Thinking Machines

In 1977 George Lucas released Star Wars and I fell in love with a tiny little droid named R2D2.

In fact, to this day, I have a full-size version that sits pride of place in my living room, a constant reminder of the magic I felt when I witnessed AI working alongside Luke and helping him save the galaxy.

Now, we’re not quite at the stage where robots are roaming around our homes, that’s if you exclude the annoying little Roombas that try to keep it clean, but AI is definitely more prevalent, be that via our smart speakers, mobile devices or streaming services.?

So how did we get here: Data! or more specifically the consumption of data and the prediction of patters through Supervised, Unsupervised and Reinforcement based learning.

Let’s investigate these different methods!?

We will start with supervised learning and as you can probably guess, it means someone looking over the shoulder of the CPU and judging whether the computer is getting things right.

This way of teaching means that each example you present is tagged with the answer so that the machine can learn through association and repetition.

It’s like teaching a child to read with phonic cards.

Now you’re probably thinking, wait we don’t hold up flash cards to a computer. But the thing is, we kind of do. Think about those apps that can identify objects, you know that’s a car, that’s a bike, that’s a person. Well, the computer only knows that because somebody has painstakingly gone through images telling the computer that that’s a car, that’s a bike, that’s a person.

But it’s not just classification, supervised learning also shows potential in regression-based problems or problems based on historical data, think about an estate agent for example who could use regression-based machine learning to calculate the value of your home.

By taking the independent variables i.e., the location, number of bedrooms and when the house was built, then using these alongside the dependent variable, (in this case the house price) they’re able to predict market value as circumstances change.

But what happens when we don’t have perfectly labelled datasets, or we ask questions that haven’t been trained, this type of learning is called unsupervised.

With this model the machine is given the data with no explicit instructions or an example of a desired outcome and simply asked to get on with it.

The neural network then tries to make sense of it all, desperately trying to find structure to the data and formulate a pattern.

Unsupervised learning can organize data in many ways, this could be through clustering i.e. matching colors, shapes or sizes, or anomaly detection whereby it detects unusual patterns.?

Think about how your bank checks for fraudulent activity or how sometimes your card won’t work abroad, this is down to a machine thinking that something is perhaps a little off.

Maybe it’s association, we’ve all bought things online and then been prompted with recommendations to consider as we check out. This is because a machine learning model has determined that these recommendations may be of interest by simply matching key attributes or vectors.

We also have a third option, reinforcement-based learning!

Now as the name suggests this training model is based on two aspects, exploration of what is not known and exploitation of what is. Think of it like playing a video game, how many times have you tried to complete a level only to run into an enemy or some other entity that cuts the game short.

Through perseverance and reacting to what has been learned you eventually complete the level. This is the very essence of reinforcement-based learning!

This cut and dry approach simply works through trial and error, the algorithm evaluates a situation, then acts, receives feedback, and adjusts accordingly to maximize long term reward.

So there you have it, three learning models that are transforming the way machines interact with data and in turn enabling technology to become more sophisticated, intuitive, and capable of augmenting human capabilities.

Just like my beloved R2D2 from Star Wars, thinking machines are becoming indispensable allies in our quest for progress and understanding the universe around us.