The reality behind the hype: Machine Learning & Artificial Intelligence
Whenever a new technology surfaces and appears to change the world, there’s always an enthusiastic bandwagon of companies and marketers trying to cash in on that hype. Think Radium chocolate promoted for “its rejuvenation power,” more recently pointless IoT devices (Like my recently disposed of wifi coffee machine) or BigData being the answer to any question.
Fortunately, this centuries technology buzzwords are far less likely to make you glow in the dark, however, blindly following the new crop of “Machine Learning” (ML) and “Artificial Intelligence” (AI) spiel can still lead to unwanted side-effects. Understanding what these terms are and where they’re useful (and potentially dangerous) is vital to separating genuine technical advancement from someone riding the coat-tails of new big-crypto-data-chain-o’-things.
Machine Learning is a Technique, not a solution – When tasked with the problem of housing people in a big city. The solution isn’t “concrete”. It may be an important component, but the solution is far more than a single building material alone.
ML on its own doesn’t magically reduce cost, improve predictions or provide anything at all… unless you know where to use it.
Not every problem benefits from Machine Learning – After the roaring success of the concrete skyscraper, I’ve decided to branch out into aircraft. I’ve all the skill and experience with concrete to make the best concrete aircraft the world has ever seen. We’ve heard the term: “If your only tool is a hammer, every problem looks like a nail”. The same applies here. People’s desperation to appear cutting edge drives them to wax lyrical about their ‘industry-leading ML’. Which (if they’re using ML at all) is potentially less efficient and at worse potentially dangerous than the algorithmic alternative.
There are risks – Introducing any form of AI into a system brings with it several extra concerns depending on the what problem you’re trying to solve. ML models require “training” (the “L” part of ML). The results of this being a system which can now ideally make some sort of decisions for you. However, this system isn’t often clear to understand. When learning mathematics you’re taught to show you are working so the teacher can ensure you’re using the correct technique. This doesn’t come for free with ML. Entrusting a computer with making complex or potentially dangerous decisions needs a lot of thought.
Tesla’s autopilot is certainly one of the most impressive examples of ML that I can think of and it still throws its hands in the air the second it’s not 100% comfortable, giving control back to the driver. Tesla’s software architects have had to consciously introduce “uncertainty” into autopilot to guard against disasters.
So, should we be shunning anyone who professes to achieve something using one of these techniques? Certainly not. It’s an incredibly exciting time for technology, we have developed tools and practices that are enabling some truly amazing things.
Any Machine Learning powered technology should have a compelling “Why” use ML for their corresponding “How” they use ML. If the “Why” is marketing buzzwords and twitter follows, they’ve probably not put enough thought into their “How”.