How to use game-changing AI to boost decision quality

It’s time to augment humans in business decision making

Note: this article is co-authored with @Tom Savels and originally published on Medium

There's a massive hype around the potential value of AI in business applications. Based on our industry experience, most of that hype is misguided. In many success stories, the "AI hero" only acts as a polishing tool, used to tweak existing processes. This is not because of AI limitations - it's due to lack of imagination. 

Instead of using AI as an add-on, we should be re-designing our businesses around the latest AI. We'll show that Game AI is particularly suited for this. That business processes can be seen as games. That decisions are like moves in a game. And that winning games is what the latest AI excels at. 

More than any other tech, Game AI has the potential to drive business transformation. 

This article serves as a prologue to a series of case studies. In each study, we'll share an example of Game AI supporting complex business decisions. Before going into the details of these studies, we'll start with this introduction. And show why Game AI for business decisions makes so much sense.

Let's dive in.

Data versus Decisions

In the debate about AI for business, there is a lot of focus on the data, but data by itself doesn't generate value. Businesses generate value by turning data into decisions. To quote Cassie Kozyrkov, head of Decision Intelligence at Google: 

“Data are beautiful, but it’s decisions that are important. It’s through our decisions — our actions — that we affect the world around us”. 

It’s through our decisions that we create value, or leave value on the table. 

High stakes, Ancient tools

Now think of your own experience in complex business decisions. Have you ever made or supported decisions that left you deeply uncomfortable? Feeling nervous that the analysis didn't cover all options? Knowing that there are more opinions and risks to understand? Your confidence affected because the input data is so uncertain? And most painful, realizing that time has run out? We sure have.

Many business leaders recognize this feeling. Yet, most have accepted that important decisions are based on incomplete analyses. They've accepted that their decisions are driven by gut feel, consensus, and human bias.

So, if you're dealing with complex decisions in your own business, ask yourself this. How confident are you in the quality of your decisions, if Excel is the most sophisticated tool you have? If the tech in your phone is more advanced than the tools that drive your decisions, something is wrong.

Enter Games

Decisions, and strategies to optimize them have been studied extensively by Game Theory. To a game theorist, a Game is something like: 

“… interaction between two or more people where the outcomes of the interaction depend on what everybody does and everybody has varying levels of happiness for the different outcomes.”

Now - that sounds like most business interactions. To a Game theorist, business decisions represent moves in a Game. And over the years, winning Games is what AI has been excelling at. 

So, what we imagined is this: 

Can we use Game AI to make winning business decisions? 

It turns out, you can.

We will show how business problems can be turned into Games. Mastering those Games with AI leads to decision quality beyond human capability. This can be invaluable - especially in high stake business decisions. 

Mastering Games with AI - a brief history

It's fascinating to consider the history of AI through the lens of Games and their complexity. As AI evolved, AI bots have learned to master more complex Games - and support more complex decisions. For those interested, we recommend “A brief history of Game AI up to AlphaGo” by Andrey Kurenkov.

Checkers and Chess were the first board games targeted by IBM to showcase AI capability. In 1996, Deep Blue was the first computer to defeat a world champion in Chess. Deep Blue’s skill level was lower than top humans. Yet, its ability to analyze millions of moves secured victory against Garry Kasparov. 

A next famous showcase for AI research was the game of Go. Go is notoriously hard because of the large number of board positions - far exceeding those of Chess. Google-owned DeepMind created AlphaGo, beating one of the world’s top players in 2016. They then moved on and built AlphaGo Zero, beating its predecessor 100 to 0. Where AlphaGo was designed around human experience, Zero mastered Go completely through self-play.

Board games do not always reflect real-world decisions though. They are turn-based, so there's no pressure to make decisions in a changing environment. Also, in board games, the AI has complete information about the environment. There's no guessing or risk-taking based on unknown factors.

The video game StarCraft emerged as the next “grand challenge” for AI research. Winning at StarCraft requires players to balance short- and long term goals. It also asks for an ability to adapt to unexpected situations. Late 2018 DeepMind’s AlphaStar was the first AI to beat a top professional player in this game.

Next, in July 2019, a team of Facebook and Carnegie Mellon University researchers built an AI called Pluribus. Pluribus beat top pros at 6-player Poker, indicating another breakthrough in AI research. What sets this victory apart is the AI's ability to learn how to bluff - a skill thought of as particularly human. Also, Pluribus ran on a laptop, a long way from Deep Blue's supercomputer.

Big Tech understands the business potential of Game AI

So why do companies such as IBM, Google, and Facebook invest so much time and money in developing Game AI? Fortune noted:

“The development … of Pluribus … has potential implications for the business world where, as in poker, people need to make strategic choices amid great uncertainty and where outcomes are often not zero-sum.”

