What do we mean by decisions?

In order to understand the different approaches, tools and architecture nuances in decision automation and management, we firstly need to understand that there are different types of decisions and transactional requirements that require very different approaches, architectures and tools. And that some decisions can be implemented correctly using different approaches. We will get into examples later but first let’s step back and recap some fundamentals of logic. There are three categories of logic; deductive, inductive and abductive. Decision automation attempts to automate all logic but how these are brought together for elicitation and automation are important in tool selection, architectures used and project approaches, skills needed etc. Let’s start with the simplest form, deductive logic.

Deductive Logic aka Rules

Inferences based on truth-preserving rules are deductive. The easiest way to think of these are as “if these conditions exist, then decide to do an action” E.g. If the applicant is under 16 and resident in California, then mark them as ineligible for auto insurance. There are many sources of these deductive rules some internal to an organization and some external. External sources may include regulation or industry standards e.g. an insurance company may have state by state variations in what it’s permitted for eligibility. A medical insurer or provider may want to use specific higher-level industry standard ICD codes for transactions. The conditions defined that lead to deductions are generally simple consisting of mostly AND and OR groupings e.g. If the driver is in (CA or MN) and 16 or older then set as eligible for insurance.

The conditions and rule dependencies can become complex, in some systems the number of deductive rules can become surprisingly large, entangled, with long dependency chains and subsequently difficult to manage. Fortunately, visual techniques, tools and approaches for eliciting, organizing and representing this logic have been developed that make even the largest rules-based systems manageable. See Decision Model Notation

Machine Learning - Inductive Logic

Inferences that MAY produce conclusions that contain new information. Inductive logic being less definitive, can be thought of as educated guesses based on historic observations, contextual knowledge, or imperial formula, largely leveraging probability these areas are commonly referred to as business intelligence, statistics, analytics, AI, algorithms or machine learning. Trend analysis or business intelligence leveraged by subject matter experts can be brought to bear to inform deductive conditions and general decision improvement. Inductive logic is a wide area with a broad landscape of tools, models and techniques and is approached quite differently to deductive type decisions. It is therefore helpful to break down the decision problem at hand first, to better understand what types of decision making, tools and models are needed to most effectively automate with the available data at hand. See Decision modeling with inductive decisions.

For inductive logic data is of great concern because inductive logic is based on historic observations. In the real world however, large enough data samples are often not as freely available. E.g. if you launch a new product or service to a new market, you have no historic data to predict fraudulent activity. Enter the subject matter expert.

The all-important SME - Inductive/Abductive without the data

With inductive decisions, Subject matter experts are invaluable in countering the problem of insufficient or inaccessible data, which is often essential to getting automation initiative ROI quickly. This is leveraged by modeling out their contextual knowledge, inductions and abductions based on real world experience and implementing them as deductive decisions.

Abduction

Kind of inference - Ampliative reasoning

Close connection between optimal strategies and Ampliative reasoning and same with deductive

Definitory Vs Strategic rules

Automation of logic in a business case usually requires combining various forms of logic and types of rules. Rule-governed goal-oriented activities can be easily understood through the analogy of game theory. Definitory being the rules that would define the game. E.g in chess what are the allowed moves a particular piece can make. What defines being in check or check mate. And strategic rules being; what makes a good move in a particular scenario, that helps most to reach the goal of winning. In a business scenario like an insurance claim, definitory rules may be regulation imposed by governments or internal product structures, and strategic rules would be those that helped reduced the loss ratio, like better fraud detection without impacting customer satisfaction.

Ampliative reasoning

Inferences that may produce conclusions -  called ampliative or inductive  formal study know as inductive logic

Epidemic logic – inferences based on notions, “knowing that”