicitly trust technology:

“Slowly, very slowly, like two unhurried compass needles, the feet turned towards the right; north, north-east, east, south-east, south, south-south-west; then paused, and, after a few seconds, turned as unhurriedly back towards the left. South-south-west, south, south-east, east…”

  • maureensharib
  • Steve Levy

    I’m going to post comments here as well as on FB and Twitter (ok, on LinkedIn too)

    Rob and I had a spirited chat yesterday – about a frog and a pot of boiling water. Right now, we’re the frogs in the pot on the stove that someone is alley heating. See the problem here when it comes to recruiting and automation?

  • Steve Levy
  • Steve Levy

    And Microsoft Research has Society, Ethics, and AI group http://research.microsoft.com/en-us/groups/sea/

  • Steve Levy

    Lisa Rokusek wrote – Anyone interested in ethics and data and algorithims would find @katecrawford an excellent follow on twitter.

  • Steve Levy

    From http://faculty.smcm.edu/…/s13/artificialintelligence.pdf


    Although current AI offers us few ethical issues that are not already present in the design of cars or power plants, the approach of AI algorithms toward more humanlike thought portends predictable complications. Social roles may be filled by AI algorithms, implying new design requirements like transparency and predictability. Sufficiently general AI algorithms may no longer execute in predictable contexts, requiring new kinds of safety assurance and the engineering of artificial ethical considerations. AIs with sufficiently advanced mental states, or the right kind of states, will have moral status, and some may count as persons—though perhaps persons very much unlike the sort that exist now, perhaps governed by different rules. And finally, the prospect of AIs with superhuman intelligence and superhuman abilities presents us with the extraordinary challenge of stating an algorithm that outputs superethical behavior. These challenges may seem visionary, but it seems predictable that we will encounter them; and they are not devoid of suggestions for present‐day research directions.

  • Michael Wilson

    Great article! It’s as if every freaking writer has forgotten the mortgage crisis and the quants and their software algorithms who perfected risk assessments-now they are going to perfect recruiting.

    • Steve Levy

      Preach Michael. Preach…

  • gerrycrispin

    Really well written. Thanks Steve. Other than the title of your article, if I read you right, you are supporting the notion that AI will ‘competently/eventually’ drive hiring practices for jobs that are in high demand with a well defined path (even compassion can be programmed to disposition those not hired and it might be awkward but, not surprisng, if their NPS scores surpassed humans). Assume that there was 1 job filled that way today, when would you predict 100, 1000, 10,000….1,000,000 are filled? I suspect (w/o knowing for sure) that 1000 openings have been/will be completely filled with technology and primitive AI cobbled together this year w 0 or at most minimal oversight…we just haven’t talked to the employers yet and written their story. No more than 4 years to 1,000,000 and at 100 hires per recruiter per year that is a big # of staffing jobs. #rpo-AI

    • Steve Levy

      Gerry, the title is all about marketing…

      If I could predict “when”, I’d be out of recruiting and playing the ponies, lotteries, and the stock market.

      As far as the logarithmic curve you asking me to create of jobs filled algorithmically, in our profession, it really does depend on (a) buzz – which speaks to your point about 1000 openings the first year – and (b) the logic of recruiting leaders to not put the cart before the horse and use automation on roles where there simply isn’t enough of similar problems to be repeatably solved. While (a) is all about the copycat, follow the “leader” mentality that afflicts our profession, (b) requires deep job knowledge by all members of the team who will be creating the process, procedure and policy flows of the jobs to be substantially automated during the sourcing, recruiting, hiring, and onboarding processes.

      The reason I’m hesitant to apply Moore’s Law logic to this problem is that there’s a world a difference between transistors and talent – and I’ve seen Soylent Green far too many times.

      But it will happen – some roles are more amenable to #AIrecruiting, others more to #AIsourcing. IMO, this is one reason why a global recruiting association is essential – a platform to discuss these groundbreaking scenarios and to plan solutions.

      • gerrycrispin


      • chad hatten

        well said

  • Randy Moore

    Great article Steve, I agree we need to use AI in a way to increase the recruiters efficiency, not take the recruiter out of the game. The human aspect of recruiting is critical.

    • Steve Levy

      Appreciate the kind words Randy. I’ll be damned if there isn’t a part of me thinking, “I’m in no way offering my imprimatur for computers to select people…”

  • chad hatten

    ourney from predicting the movement of black and white stones to the behavior of people replete with an unending number of human variables is a huge responsibility for our profession. Rather than get all worked up over a technology to replace people, think about how #AIrecruiting can serve to re-focus organizations on how important recruiting should be, to not only the solvency of companies but to the lives of the people we touch. It is th