Arthur Samuel was among the first computer scientists to develop research into Artificial Intelligence (AI) and Machine Learning. Having graduated from MIT, he went on to work for IBM. While there he built the Samuel Checkers-playing Program which was the first successful self-learning AI so advanced in fact, it could beat him in a game of checkers and became the starting block for all AI research. In 1959 he famously defined Machine Learning (ML) as the:
“Field of study that gives computers the ability to learn without being explicitly programmed".
However, this definition may be slightly too vague. An alternate definition of ML was proposed by computer scientist Tom Mitchell in his book “Machine Learning” (1997):
“A computer program is said to learn from experience E with respect to some task T and some performance measure P, if its performance on T, as measured by P, improves with experience E.”
This definition is more precise but slightly confusing so let's break it down a bit. If we take Samuel’s Checkers-playing program as an example, the programme will learn to improve based on its performance (P) as measured by its ability to win at the tasks (T) focused around checkers. It will do this via experiences (E) obtained by playing games against itself. To have a well-defined learning problem, we must identify these three features: the type of the tasks, the measure of performance to be improved and the source of experience. Once these have been determined the Machine Learning can begin!
Now that we have roughly outlined what ML is, it’s time to examine how companies use it to their advantage.
Companies nowadays are personalising their customer experiences more and more as they seek to attend to three crucial human drivers: the desire to bond and affiliate, the desire to learn and grow, and the desire to be in control and prepare for the unexpected. By tailoring the customer experience towards fulfilling these drivers, a positive emotional response is elicited and associated with the experience as a whole. This response then feeds into the subconscious decision-making process of the individual, encouraging them to have a positive impression of the entire organisation.
Research has also found that the entire experience does not have to be completely positive from start to finish for this effect to be achieved. Instead, customers are more likely to remember just the high-point and end-point of each interaction; referred to as the `Peak-end Rule’. By personalising the experience to the customer we create these small but important “peaks” that fulfill the desires and sufficiently impact the mind-set of the customer.
‘Why have we gone from discussing Machine Learning to customer psychology?’ you may be asking. The answer is that companies use ML to create these highly personalised customer interactions! Storing and tracking all customer data to then manually implement into marketing strategies would be a herculean task if done without the use of such technology.
Here are a few examples of exactly how companies use machine learning to personalise their customer engagement.
While AI is used on Facebook daily to filter out spam, poor-quality content and to monitor the spread of false information, it is also used dynamically in other areas. Have you ever spoken with a company on Facebook messenger for instance. Most of the time the thing responding is a chat-bot rather than a real person. A chat-bot is a piece of software that can conduct a conversation, it can either contain pre-written responses or is developed using ML and AI to ‘think for itself’. While some chat-bots are highly advanced, they have not been able to pass the Turing test of intelligent behaviour as of yet.
It is possible for any company to create and submit a chat-bot to Facebook, this is beneficial as it means that even small start-ups can have a larger customer service presence even with limited engineering resources. A good representation of this service is demonstrated by the Whiskey experts at 'Macallan'. They have implemented a chat-bot on Facebook that provides whiskey recommendations using a multiple choice format. Customers can also ask the bot for advice or information about the whiskeys. As of 2018, there were roughly 300,000 chat-bots active on Facebook, that was three times as many as the year before and this number is likely to have risen dramatically since then.
IBM has developed their own AI, Watson, built on a machine learning system. Developed in IBM’s DeepQA project led by principal investigator David Ferrucci, Watson is a question answering computer system with the power to process and respond in Natural Language. Though originally built to answer questions from the quiz show ‘Jeopardy’, it was quickly advanced enough to beat two previous champions of the game. At this point the company looked to commercialise the system and in February 2013, IBM announced Watson would be used for utilisation management decisions in lung cancer treatment at Memorial Sloan Kettering Cancer Center, New York City. The system was a huge success and IBM’s business chief in 2013 stated that 90% of nurses in the field using the system follow its’ guidance.
Watson has demonstrated an aptitude for making highly accurate recommendations for the treatment of certain types of cancer. Studies have found that Watson was not only able to reliably determine ineligible patients for clinical trials, but can also identify actionable mutations from next generation tumour sequencing data that was unidentifiable by the manual analysis of an expert molecular tumour board. The implementation of an AI system within the healthcare system can no longer be overlooked. Of course a system such as this would also be hugely beneficial in the retail sector as well as the hospitality industry, but there is a key factor that encourages the use of Watson first and foremost in the healthcare sector. Data. Specifically the incessant influx of health data that is being produced daily. It has been stated that medical data is expected to double every 73 days from 2020 with each person generating enough medical data in a lifetime to fill 300 million books. If this is truly the case, a system for managing even a portion of this is more crucial now than ever.
Watson is now packaged up and being offered on a license basis, it is one of the first AI applications being sold in this manner.
----------- Part 2 next month! -----------
Bernard Marr 2019, Bernard Marr & Co., accessed 6th February 2020, <https://www.bernardmarr.com/default.asp?contentID=730>
Customer, accessed 5th February 2020, <https://engagecustomer.com/the-importance-of-personalisation-in-a-customer-experience-strategy/>
IBM Community, IBM, accessed 7th February 2020, <https://www.ibm.com/developerworks/community/
IBM Watson Health, IBM, accessed 6th February 2020, <https://www.ibm.com/watson-health/about/get-the-facts>
Kemi Akenzua 2019, Digital Initiative, accessed 6th February 2020, <https://digital.hbs.edu/platform-digit/submission/hello-barbie-nlp-and-the-first-conversational-doll/>
Lexie Kane 2018, Nielsen Norman Group, accessed 5th February 2020, <https://www.nngroup.com/articles/peak-end-rule/>
Libby Plummer 2017, Wired, accessed 7th February 2020, <https://www.wired.co.uk/article/how-do-netflixs-algorithms-work-machine-learning-helps-to-predict-what-viewers-will-like>
Netflix, accessed 7th February 2020, <https://help.netflix.com/en/node/100639>
Paige Cooper 2019, Hootsuite, accessed 7th February 2020, <https://blog.hootsuite.com/facebook-messenger-bots-guide/>
Sib Mahapatra, Toptal, accessed 6th February 2020, <https://www.toptal.com/insights/innovation/how-top-companies-use-machine-learning>