The arrow of time. We fly from past, to live in the present and anticipate the future. Sir Isaac Newton would be proud of us. By unravelling laws, that where always there, the means to anticipate the future was illuminated.
In civil aviation, we have devised and grown a whole regulatory system that depends on learning from the past, doing calculations today and flying with a belief that we know what’s going to happen next. Flying is predicated on a reasonable degree of predictability. There’s clear logic in this way of thinking. Just imagine powering up a couple of massive jet engines and starting a take-off roll without being extremely confident that at a certain speed the laws of physics will do their part and the ground is left behind.
We don’t establish a reasonable degree of predictability by looking at a crystal ball or taking up alchemy. Yes, we do still depend on reasoned expert opinion in addition to doing calculations. The minute those expert opinions start to shift away from grounded reasoning and careful deliberation then danger is afoot. This is one of the arguments for treading carefully when political opinions start to come to the fore. The laws of physics are not established by a public opinion poll. Nevertheless, it’s equally polarising to say that there’s no political dimension in the aviation regulatory system.
Anyway, that’s not the subject that was on my mind. Conversations about Artificial Intelligence (AI) are more prolific than those about self-help books. Even the shelves of popular high street bookstores are starting to fill up. The non-fiction titles with AI, either as the main subject or as an adjunct are numerous. It’s the fashion to write something literate or purely speculative.
I’ve mentioned the word “determinism” before. It can be interpreted philosophically or in a more scientific and technical manner. Determinism is a belief in the inevitability of causation. That chain of cause and effect that is so familiar to anyone reading an aviation accident report.
Understanding what causes something to happen in a moment in time goes back to my initial subject of a reasonable degree of predictability. In aircraft certification, no matter how complex the system, when presented with a system safety assessment we expect a comprehensive and reasoned set of statement. Predictions about the “what ifs”. What if an aircraft part fails and what happens next? What happens in combination with other failures?
This is where AI is potentially problematic. All the reasoned arguments in the world go out of the window if a system, subject to the same conditions, behaves one way on a Monday and differently on a Friday. Not to mention the weekend. I could say, AI is remarkably human in that respect.
The subject that was on my mind is not the inner working of complex aircraft systems. Certification experts are on that one. It’s possible to put boundaries around the behaviour of some aircraft systems. What’s more fascinating is the evolution of AI interactions with us mere mortals.
Let’s say I have the responsibility for return to service of a transport aircraft that has been subject to maintenance. A pile of documentation will provide the evidence that the work conducted has been correctly completed. It conforms. Amongst that paperwork might be an output from an AI driven diagnostic system that flashes a green light to say everything is fine.
Now, playing with the “what ifs”. What if it’s not fine given that the conditions experienced were way outside the AI systems training and it does a creative hallucination. The person signing the release to service documentation would have no idea or facility to question the green light. But it’s their signature that matters in the process of return to service.
There is a point of concern.
POST: There’s a lot going on out there Enhancing aviation safety with artificial intelligence: A systematic literature review on recent advances, challenges and future perspectives – ScienceDirect
John Vincent 