Never Automate Ignorance

Never Automate Ignorance

AI has enormous potential in engineering. But before we automate a process, we should first understand what we are asking the technology to improve.

AI is going to change engineering, and, in many areas, it already has.

The technology is developing quickly enough that the question of whether engineers should use artificial intelligence is becoming increasingly academic. We will use it. Our clients will use it. Our suppliers will use it. Our competitors will use it.

The more important question is therefore not whether AI belongs in engineering, it is how we use it technically and ethically.

Engineering has always progressed through better tools. Improved modelling, survey technology, automation, robotics, communications and data processing have steadily expanded what engineers can do.

AI should be considered in much the same way: as another potentially powerful engineering tool.

But tools are only useful when we understand the problem they are intended to solve.

And that leads to a principle that may be worth remembering as engineering organisations accelerate towards AI:

Never automate ignorance.

If we do not understand a process, its limitations, its failure modes or the information upon which it depends, automating it does not necessarily improve it.

It may simply allow us to make the same mistake quicker, more consistently and at greater scale. All the while often not even understanding if a mistake has been made at all.

Start with the problem, not the technology

There is an understandable temptation surrounding new technology to begin with the question:

Where can we use AI?

From an engineering perspective, that often puts the process in the wrong order.

The first question should remain the same one engineers have always asked:

What problem are we trying to solve?

Only then should we determine whether AI, automation, robotics, conventional software — or simply a better process — represents the most appropriate solution. That distinction matters.

How often have we witnessed a technological development becoming successful according to its own internal measures while producing relatively little operational value?

A system is implemented. A dashboard appears. Data is collected. But without a clear operational purpose, the information may simply sit there waiting for somebody to determine what it means.

Has anything become better by anything remotely measurable?

Has quality improved? Has risk reduced? Has an operation become safer?

Have we reduced waste, rework or unnecessary vessel time? Have we reduced cost and increased revenue?

If the answers are either unknown or simply no, what exactly have we achieved?

What should technology improve?

There are many ways of measuring technical progress, but for engineering applications I increasingly come back to four broad outcomes:

Value. Quality. Safety. Environment.

Not as competing priorities, and certainly not in order of importance, but as a practical way of testing whether technological change is achieving something worthwhile.

Value

Does the technology create genuine value?

That might mean reducing cost, increasing productivity, improving utilisation, creating a new capability or allowing work to be performed that would otherwise be uneconomic or impractical.

Importantly, value should not simply mean doing the same thing with fewer people.

Sometimes technology will reduce manpower but often it will allow the same people to make much more informed decisions.

Quality

Does the technology improve the quality and reliability of the outcome?

Does it help identify problems earlier?

Does it reduce rework?

Does it improve repeatability?

Does it prevent known mistakes from recurring?

This is where AI could have a particularly important role in addressing something engineering organisations have struggled with for decades: the Cost of Non-Quality.

Safety

Can technology reduce human exposure to hazardous activities?

Can it improve situational awareness?

Can it provide engineers and operators with better information before a decision is made?

Remote operations, robotics and autonomy already demonstrate what can be achieved when technology is applied intelligently to offshore and hazardous environments.

But safety also includes ensuring that technology does not gradually remove human oversight from situations where human judgement remains essential.

Environment

Can technology reduce environmental impact that may involve reducing vessel days, fuel consumption, unnecessary mobilisation, waste, repeat operations or physical intervention.

Sometimes the most environmentally effective operation is simply the one we do not need to perform because better information allowed the correct decision to be made earlier.

Engineering organisations already possess enormous intelligence

One of the most interesting opportunities for AI may have very little to do with autonomous machines.

Engineering businesses have accumulated extraordinary amounts of knowledge in the data we have been collating for years.

Project reports, lessons learned, non-conformance reports, near misses and Incident investigations are a small example of a collection of decades of experience, often gained at considerable cost.

Yet organisations are remarkably good at recording knowledge and surprisingly poor at retaining practical access to it.

