Understanding the difference between ‘black box’ AI and model-based AIUnderstanding the difference between ‘black box’ AI and model-based AI
An opinion piece by the architect of Aerogility, provider of intelligent decision-support systems
June 21, 2022
Artificial intelligence (AI) and machine learning are two related and interconnected technologies that possess the ability to perform infinitely complex and time-intensive tasks such as predictive analysis without the need for human intervention.
Both AI and machine learning have been successfully applied to a wide range of applications, and while they are undoubtedly brilliant computational tools, adopted by many businesses and organizations, they both exhibit a similar, central issue known as the ‘black box.’
Not to be confused with the black boxes dedicated to logging flight data, this type of black box occurs in any computational system where the user can see both the input and the output – but has little oversight or understanding of the operational processes that lie in between. Because of the opacity in the processes – or because of the user’s inability to comprehend those processes – this causes the user to mistrust or doubt the results.
Model-based AI is a process that creates a rich representation of meaning across concepts that can then be manipulated explicitly, traced and tracked. This makes it easy for the user to understand – and therefore have confidence in – the underlying rationale behind the decisions made by the technology.
Model-based AI represents concepts, objects or ideas present in the real world with meaningful computational representations known as ‘agents.’ Each key element in a business or organizational system – such as an asset, a facility, a resource or a decision-maker – can therefore be represented as an agent and configured to act in a proscribed way.
The model is the result of each of these individual agents operating and interacting with each other over a period of time to simulate the overall activity of the operation. The agents create data that can be aggregated into a set of outputs that determine key performance indicators such as cost, availability, capacity or utilization.
By changing the way an agent is configured – for example increasing the operational activity of an aircraft or changing the availability and location of key spares – numerous alternative hypothetical approaches to the operational activity can be examined and compared, using what if?’ analysis.
The aerospace and defense industries increasingly rely on the use of AI to schedule and plan complex, multi-year maintenance and repair operations across thousands of aircraft. Model-based AI allows a user to ask the system a hypothetical question such as: “If engine X or aircraft Y was removed from service, what impact would that have on the rest of the fleet?”
To solve this problem, model-based AI uses labeling to recognize the concept of ‘a flight’ as something we understand within the system, together with its associated relationships - such as between as aircraft and a flight.
The system can house a complex system of entities – for example an aircraft, an engine and a maintenance operative – and represents these computationally. As it is explicit, the relationships between entities can be verified, and therefore, trusted.
Model-based AI vs. black box AI
Black box AI relies on collecting a large amount of ‘Big Data’ based on historical transactions before identifying key patterns in that data, and then simulating how the organization will perform in the future based on trends identified in the data.
The system generates a result, but the user has no oversight or understanding of how the result was attained. This can become particularly problematic when the output demonstrates unusual or outlying results, as it can lead users to a vicious cycle where they become increasingly cautious and distrustful about using this type of technology.
Model-based AI, on the other hand, is a predictive tool that allows the user to interrogate the data and the output conclusions and does so meaningfully. It does not assume the future can be interpolated from the past; it operates according to a behavioral model, therefore the actions are the result of the combination of individual agents interacting together.
The user can configure the capability and behavior required from each agent. Each agent ‘actor’ in the simulation can be examined to see what actions it took, and the justification for such a decision. This can then be compared against observable, real-world situations to see if the agent made the correct decision given the current available options.
A further benefit of model-based AI is that it allows the user to model non-discrete, operational activities and ‘black swan’ events – unforeseen events that stimulate or provoke rapid change and create potential ‘crunch points.’ A good example of this would be if a new contract was taken on at a future date which drastically changes the composition and size of the demand on sustainment. Equally, it could model a scenario when demand increases, then maintenance activity increases, then maintenance capacity maxes out, so bottlenecks form and aircraft groundings spike up.
Why model-based AI should be adopted
For optimal implementation of AI solutions, model-based AI is the key to avoiding the black box problem.
Using a model-based AI approach eradicates the problem of mistrust in the black box, as each decision made by the system can be traced and monitored to provide users with a greater understanding of the reasoning applied. These results are ‘safe’ as there is an underpinning guarantee that it will perform the task that is required.
There is no question that AI can crunch through an immense amount of data to provide results faster than any human actor or actors. However, when life, limb and national security are potentially at stake, those same actors need more than just the ‘what,’ they need the reassurance of the ‘why’ provided by model-based AI that allows them to comprehend, believe and have confidence in the output.