With so much talk about Artificial Intelligence (AI) at the moment – from the UK’s announcement of nearly £1 billion for AI sector development to the advent of innovations like that Tess the mental health chatbot – it can feel like everyone who’s into AI is a genius, and those on the outside have no hope of catching up. We get the feeling, but don’t think that needs to be the case.

We want to use this blog to break down some of the terms you hear about so that it’s easier to figure out how you could adopt AI in your organisation. We’d also like to show you that AI and its different techniques are reasonably straightforward to understand, and that everyone has the ability to contribute to AI development. We’ve also found some good case studies to help bring the terminology to life. We believe AI is exciting and offers organisations new and creative ways to transform how they work and the services they offer.

 

Let’s start with the obvious one… Artificial intelligence

One of the best definitions of ‘Artificial intelligence’ is by Merriam-Webster,
which defines it as:

“The capability of a machine to imitate intelligent human behaviour.”

This covers a range of behaviours that machines are increasingly capable of, including problem-solving, learning and decision making. AI is an umbrella term – so just because something is labelled as AI, that doesn’t really tell you what it does that is ‘intelligent’. When in doubt, always ask!

A lot of news coverage looks at how AI (usually automation) will result in huge job losses. We believe the goal of AI is not to replicate or replace humans. Rather, we would suggest thinking about AI in terms of its unique computational capacities, and the ways it can augment human skills. For example, DeepMind used their AI software to predict the future temperatures of Google’s servers (keeping computer servers cool is a nightmare for many companies, and an even worse nightmare for the environment) and used these findings to reduce the need to cool servers by 40% and therefore almost halved their carbon footprint. In this situation, the AI software’s computational power found new patterns and evidence that its human overlords used to make a positive and impactful change.

As we’ve already mentioned, ‘Artificial intelligence’ is used as a catch-all term for a number of things in this space, some of which include:

 

Machine learning

This is an AI technique where computer “systems have the ability to automatically learn and improve from experience without being explicitly programmed” – basically, a system that has learned what it needs to do rather than being programmed step by step (source). It has many potential uses – for example, detecting fraudulent behaviour by learning and recognising the characteristics of fraudulent activity.

A great example is of games company Unity’s simulated environment where they used Machine Learning to teach zombie teddies how to avoid each other. The teddies weren’t programmed with the ability to detect and avoid each other, they were instead given the overall goal to avoid each other and put inside an environment where they could learn how to do it. It’s a really efficient way of teaching software what you want it to do, while retaining the capacity to learn even more in the future. Here’s another Unity example of teaching a chicken how to cross the road (chickens from the 14 minute mark). The downside is that computer scientists are not always clear on exactly how the software learns, which makes transparent decision-making hard.

 

Robotic process automation

This is a way of automating processes by using ‘software agents’ or ‘software robots’. It’s not strictly AI, rather a very sophisticated form of automation. It’s still an important relation of the AI family as it can be an enabler for other AI techniques. It teaches the software agents each step in an electronic process or workflow – such as receiving completed forms, checking the form for completeness and updating a system record accordingly (source). So basically – tiny robots zooming around your computer systems doing your processes for you.

Introducing this is very effective when dealing with a large-scale process that requires the same step to be completed repeatedly – for example data entry. Typically, this would then free staff up to focus on higher value tasks such as dealing more attentively with serious cases.

 

Natural language processing

Natural language processing is the ability of a computer to recognise, interpret, analyse and respond to human language. This could include the ability to read the inputs on an online form or understanding what’s been said in a web-chat window (source).

A negative example of this is Microsoft’s Twitterbot ‘Tay’, which used natural language processing to understand and respond to what other Twitter users tweeted to it. Adding in some Machine Learning enabled Tay to learn and change its personality and responses based on these tweeted conversations. The result? Tay had to be turned off in less than 24 hours, as being spammed by racist, misogynistic and Holocaust-denying tweets turned it into a very nasty bot indeed.

 

Intelligent data extraction

Simply put, this is where a computer is able to draw out a specific bit of data from a document, and upload it to a database. For example, you might want to draw out a National Insurance number from a P45. Yeah it’s pretty straightforward, and super useful for routine processes – works well with RPA.

 

Neural network

A neural network is a technique for identifying ‘underlying relationships in a set of data by using a process that mimics the way the human brain operates’. What this means is that a neural network is able to deal with inputs that are slightly different from each other, and find the appropriate or best possible result (source).

So, that earlier example of DeepMind using AI to reduce the need to cool Google’s servers was achieved using a combination of AI techniques – including a neural network. It basically set up a series of ‘neural’ nodes that became weaker or stronger depending on that amount of information they processes – and in the case of DeepMind’s work, the stronger nodes were the ones that flagged the future server temperatures.

Each of these techniques have their value and limitations, depending on the type of problem you’re trying to solve, or service you’re aiming to offer. This is by no means an exhaustive list, but we feel it covers some of the most important elements to have a grasp of.

 

We can help you start your AI journey, by running seed workshops and discoveries to get things started. If you would like to learn more about this, and the ethics and opportunities around harnessing AI, don’t hesitate to get in touch with josephine.young@methods.co.uk and/or mark.nutley@methods.co.uk.