The same AI that powers Pluribus may outperform humans in real-world strategic decisions. For instance, political negotiations and contract bidding share many similarities to Poker. What they have in common is that AI has to take risks and make short-term sacrifices for long-term gains. 

DeepMind applied AlphaGo Zero techniques to improve the British power grid efficiency. It applied the same algorithms to reduce the electricity cost of Google’s data centers.

Big Tech understands that Game AI has massive potential value in real-world applications. Assisting humans in complex decision making is one obvious high-value application. Why? Because many companies still rely on flawed human analyses for high stake decisions.

Decision Quality … or not? Flaws in human decision analyses

We have worked for most of our careers in the energy industry. Decisions in this industry are almost always very complex. Decision-makers often face many options, uncertainties, risks, opinions, and value drivers. And the consequences of making the wrong decisions are often very high.

So to help decision-makers, the industry has adopted a tool to assess the quality of decisions. The Decision Quality framework is a structured approach to analyze decisions. It was developed in the '60s and can be thought of like a checklist to map all facets leading to a decision.

Having such a checklist is great, but of limited use if we populate it with poor or incomplete input. Too often, we rely on slow, iterative and manual workflows. Our workflows prevent us from exploring the full option space. They limit our understanding of risk and value trade-offs. We usually have an outdated toolset and more often than not, we introduce various types of bias and run out of time.

Let's look at an example.

Flawed decision making in the real world

Assume you're in charge of the maintenance of a refining operation. You're facing 50 independent maintenance activities that need scheduling. Only one can be executed at any time. This gives about 10^64 different permutations for the planning team to work through. Each activity has different priorities about safety versus production criticality. Some activities need specialized skills that come with a high mobilization cost. To make matters worse, you're facing ongoing inspections. These may lead to a change in priorities or extra daily maintenance activities. And what is at stake? Suboptimal planning may lead to significant safety exposure, higher cost or unnecessary deferment. In a plant that generates millions of dollars of revenue every day, this is a big deal.

Most industrial assets have maintenance scope that includes many more than 50 activities. The number of permutations easily exceeds the possible board combinations in Go. Yet, the most sophisticated tool your planning team has is an Excel spreadsheet.

So as a maintenance manager, your job is to decide on the best maintenance strategy. Following Decision Quality principles, you aim to explore alternative scenarios. You wish to understand trade-offs between safety, cost, and deferment. Yet, asking this from your planning team is impossible - they are just not set up to give those answers.

So what do you do? You rely on intuition and experience (and bias) of the planners. Everyone knows that the resulting plans are suboptimal, but this is accepted. No-one knows how to improve - since no-one knows what good looks like.

Rethinking jobs around new technology

Many companies still rely on the same tools and processes to support complex decisions as in the '80s. Excel is still the dominant tool to support decision making. As @Tom Goodwin wrote: 

"As much as we talk about AI taking jobs, the reality is that almost every job that primarily uses Excel could have been replaced or augmented by rudimentary software about 10 years ago. The issue is that we add technology to old processes, we don’t rethink what jobs should be around new technology."

Imagine how different jobs would be if redesigned around the latest technology. How much more satisfying, interesting and fun that would be. Teams would have more time to think strategically. They would be able to test trade-offs and to actually help decision making. Would decisions have higher quality, and the plans generate more value? Absolutely - and often with step-change improvements.

What does this mean to you? Challenge the status quo.

Suppose you'd start a new company today. You need good plans to make strategic decisions. You also know that your planning team faces more permutations than atoms in the universe. Would you give them an Excel spreadsheet as their best tool? Or, would you choose something more sophisticated to help your decision making?

We recommend that you take your most complex business problem and take it to the drawing board. Give yourself some mental white space to think. Challenge yourself how to fundamentally redesign and improve the underlying processes. And re-design them around the latest technology. 

If Big Tech is using Game AI to make better decisions, why wouldn't you?

This article is the introduction to a series. In each next article, we'll share a case study of Game AI supporting business decisions. Business decisions are like complex games in many ways. The same techniques can improve decision speed and quality. 

Of course, there will be resistance to change, at all levels in the organization. So, we would like to touch upon change management as well. The fear of “redundancy” will be in the minds of many professionals. To quote Cassie Kozyrkov again: 

“… if you think that AI takes the human out of the equation, think again! All technology is a reflection of its creators and systems that operate at scale can amplify human shortcomings, which is one reason why developing decision intelligence skills is so necessary for responsible AI leadership.”

Most complex decisions do not have a mathematical optimum. Trade-offs between risk and reward are often driving the choice. Experience, facilitation and gut feel will continue to play a role in decision making. Humans will need to remain in control. Yet, the right AI can help unveil the full option space. It can translate different opinions into alternatives. And it can quantify those trade-offs. 

That AI is already available, thanks to the role of games in pushing AI research forward. 

We are very interested to hear your views. Feel free to reach out on [email protected] or comment on this article.