Assistive AI, Not Autonomous AI, Is The Path To Improved Operational Efficiencies
In my 30 years of experience designing and delivering large-scale, high-profile software, I’ve served as architect and director of engineering at Microsoft Research and founded the Knowledge and Reasoning group within Bing. Most recently, I co-founded and currently serve as president of Maana, a company that has pioneered digital knowledge technology and is helping some of the world’s largest industrial companies achieve their digital transformation goals.
The consistent objective of these organizations is to minimize costs and maximize benefits of each individual facet of their operations. To achieve this goal, they are increasingly turning to artificial intelligence-based solutions. When looking at AI, decision makers must distinguish between autonomous (or automated) and assistive (or augmented) solutions. My goal in this article is to explain the difference between these two forms of AI, especially as it pertains to accelerating the demonstration of business value for optimizing operations.
Assistive AI Helps Answer Higher-Level Business Questions
Fully autonomous solutions have a place, but they are generally quite a ways away from being able to support higher-level business operations. Organizations often use autonomous AI at the edge of their networks to automate the operation of remote field equipment. Automated AI allows for analysis of sensor data from these machines and automates the operation of equipment such as machine shutoff when those sensors detect a potential failure. Organizations are not quite ready to hand over important operations and business decisions to automated AI.
Assistive solutions, on the other hand, support humans making complex decisions by assisting with the following processes. We’ll use the example of a drilling engineer:
- Observing the world: In this case, AI can literally look around at years of operational and geological data related to a company’s wells and drilling rights, as well as public data such as geophysical data related to the land.
- Reasoning about it to move toward some goal or goals:“Reasoning” describes how a question of whether to drill an exploratory well on one of these leases is answered. Assistive AI allows a human to ask literal questions of a machine in an attempt to derive answers to their problem.
- Making justified and explainable recommendations to users: AI algorithms then create models that provide answers to those questions and can help the drilling engineer recommend whether to drill an exploratory well based on 10 years’ worth of data. A human can not in any reasonable time evaluate and analyze that much data, but AI models provide those recommendations to the drilling engineer.
- Supporting the user to explore hypotheticals: Based on the results of this exploratory well, the machine can help the engineer look at what returns a production well might provide on the investment.
- Recording decisions and actions taken: The AI algorithms record the decisions and actions taken by the drilling engineer and then the engineer either accepts the recommendation made by the models or rejects them and makes his or her own decision.
- Measuring impacts and outcomes to learn and improve over time: AI’s greatest value is its learning attribute. It measures the outcome and KPIs based on the decisions and actions taken by the drilling engineer and uses that to learn and improve recommendations for better impact on the KPIs. The reasoning process builds up a bank of knowledge — an answer (solution) to a question (problem).
Assistive AI helps humans make better and faster decisions, but it uses human inputs, as well as existing data, to provide the answers. For example, it can mathematically encode the decision process a drilling engineer goes through when making his or her assessment on whether to drill. Ultimately, it is the subject-matter expert who makes the final decision. It is about augmenting the decision the human is making using large volumes of data.
Models, Not Static IT Applications, Offer Improvements In Operational Efficiencies
The goal of any organization is to tackle problems that deliver major business value (minimize costs and maximize profits) by achieving operational efficiencies. To best integrate AI into the process of delivering business value, leaders must shift their organizations’ mindset from traditional IT “static” applications and machine learning and analytics, using specific trained models, to a perspective that supports many interacting models at multiple levels working together to optimize the enterprise, learning and improving over time.
Historically, CIOs run an organization’s infrastructure — knowledge management, enterprise search, SharePoint, etc. Those are all static experiences rooted in IT applications. Implementing AI with this mindset only perpetuates a reliance on simply seeing information or asking a machine to present a visualization to help make a decision. Instead, leaders must push for a shift in thinking toward knowledge models that provide recommendations.
In the drilling engineer example, each part of the decision process represents something that a machine can model with human assistance and answer. The relationships between those models help demonstrate to decision makers the interconnectedness of the processes they must improve.
Assistive AI Helps Humans Demonstrate Business Value More Efficiently
Business units put forward projections designed to demonstrate how those units will deliver shareholder value. From there, leaders can identify specific decision flows that have the greatest impact on gross margins and leverage AI to digitize those decisions flows for greater operational efficiencies. This application of digitizing decision flows is the essence of digital transformation.
When it comes to digital transformation and optimizing operations, assistive AI delivers the greatest and fastest value. Organizations attempting to digitally transform their operations should focus on this type of AI, where subject-matter experts can mathematically encode their knowledge and combine that with the most relevant data. Machines can aid in the consolidation and analysis of these various operational facets and ultimately help improve operational efficiencies faster than humans alone can.