Chess has seen a resurgence in popularity of late, driven by the rise of internet streamers (shout out to Hikaru and Eric Rosen) and the hugely successful Netflix series The Queen’s Gambit. As someone who’s both a chess fan, as well as holding an interest in the uses of Artificial Intelligence, the world of computer chess engines has long fascinated me. Those of you of a certain age may remember the famous Kasparov vs Deep Blue match, the first time a computer was able to beat the reigning chess world champion. It was seen as a milestone in the advancement of AI, a sign that computers were beginning to be able to compete with humans in what is traditionally seen as one of our great intellectual pursuits. Since then, the power of computer chess engines has grown dramatically, to the point that now an engine running on an every day laptop could beat any human.
The way chess engines have traditionally worked is by using a combination of brute force calculation power (calculating tens of millions of positions a second), and heuristic functions. A heuristic function is a way of evaluating a chess position, by weighing various factors (such as the pawn structure or how safe the king is), which allows the engine to decide on the best move. Crucially, this function, the factors it considers and the way in which evaluates a given position, is entirely defined by the (human) programmer.
In 2017, DeepMind, an AI subsidiary of Alphabet (Google’s parent company), created AlphaZero, a computer program that used deep learning, a type of machine learning based on artificial neural networks. AlphaZero was encoded with the basic rules of chess but wasn’t given any further instructions on how to evaluate what constituted a “good” position. After 9 hours of self-play, where it played 44 million games against itself, constantly learning and improving, it was pitted against Stockfish 8, widely considered to be the strongest chess engine in the world. Over 1000 games, AlphaZero won 155 and lost only 6. Stockfish, the strongest engine in the world, using the combined knowledge and strategies of nearly 1500 years of human analysis, had been beaten 155-6. After 9 hours.
The way in which AlphaZero played was both familiar and extraordinary at the same time. The openings it used would be familiar to any club-level player, demonstrating that humans hadn’t missed a trick when it came to the first 8-10 moves. However, it also did things that no human or other chess engine had considered. For example, it had a tendency to move its bishops back to the first rank, which often looked like a waste of time (or lost tempo for the chess players amongst you) but in the long run paid dividends. In general, it was seen to play more strategically than traditional chess engines, looking at the “big picture” in a way that they never had.
At this point you may be wondering how any of this applies to HR. There’s a lot of excitement about the use of AI within HR and rightly so. Chatbots and Virtual Agents are allowing organisations to interact with their employees at a speed and volume that wasn’t possible before. As these tools get more sophisticated, they’ll begin to bring insights into employee’s wellbeing, career development goals and job satisfaction. HR is about humans, and that will never change. AI shouldn’t replace human interaction, but enhance it; even today process automation is taking the administrative burden away from HR, allowing them to play a more strategic role within the organisation. Combining virtual agents with process automation will eventually lead to a time when activities like onboarding, payroll and absence management are almost exclusively handled by AI.
However, what really excites me is the possibilities that machine learning could bring to light. There is a wealth of data stored on our HR systems, performance ratings, salaries, absence records. And whilst we are able to direct analytics to look for patterns that we can theorise (such as looking for a connection between university degree, salary and performance rating), what if we could throw all the data at a deep learning AI and get it to find connections we’re not even aware of. In the same way that AlphaZero has found a better way to play chess, could a HR-focused AI show us a better way to identify top talent, optimise processes or define our organisation designs?
Unfortunately it’s not quite as easy as that. Chess has a well-defined, relatively simple set of rules that AlphaZero is able to operate in, the real world isn’t as easily defined. There are things you can do to make it easier for AI though. It begins with your data, as that is how the AI is going to interact with our world. Ensuring your data is of a good quality, that you’re capturing pertinent information, keeping it up to date and storing it in a way that is easily understood will allow AI to understand who we are and what we do. Next you need to understand what outcomes you’re looking for (AlphaZero simply wanted to win). This might be wanting to hire more employees who receive the highest performance rating, lowering the attrition rate of your top talent or decreasing your onboarding times.
Finally, you’ll need some AI specialists to help you take that data and turn it into something that an AI can work with to try and achieve those outcomes.
What might we find? Are there HR equivalents of moving the bishop to the first rank? Perhaps employee retention is less dependent on salary or career opportunities than on allowing more breaks during the day? Maybe absences could be reduced by lowering the prices in the staff canteen? The point is, we have no idea what it might discover, but if 1500 years of chess theory can be improved upon in 9 hours, surely there are aspects in the relatively recent discipline of HR that AI can enhance given a little time.
AI will change the way we work in ways that we are only just beginning to understand. What will it help us learn about ourselves?