DecisionCamp 2019, Decision Manager, AI, and the Future

A few days ago my fellow Red Hatters Mario Fusco, Matteo Mortari, Mark Proctor, Donato Marrazzo and myself (Edson Tirelli) had the opportunity to attend Decision Camp 2019. Following the tradition from previous years, this is a conference focused on Decision Management and related topics, with an emphasis on practitioners, vendors and users of the technology. In other words, a 3-day conference that packs a lot of content, mostly technical and strategic.

This year in particular the agenda was packed full of interesting and relevant topics, ranging from human centric topics (like coordination of collaborative decisions), to compelling use cases (like airport gate scheduling), to glimpses of what is coming on the DMN standard (like temporal reasoning and discussions about DMN 2.0).

Red Hat delivered two presentations: 

• Decision Management + Machine Learning: a standards based approach : Matteo Mortari and myself presented how the purposeful use of the triple crown standards (CMMN, BPMN and DMN) in combination with PMML (Predictive Model Markup Language) enables businesses to leverage machine learning to automate complex decisions in a vendor neutral and effective solution, while promoting transparency and simplifying explainability. 
• How and Why I Turned a Rule Engine into a First-Class Serverless Component : Mario Fusco and Matteo Mortari presented how the use of the latest advancements in JVM, cloud and containerization technologies made it possible to re-architect the Drools project into a cloud-native, first-class serverless component.

You can find a good review of all the presentations on Sandy Kemsley’s blog. All the slide decks are also available for download from the Decision Camp website.

Trends in Enterprise AI and Digital Decisions

One of the highlights of the conference was the keynote presentation by Mike Gualtieri from Forrester, on “Trends in Enterprise AI and Digital Decisions”.

During the presentation he touched on several subjects, past, present and future, but the message was clear: effective Decision Automation (or Digital Decision, Business AI, or any name you choose to use) is not a one-trick solution. It requires a number of technologies to be combined in order to truly deal with today's enterprise challenges. 

AI is a hot term in the market right now, but Machine Learning (ML) without the framework of Digital Decisioning (DD, also called Decision Management) and the optimizations from a Constraint Solver (that he called Mathematical Optimization (MO) engine), is too unpredictable and opaque to be effective. In his words: AI = ML + MO + DD.

His argument is that AI is not only a requirement, but inevitable, for any company looking to become a leader in their industry. He does caveat that statement by explaining he is referring to Pragmatic AI (or Applied AI), that is focused on solving specific problems, and not the Pure AI that we sometimes see on Science Fiction movies.

He continued then explaining that although Pragmatic AI (in the context of Machine Learning)  is a game changer, it is essentially a model, with probabilistic predictions as outputs and the quality of the result depends on both the model as well as the quality of the data used for training. Being essentially probabilistic, in rare occasions it can also produce outlier results that make no sense for the business, or, more often, answers that are sub-optimal.

That is where the combination with Mathematical Optimizations (MO) and Digital Decision (DD) technologies can hugely improve the success of these solutions. Machine Optimizations can be used to constrain or improve the results generated by ML predictive models, eliminating those outliers and improving the sub-optimal decisions. Digital Decisioning can leverage those solutions for explainable automated decisions, while keeping humans embedded on the process as DD is effectively a translation of human knowledge into executable computer models.

Lots of insights and food for thought, but also happy to see that his advice and observations match to some degree what we’ve been doing with the Red Hat Business Automation platform.

Red Hat Business Automation ships with a Mathematical Optimization engine (Optaplanner) and a Decision Engine (Drools) out of the box. Optaplanner leverages business rules for the specification of constraint and scoring rules, and can as well be used for decision optimization and resource planning. The big news we live demoed at Decision Camp is that starting with version 7.5, users can now transparently integrate Machine Learning models through PMML (Predictive Model Markup Language), without any glue code, directly into their DMN (Decision Model and Notation) models. That level of seamless integration is critical to reduce time to market, transparency and efficiency of automated decisions.

All in all, a great conference! Thank you to the organizers for putting together such a strong line up of content.