Artificial intelligence has almost as many definitions as the number of books and courses written about it. Many of these definitions may feel a bit abstract, especially to anyone new to the field. The breadth and abstraction of these definitions are all very well for research and academic discussion. But they’re not much use to organizations considering AI as a technology to address their business needs. For the benefit of these organizations, this post keeps it simple: we suggest that AI is an extremely helpful paradigm for automating repetitive decision-making.
So let’s drill down and see why this suggestion is so practical.
What is Decision-making?
Decision-making is the act of selecting between several alternatives. Some decisions have single occurrence, while others are repetitive in nature. Throughout this post we will call each such occurrence a decision-making instance.
Many business processes are, by their very nature, repetitive decision-making processes. Here are few examples:
Ensuring that every user receives the appropriately targeted marketing message (there are several potential messages, so someone has to decide which one to send to each of the many hundreds and thousands of users. Now, that’s what we call repetitive!)
approving or rejecting credit card transactions (there are many, many transactions)
processing insurance claims (e.g., approving or rejecting a claim, or even deciding on the amount that may be paid as a settlement)
initiation of a preventive maintenance procedure on a production machine
deciding on the forces that need to be applied to the joints of a robotic arm so that it completes a maneuver
In the Stone Age, before computers (BC), all organizational decisions were taken by people. Now that we can computerize and automate decisions, we can scale up operations without having to rely on a linear increase in costs. And, in many cases, the decisions are better.
The simplest way to automate decision making is to incorporate the knowledge of a human expert as a means of pre-tailoring the most appropriate alternative (decision) to every potential decision-making instance. For example, in software QA automation, we often define automatic testing procedures. These procedures allows us to continually deploy and integrate new software.
Although extremely helpful, the pre-defined-decisions approach suffers from two main limitations:
a) In many cases, the number of possible decision-making scenarios is huge (and even infinite), which does not allow manual planning
b) The decision-making instances might be so complex that even the most experienced human can’t cope with them. For example, just a few years ago, the computer program AlphaGo defeated a human world champion Go player. The game of Go involves repetitive decision making. Clearly, no human being could have pre-decided a game plan that would beat a world champion
Through the use of Machine Learning algorithms, AI offers a new and powerful way to automate decision making: learning a decision-making policy through inference either from real-life examples or by trial-and-error. For example, AI algorithms can analyze an enormous number of credit-card transactions (which retrospectively can easily be labeled as either fraudulent or legitimate) and learn a surprisingly strong automatic transaction-approval policy. Other AI algorithms can monitor important machine metrics and indicate when something changes (which might imply a need for preventative maintenance), and so on.
Referring to AI as a paradigm for the automation of decision-making can help in validating the need for AI. If you think that your organization might find value in using AI, but you’re not sure, first of all, ask yourself what is the repetitive decision-making process that you want to automate. If you don’t find the answer, look for it. If you still can’t find it, you are probably in one of the following situations: a) don’t need AI technology, or b) need help in translating your business ideas into an AI project.
If you’ve discovered a decision-making process that would benefit from automation, ask yourself how hard and well performing it would be to pre-tailor the right decision for every decision-making scenario. If it’s not going to be too demanding, we warmly recommend you automate it in the old traditional way.
Takeaway: AI is all about automating decision-making through inference. The first step towards delivering business value with AI is to accurately recognize the decision-making process and make sure there’s no easier way of tackling it.