Model-based AI and corporate decision making
Digitalisation is not new.
The transformation has been an evolutionary journey. At first, people interacted with a single online system, now, this is just one element in a chain of actions.
We now accept that in enterprise software, we cannot control all processes, especially the touchpoints with the outside world. The world has moved on and software is now designed for digital ecosystems.
Digital ecosystems and the digital enterprise
What is an ecosystem? National Geographic defines an ecosystem as “a geographic area where plants, animals and other organisms, as well as weather and landscapes, work together to form a bubble of life”. It is easy to see why the concept of the digital ecosystem has gained so much ground.
No longer “accountancy with dials”, business software is now an exciting and dynamic world of new technologies, services, business models, and possibilities. Some concepts never take off and others gain traction at bewildering speeds in a cycle of digital innovation. We are now lightyears away from rigid software processes and instead live with dynamic digital ecosystems where relationships between providers, customers and value-adders constantly interact and evolve.
The digitalisation of mass consumer markets has led the way. It has achieved phenomenal amounts of value and generated so much data that it spawned a correspondingly large data science industry – reviving the ambition of data-driven AI, which for a while seemed too complex to become reality.
Complex, engineered assets in the manufactured world
In the engineered world of big, complex industrial assets, digitalisation is challenging and must work in parallel with the real-world constraints of physical, manufactured products – but the transformation is no less radical.
Highly engineered assets generate continuous data. Using the original CAD/CAM design plus all the configuration data from manufacturing and real-time updates, a digital view of the real-world thing is created: a digital twin.
For example, in aviation, every time a modern aircraft flies it generates terabytes of data – reporting the status of the aircraft almost down to the molecular level. Using deep machine learning algorithms, we can crucially predict when it may breakdown or need maintaining. But there is a problem. AI may be excellent at processing huge quantities of engineering data quickly and informing powerful insights, but too many data signals too quickly can swamp an operation and overload a process managed by people. We need to filter what we are being told to make sense of it.
The fact that an aircraft is telling you that it might need repairing at some point soon is not the only consideration. The human decision-maker knows that you also need to look at the bigger picture; like the capacity in a hangar, lead time for spares, available skills, and fundamentally, the need to fly a mission or carry passengers.
In other words, the world is made up of ecosystems, each with their own purpose and objectives. Human decision-makers need a model of how this works over time to make sense of all the data signals they are receiving.
Another challenge is the dimension of time. In aviation, or any other complex asset-based industry, we may make decisions today that can have an impact several years into the future. We may create a plan that is extrapolated from a trend in our data and then something unexpected happens. We ground our fleet for several months because of a pandemic or ash cloud, or we introduce a brand-new technology to reduce our carbon footprint and much of our previous data and learning is invalidated.
Model-based AI and digital twins
It was trying to resolve some of these issues, and the need for more explainable decisions, that has led to the development of a new generation of software under the heading of model-based AI.
The results from this type of software are more transparent and generates conclusions that are properly understood – essential for the corporate world.
At Aerogility, we utilised a type of model-based AI called multi-agent or intelligent agent software and created a highly innovative forecasting and planning software solution.
The software is used to implement holistic models of the real world, representing digital ecosystems and the actors that operate in them. Software agents take on the role of all the assets, processes, and decision-makers in the model and perform their purpose autonomously, interacting with each other to achieve operational goals in their own virtual world. The models are what-if simulations, with different versions of the agents resulting in different outcomes for comparison. The impact of decisions can be seen before they happen, and the side-effects understood, enabling decisions and plans to be optimised and adjusted accordingly.
The models are a variation on the digital twin theme, although they are not trying to be real-time data repositories for individual assets. The models are designed to capture the big picture – where people, systems, equipment and shared infrastructure all interact with each other. This means a management team can explore the future consequences of their actions and optimise different business policies and resource decisions in advance in a safe and trusted virtual environment.
Transparency is the AI world
The transparency of model-based AI and the explicability of simulation results is a game changer for corporate decision-making. We already have autonomous control of vehicles, and potentially a next step is the automation of decision-making between digital ecosystems in the autonomous enterprise.
For this to happen, it is vitally important to have complete confidence in the decisions being made. Gaining this confidence by understanding the optimisation and decision-making process is an important step towards safe and trusted AI – an essential precursor for widespread acceptance in the corporate world.