Everyone Wants Experience. Who Will Create It?
AI may remove some of the junior work through which engineers traditionally gain experience. But the same technology could help us develop engineering expertise more deliberately and perhaps faster.
Modern aircraft can fly more precisely and consistently on autopilot than humans for much of a journey. Yet pilots must still be able to take over when automation reaches its limits.
Automation removes many of the tasks humans once performed routinely while leaving them responsible for monitoring the system and handling difficult failures. In doing so, it can also remove opportunities to acquire and maintain the skills needed when automation reaches its limits.
Lisanne Bainbridge described this problem more than forty years ago in her 1983 paper Ironies of Automation.
Aviation responded by making experience deliberate. Pilots train in simulators, practise abnormal situations, undergo recurrent training and progress through supervised operations. Aviation creates those experiences deliberately through training rather than relying on rare failures to provide them.
Engineering may soon face its own version of the pilot problem.
The hidden output of junior work
For decades, junior engineering work produced two things.
The visible output was an engineering result.
The other output was a more experienced engineer.
Junior engineers performed calculations, prepared reports, investigated problems and gradually took on more difficult work. Through repetition, feedback, mistakes and increasing responsibility, they developed the judgement expected of experienced engineers.
Generative AI changes this equation.
AI can already perform parts of that work at very low marginal cost. Even after including the time of an experienced engineer to instruct the AI and review its output, the combination may sometimes be cheaper or faster than assigning the work to a newcomer and supervising them.
The project still needs the engineering result. It may no longer need the junior engineer to produce it.
That could remove part of the mechanism through which engineers have traditionally developed experience.
The validation problem
For a while, this may not create an obvious problem. Today’s experienced engineers already possess much of the knowledge needed to supervise AI. They can benefit from its productivity while recognising when its output is wrong.
A calculation can be mathematically correct and still use the wrong load case. A requirement can be perfectly written while missing the safety intent. A certification argument can look convincing while overlooking an interaction that an experienced engineer has learned to question.
The harder question is how somebody develops that judgement if AI performs more of the work through which previous generations acquired it.
Experienced engineers still have to become experienced somewhere.
Can AI help create experience faster?
Perhaps the old apprenticeship model was never particularly efficient.
We gave newcomers work because the work needed doing. Some of those activities eventually developed expertise. Learning was largely a by-product of production.
Once AI begins to separate production from learning, organisations are forced to ask a question that was easier to ignore before:
If our objective is to turn a graduate into an expert engineer, what experiences should we actually give them?
The answer may not be thousands of hours spent producing routine deliverables.
AI could create technical cases specifically designed to develop judgement. A junior engineer could investigate deliberately flawed analyses, challenge requirements with hidden inconsistencies, choose between plausible design alternatives or defend assumptions against an AI acting as a critical reviewer.
The system could vary the cases, introduce unexpected information and adapt their difficulty as the engineer develops. Crucially, it would sometimes withhold the answer.
A simulator does not replace real flying experience, but it concentrates certain kinds of experience. Pilots can encounter failures and combinations of events in a few hours of training that they might not encounter during years of normal operations.
AI could do something similar for engineering.
Years of responsibility on real programmes provide forms of experience that simulation cannot reproduce. AI could still expose junior engineers to carefully selected problems much earlier instead of waiting for a programme to provide the right sequence of experiences by chance.
AI would then be doing something more interesting than automating junior work. It would be helping us redesign how expertise is created.
From by-product to deliberate investment
If AI weakens the connection between production and learning, organisations will have to invest more deliberately in development.
There will be a temptation to take the productivity gain today and leave the expertise problem for later. But making the problem visible could also force us to become more serious about what engineering development is supposed to achieve.
Which experiences actually distinguish an expert from a beginner? Which mistakes are useful to encounter? Which situations develop physical intuition, regulatory judgement or systems thinking?
Those are questions we can start designing around.
Generative AI may force engineering organisations to make the development of expertise a deliberate investment. Used well, the same technology that removes some junior work could also help create experienced engineers faster.
References
AI helped to research and write this article.