Although the information exists, the problem is finding it and applying it when it matters.

Organisations forget

An experienced engineer encounters a problem; the team investigates it; a solution is subsequently developed and a report written.

There then follows a session on lessons which is diligently recorded before the project finishes and people move on.

Five years later, another project encounters a remarkably similar problem whereby a different group of engineers then begins investigating it again.

Eventually they may reach much the same conclusion as the first team implementing a very similar solution. The organisation has therefore paid twice to solve the same lesson, sometimes many times.

This is not necessarily because lessons-learned systems do not exist, most organisations have them.

The difficulty is turning a database of historic information from a static archive into something that can genuinely influence an immediate concern.

AI as organisational memory

Imagine an engineer developing a proposed methodology for the installation of subsea power cables in support of an offshore wind farm development.

Instead of relying solely upon personal experience and the immediate project team, that engineer could instantaneously interrogate decades of organisational experience.

AI does not need to engineer the solution. Its value may simply be in helping the engineer discover what the organisation already knows.

The value of the technology would not come from replacing engineering judgement. It would come from increasing the amount of relevant experience available to the person exercising that judgement.

The implications could be significant. Better access to organisational knowledge may help reduce repeated mistakes, unnecessary rework and some of the wider Cost of Non-Quality issues associated with lessons that were learned once but not effectively transferred.

That subject deserves consideration.

But there is an important condition attached to all of this: the information being interrogated must itself be reliable.

AI cannot repair bad foundations

There is an obvious danger in believing that a sufficiently sophisticated system can overcome poor data or poorly understood processes.

If our NCRs are incomplete, AI will analyse incomplete NCRs. If our lessons learned avoid difficult conclusions, AI will learn from sanitised lessons.

The technology will invariably inherit the inconsistency of the input. AI could make good information considerably more valuable. Equally, it may make poor information considerably more dangerous.

If the organisation has historically rewarded the recording of activity rather than the quality of learning, digitising the process will not magically correct it.

The engineering principles surrounding data quality, validation, traceability and verification therefore become more important rather than less.

Human judgement still matters

AI, like many new technologies, is exciting. But we should remain appropriately cautious.

Engineering is rarely performed with perfect information. Experienced engineers routinely make decisions while balancing incomplete data, competing risks, practical constraints and consequences that cannot always be reduced to numbers. That judgement has value.

Sometimes a project team knows that something does not feel right long before the evidence becomes obvious. That intuition is not magic; it is often the product of experience, intelligence and education.

Although AI may eventually become extremely good at recognising similar patterns, we should be careful about removing human judgement before we fully understand which parts of that judgement we are replacing.

The strongest engineering systems may therefore not be those in which AI replaces engineers.

They may be those in which the machine and the engineer compensate for each other's weaknesses.

The objective is better engineering

There is a risk that conversations around AI become divided between two extremes.

One presents the technology as an existential threat, the other presents it as the answer to almost every organisational problem.

Neither is particularly useful to an engineer trying to decide what to do late into a shift offshore.

The technology exists, it will improve and its use will grow. Our responsibility is therefore to apply it intelligently: identify genuine problems, understand them properly, and only then decide what should be automated.

A different question

Perhaps, then, we should stop asking:

“What can AI do for engineering?”

and begin asking:

“What engineering problems have we repeatedly failed to solve — and can AI now help us solve them?”

Safety, environmental impact, asset utilisation, inspection, planning, decision support and remote operations provide many of them.

There will undoubtedly be applications we cannot yet imagine.

But enthusiasm for what the technology might become should not prevent us from applying basic engineering discipline to what we build today.

AI could become one of the most powerful engineering tools developed during our working lives. Used well, it could allow future engineers to benefit from decades of experience accumulated before them. Used poorly, it could simply allow us to repeat old mistakes with considerably greater efficiency.

So perhaps three words are worth keeping in mind as we develop what comes next:

Never automate ignorance.