When I go into my local coffee shop, they greet me and ask me if I want my usual coffee (latte if you are offering). Usually I say yes, but occasionally I go for something different like a mint tea. This usually gets the response of a raised eyebrow or a comment about trying to cut down on caffeine. They have hundreds of customers, but they know who I am, know my habits and therefore recognise when I’m departing from my usual pattern of behaviour.

That is great customer service using a simple algorithm. You may not recognise that as an algorithm, but what they are subconciously doing is a form of pattern recognition. X chooses something other than a latte = anomaly. Of course great customer service is much more than recognising whether I order something different. It is also about recognising when I’m in a hurry or on my phone and don’t have time for a chat.

There has been a lot of hype over the years about Artificial Intelligence (AI). Some of it has been justified with some billion dollar companies such as Google and Autonomy (since bought by HP) created from research in the field. Over the last couple of years, the technology to build applications that are powered by artificial intelligence has become highly accessible and much lower cost.

This has led to many organisations starting to explore the possibilities around AI and how it might help them to serve their customers better, become more efficient or develop new products more effectively. There are many different potential applications of AI. This means that applications using AI across organisations tend to be distributed across an organisation and not well co-ordinated.

Increasingly, this machine intelligence is becoming a key part of how organisations operate. This means that the machines are becoming as important to the smooth running of the business as its people. Most AI implementations use a number of different types of AI together. For example, one algorithm recognises the language something is written in, then a second recognises the meaning of the words. These algorithms form an intelligent production line process in much the same way that doctors refer patients to different specialists to aid with diagnosis or factory workers specialise in a specific step in the process. Each of the medical specialists is trained to have in depth knowledge of their chosen field and they are therefore very good at recognising the symptoms for that condition. They are much less well tuned to recognising symptoms of conditions in other fields of expertise.

This concept of specialisation is very important for the successful development of AI in the enterprise. Sustainable advantage can be created for an organisation by training specialist bots using data that is unique to your organisation and then getting them to work effectively together. If you were thinking about this in terms of Human Resources, you would be asking yourself what specialists you need in your organisation and where do they fit within your organisational structure and business processes. You would then ask what training they will need and how you will measure their performance. The same questions are equally valid for your bots.

This approach to bots is increasingly relevant now because of the way Modern artificial intelligence bots learn. This was demonstrated recently in negative headline grabbing form by “Tay” a chatbot created by Microsoft and let loose on Twitter in March this year. Tay learned from the tweets of others and started tweeting based on this. Within 24 hours, Tay was tweeting racist, sexist tweets. Much like a child, Tay was exposed to racist and sexist tweets and was not able to distinguish between acceptable and unacceptable tweets. Think of Tay as the rogue employee.

What does this tell us about how AI learns? Well, it is all about the data…. If you train a bot using garbage data (racist tweets), you will get garbage out. If you manage the training well, within a safe, controlled environment, you are much more likely to be successful. So in much the same way, if you train your staff well with clear guidelines on what is acceptable and what isn’t, you will have happy staff and customers.

Much like your human resources, generally the more you train bots, the better they get. However, until you have a clear framework in place to measure and manage the performance of your bots, you won’t know which ones are performing well and which are not.

Of course, you need good people to train good bots…

Article by Giles Blackburn