Operations Note # 2 – The Human Art of Separating AI Fact from Myth

Is Skynet here? For those who are not fans, Skynet was the terrifying Artificial Intelligence from the Terminator series of movies. Responding to my own question: Skynet is definitely not here. Artificial Intelligence is, though.

Artificial Intelligence, or AI, has made enormous strides in the past several years. DeepMind (now a part of Google) made headlines in 2015 and 2016 when its AlphaGo mastered the game of Go enough to beat world-class human players. It did so through the use of Monte Carlo algorithms, which are in essence repeated simulations based on assigned probabilities. In the years following, this success has been built upon through DeepMind’s subsequent development of AlphaZero, which has mastered not only Go, but also Chess and Shogi. This latter achievement was done through a fundamentally different approach in which the system is programmed with the rules of the game and then teaches itself strategy by playing against itself over and over, millions of times to see what works and what does not. Each game played helps it refine its approach. Practice makes perfect.

Does this apply to operations? Well, no. And, yes. This technology is expensive to implement and energy intensive for what it is accomplishing. I would advise against rushing out the door to apply it right now (October of 2020). The operative words in that last sentence, though, are “right now.”

We are seeing other, different real world applications of AI that are immediately available. For example, IBM’s famous Watson that once won Jeopardy is now being applied to practical business needs such as organizing information across a business and making better chatbots. After the excitement of Chess and Go, this may sound anticlimactic, but it is not.

This is a very serious development that fundamentally impacts how organizations interact with customers or the general public while controlling costs. We have all interacted with ineffective chatbots that subtly fail to direct us to what we need until we type “GET ME A HUMAN!!!!” or, in the case of their voice activated phone tree cousins, yell that into the phone. Building a better chatbot involves a sophisticated interplay of standard software development, natural language processing, and AI, but it pays off with happier customers.

So what we are seeing is that AI very much applies to operations, but that it is used in a more mundane fashion than would be expected by some of the breathless proclamations that inevitably flow from hype. This is not to deflate the importance of AI. Imagine being able to tell your great grandparents decades ago that they would be talking to a machine and that it would more or less correctly answer your questions. Imagine now that you told them on April 1st. What was science fiction not too long ago is daily life now.

Where AI is best applied in today’s environment is in addressing well defined, repetitive, and relatively simple problems. Or put into other words, in this case from an article specific to financial auditing:

"That’s where AI will come in. The audit function has historically been on the leading edge of the AI development and adoption curve. This makes sense, of course, because auditing contains a great deal of repetition, with auditors spending their days chasing data from clients, cross-checking that data with transaction records and inputting the results into an auditing software system. These types of tasks read like a playbook for AI automation, and – in many cases – firms have succeeded in streamlining the process with AI."

The article from which this paragraph is drawn is a recent Forbes article by Brian Peccarelli of Thomson Reuters discussing how AI will be a natural technology to help as the Big 4 accounting firms need to separate their audit functions from other functions by June 2024 as required by the Financial Report Council, a UK government body. The separation of functions and financial reporting resulting from this decision will also negatively impact technological synergies, making AI’s ability to take over many mundane tasks of auditing a useful trait for those firms as they look to regain that efficiency.

We see a similar adoption of AI in the legal profession to help with the discovery process as well as such tasks as “reviewing documents, managing contracts, predicting case and sentencing outcomes, and even automating tasks like parking ticket disputes,” as is described by the linked article by Ajith Samuel in Law Technology Today.

The best advice I can offer in regard to looking at AI is to not believe the hype, but to watch the hype nonetheless because a significant percentage of it will become reality sooner rather than later. If you need an example, look no further than self-driving cars. Who knows? It may only be a year or two before someone packages AlphaZero’s learning approach into something your team can configure for application to well defined functions, such as logistics, commodities marketing, or allocation of scarce production capacity. You will want to know what is coming so as your business changes and AI evolves, you can wisely anticipate how to adapt your strategies to leverage innovative technologies.

In the short term, examine what is being successfully implemented elsewhere to well defined situations similar to what your organization faces. Use what you learn to drive a conversation of where AI can realistically add value to your organization. In both of the above quotations the business problems, like the rules of Chess or Go, are clear and unambiguous. If you can clearly define and delineate your business problem, determine that those problems are characteristic of those in which AI functions well, and you have a volume of work that justifies adopting these technologies, then by all means look into AI.

Finally, be sure that your team puts forth the effort to understand AI and other technologies. As systems get better at training themselves, their maddening opacity as to how they learned what they learned opens up nontrivial dangers to your organization. AI is susceptible to manipulation, sometimes in ways that can backfire spectacularly on you. If such poor decision making can be induced into AI systems deliberately, it can also probably be induced accidentally in ways that have not yet been foreseen. Your organization will need to have the wisdom to build upon the profound potential of AI while controlling risk against dangers that are still not understood.

I know that this advice on how to best apply AI to operational excellence is unfortunately vague and open to interpretation. That is unavoidable. How you choose to apply AI is just the kind of decision, in fact, at which AI would fail. It is a decision requiring human judgment.

Please also check out my previous Operations Note:

• Operations Note # 1 – The Power of Direct Observation