AI
20101161 - businessman hand draws gear to success concept

Last month, I discussed some of the basic categories of Machine Learning (ML) and how they might be applied in our asset-centric world. At the end of that blog, I promised one more in the series – this time attempting to dispel some of the more persistent myths around ML. Is it a locked black box with algorithms I can’t understand and therefore shouldn’t trust?  Is it the same as robotics, or artificial intelligence?  Is ML (or deep learning or AI) going to take my job?  Or is it going to take over the world and kill us all?

Progress gets us every time

Let’s get past the more existential of those questions first.  Yes, ML or Deep Learning or AI may well eventually take your job.  Get over it.  This is how progress works.  Smart Meters are stealing honest hard-working Meter Readers’ jobs all over the world as we speak.  Domestic refrigeration made all the ice harvesters redundant by the first half of the 20th Century.  Hey, the evil that is agriculture pretty much destroyed the future employment prospects for all the honest hard-working hunter-gatherers too.  Ever has it been thus.

We’re (possibly) all doomed

And is AI going to take over the world and kill us all?  Well, that’s a harder question.  Maybe it is.  Roko’s basilisk postulates a future sentient AI that will be able to reach back into its past and punish all those that did not actively work towards its creation.  That may just be a thought experiment, but what about the Governments around the world working on military AI applications right now that are specifically designed to kill anyone they have a disagreement with?  That stuff’s real, right?1

Now, given that this is a utilities-focused website, here is perhaps not the place to discuss the difference between free will (autonomy) and prosaic ability - especially with respect to machines killing people.  But that one misunderstanding, combined with the unhelpful amalgamation of AI and robotics and ML and any new technology we’re uncertain about does form the basis of almost all misinterpretation of the wide, nebulous thing “AI” has apparently become.  Especially the misinterpretations keeping you awake at night.  If you’re interested then do investigate further or get in touch and we can talk about it more.  But for now, let’s get back to practicalities and consider the last and hopefully the most relevant question I’ll discuss this month: is Machine Learning a black box solution I can’t understand and therefore shouldn’t trust?

The black box model

In recent years there have been a few highly successful analytics businesses selling results - and often very impressive results at that - based on a business model that goes something like this: give our smart people access to your data and we’ll do some clever Machine Learning and AI and other cool things with it and show you how you can [insert important improvement to your business here].  For their client businesses, as long as the ROI is right, that model is absolutely fine.  And good for them.  Thing is, engineers often hate it.  Especially engineers being asked to make decisions that potentially affect the safety of their operations, based on analytics they can’t see and touch and understand.  If something goes wrong based on the results from a black box, who is to blame, exactly?

Does it really have to be this way?  The answer is a resounding no.  ML is not magic or secret or unfathomable.  It’s typically not even computers creating brand new neural networks resembling biological nervous systems to solve problems in highly complex ways.2  Mostly, ML is about a computer applying some very well understood and understandable analytical techniques and learning which techniques applied in which ways lead to desirable results.

Opening the black box

Your analytics provider should be able to describe and demonstrate to you both the analytical and the learning techniques they’re using in an ML implementation.  In some cases, you’ll discover that the solution being described for the particular issue or opportunity being addressed is actually quite a simple and eminently understandable one.  Others might be more complex, of course.  But often, they’ll just be more complex implementations of the same principles you saw and understood before.

If you are embarking on an advanced analytics journey, decide early what level of understanding of the techniques being deployed you need.  If you can use cloud subscription-based ML tools to predict fraud demonstrably better than your current models, do you really need to know every detail of how they do that?3

If you are happy to engage a business that employs the black box model, that’s entirely fine too.  But if you don’t trust a black box, let’s be clear what it is you really have no faith in.  A black box is all about obscurity.  I don’t trust obscurity either.  Machine Learning (and perhaps a few smoke bombs and well-placed mirrors) may be the tools inside some vendors’ black boxes, but Machine Learning itself is not the black box.  Take the time to understand ML and I’m confident you will come to see its value and give it the trust it deserves, as a set of tools and techniques that can help you run your business more effectively.

So there we have it.  In the past three blogs, I’ve introduced Machine Learning; discussed a few of the main types of ML in the context of asset businesses; and made an attempt to dispel (or confirm) a few of the biggest myths around the most hyped areas of advanced analytics.  I hope you’ve found them useful.  If you have - or even if you haven’t and disagree entirely, I’d love to hear from you.  But make it soon.  Before the robots get us.

 

1 Yes.  Yes it is.

2 OK, so actually, one branch of ML is exactly that.  But even that stuff isn’t unfathomable.

3 Maybe you do.  Your call.

 

About the author:

David Socha is Teradata’s Practice Partner for the Industrial Internet of Things (IoT).  He began his career as a hands-on electrical distribution engineer, keeping the lights on in Central Scotland, before becoming a part of ScottishPower’s electricity retail deregulation programme in the late 1990s. After a period in IT Management and Consulting roles, David joined Teradata to found their International Utilities practice, later also taking on responsibilities in Smart Cities and the wider Industrial IoT sector.

 

Image Credit: 123rf.