Artificial Intelligence – rare common sense

Artificial Intelligence – rare common sense

Last week I enjoyed the company of a room of IT leaders in Birmingham, and spoke on the subject of Practical AI. Artificial Intelligence is suddenly the topic du jour. Whether it’s sinister activities by the likes of Cambridge Analytica, or maybe-not-quite-as-sinister demonstrations of systems like Google Duplex, it’s a hot topic.

My focus, given that I was looking at the practical applications of AI for businesses this year, was Machine Learning. This is an AI technique that gives computers the power to learn and to essentially ‘reprogram’ themselves based on their experience. I have seen this in action over the past few years and it can be very impressive. Indeed, the key tech behind Alexa, Cortana, Siri etc is Natural Language Processing, which builds on Machine Learning (ML) techniques. This is the reason you can ask one of these devices lots of different questions about the weather, how warm it is outside, if it’s going to rain tomorrow, and get the same response. They have not been programmed to respond to these individual statements. But they have been ‘trained’ to distinguish questions and find relevant answers.

One type of ML is called ‘unsupervised learning‘. In this technique the computer will look for similarities in data in order to make some sense of it. A good example is Google News. An unsupervised learning algorithm tags news stories by discerning the main themes, and then when you search for ‘Cricket’, you see lots of articles about cricket. Not because they have been tagged by an editor, and not because Google went off and searched all articles for the word ‘cricket’ when you typed it in. You see them because they have already been processed and tagged by a ML algorithm. This is also known as ‘clustering’. So, if you throw your customer data at a clustering algorithm it will tell you things about that data. It might tell you that your younger customers are less brand-loyal. It might tell you that you sell more ice-creams in the summer. It might tell you that maintenance in high-traffic areas of your building costs more than others.

At this point you might wonder why you would bother. You know all this. Of course ice-creams sell more in hot weather. It’s common sense. Why are we wasting our time on this nonsense?

My experience is that common sense is a dangerous tool. Because ‘everyone’ knows it, it’s assumed. Are you sure that every one of your ice-creams sells more in hot weather? Are you missing a trick because there is an outlier? If you rely on ‘common sense’ you will carry on the same path as you always did. This is not a survival trait.

Common sense is also something that drifts, IMHO. When I use that usenet acronym, what do you see? It means In My Humble Opinion. Unless it doesn’t. Buzzfeed recently ran a survey in which they found that more than half of their readers thought the H stood for Honest. This really floored me. In My Honest Opinion is a very different statement to In My Humble Opinion. I have used one phrase, but it’s been taken to mean another, despite that being utterly, completely wrong!

Now your computer is able to learn ‘common sense’ things about your business.  But it doesn’t have any politics, or emotional ties to a particular product or location.  It can tell you about trends you haven’t spotted or exceptions to your golden rules.

Are there other, maybe more important, assumed pieces of knowledge in your organisation or industry? Are people really acting on information in the way you would expect – is common sense really that common? Perhaps we need the machines to tell us what really is common sense every now and again.

1 Response

  1. AI only has bias that’s built into it, it’s engineered and can be developed and improved through ML..our bias and assumptions are experiential and emotional, not engineered and can obstruct thinking and decisions…collective bias or ‘common sense’ can be deceptive…

Have Your Say: