REPAIR OR REPLACEMENT?

The role of machine learning in our industry

We can probably all agree that engineers won’t be replaced by robots in the near future, but the exact degree to which we should allow machines to run the show is still up for debate. Machines are clearly proving their worth in other industries, with applications ranging from medical diagnostics to autonomous vehicles. So in our own industry, should we allow critical decisions to be made by algorithms? We are certainly making headway in this direction, with machine learning techniques already supporting some of our decision making in pipeline integrity management. This is what I am talking about at IPC 2018 tomorrow.

The key point here is that machine learning is still supporting our decisions, not making them on our behalf. For now, I believe this is where our focus should lie – with the concept of ‘Augmented’ Intelligence, as opposed to ‘Artificial’ Intelligence. Whereas the original version of AI aims to replace human cognitive abilities with machines, this alternative AI values the human factor. In my opinion, the purpose of machines should be to enhance our abilities and ‘repair’ our flaws, but not to replace us.

I do think, however, that if humans and machines are going to work together successfully, we need to become much more aware of what each party brings to the table.

Machines clearly have a talent for data analytics; they can rapidly process data, learn patterns, and apply what they have learned to unseen cases. Most of the time this works well. But some of the things we try to predict in asset integrity management – corrosion growth, or cracking, for example – are phenomenally complex, and rogue behaviour occurs more often than we might like. Even more concerning are events that an algorithm cannot possibly predict, because they have never been seen before. These so-called ‘Black Swan’ events often have the greatest consequences of all. As the industry moves towards a zero incidents mindset, we cannot just assume that machines have all the answers.

In that respect, I would strongly argue in favour of the human factor. Humans have qualities that today’s machines cannot easily reproduce: individual and collective experience, ethics, intuition, and an empathic understanding of human behaviour, to name a few. I can see us running into problems if we discard these traits altogether from our decision making processes. There’s also the issue of trust; most of us feel much more comfortable knowing that a competent human has made a decision.

We don’t have all the answers either, of course, and we should be mindful of our limitations. Personality, ego and cognitive bias can make us act irrationally, and intuition (even that of experts) can just be plain wrong. Mere visualisation of high?dimensional data is off?limits to us, let alone discerning meaning. And even the best of us have appalling memories in comparison to a computer. We need the objectivity and processing power of machines to support us as engineers.

So, to wrap it up, we should continue the debate about the role of machine learning in our industry, but keep our minds open. Debating ideas is a very human way of finding optimal solutions.