There’re claims that Artificial Intelligence (AI) will make transport safer. It’s to put a positive spin on the introduction of AI. Implying that existing safety deficiencies can be addressed with the power of AI.
It’s difficult to disagree with this simple assertion. There’s a list of risks that continue to be troubling. With directed design effort there are functions that AI can perform that mean it can have an advantage over conventional systems. With good design, no doubt high performing systems can be constructed.
In aviation, for example, if I consider the top five fatality risks, there’s a persistence of specific categories. We never seem to get away from loss of control in-flight (LOC-I) being high on that grim list. Runway related issues persist, and the hardy perennial of mid-air collision remains. Over the years progress has been made addressing controlled flight into terrain (CFIT), but that category of destructive events never disappears.
It’s fascinating to see that the industry thinks that AI itself is a risk[1]. High probability but low impact. This is considering a broad description of risk rather than a safety focus. Here the concern is related to the difficulties of practical implementation of this new technology.
Marketing people will big up the possibilities brought about by AI. This is what’s going on in relation to the most recent mid-air collision fatal accident. With sound justification given how crude elements of air traffic management are in specific locations.
We will never entirely displace “see and avoid” as a means of collision avoidance. Scanning the horizon looking for other air users. In my opinion, relying on this technique in relatively busy traffic areas is unwise, to say the least. This is where airborne AI assistants have much operational safety potential. Sucking up multiple information sources and processing masses of information to give accurate and instant advice. Such systems can be designed to give real-time updates not only to improve situation awareness but give avoiding action guidance, or even automated responses.
Let’s get back to the general assertion that AI will make aviation safer. On this one I’d be more cautious. For example, looking at LOC-I incidents and accidents there’s a complex mix of causal factors, and circumstantial factors. In addition, there’s the complexity of potential recovery actions too. Solving problems in 4-dimentions whatever the weather, whatever any damage incurred and however pilots react. This is where the probability numbers start to stack up.
That catch all disciplines “human factors” makes outcomes particularly difficult to calculate. Accidents are known where pilots and automation fight each other to produce bad outcomes.
AI is a machine. It will speedily crunch numbers in a mechanical manner. An extremely advanced manner but without emotion or, yet, not matching the imaginative capabilities of the human brain. Or for that matter the sophistication of human senses.
Would exceptional capable AI have saved Swissair Flight 111[2], for example? Sadly, I think not. On the day, likely an automated airborne system would have made the same decisions as the pilots. Decision making without the sense of precisely how the aircraft fire was developing would still have been hamstrung. I could raise other cases too.
Will AI make transport safer. In part. Not as a universal cure all.
[1] https://www.iata.org/en/publications/economics/reports/risks-2025-brief/
[2] https://www.bst-tsb.gc.ca/eng/rapports-reports/aviation/1998/a98h0003/a98h0003.html
John Vincent 