The terms ‘AI’ (or artificial intelligence) and ‘ML’ (machine learning) are often conflated, but this is incorrect – while machine learning is an application of AI, they aren’t the same thing.
AI is typically focused on mimicking human decision-making to accomplish tasks in an intelligent manner. These tasks could include holding a conversation through a “chatbot” or piloting a driverless car.
ML, on the other hand, stems from the idea that computers can mimic human learning – that it isn’t necessary to teach computers about everything, since it’s possible to teach them how to learn from themselves. This way, computers can act without being specifically programmed.
Typically, this requires the input of data. By assigning data to specific categories, we can teach machines to draw connections and predictions from the data provided.
Of course, there are lots of challenges. A good example would be teaching machines to recognise handwriting. Take, for example, the letter “a”. When people learn to recognise letters, we identify the “a” shape as being linked to things like phonetic output. This is because when we see a similar shape, our brain unconsciously compares it to an example we’ve already seen. If it’s similar enough, we recognise it as an “a”, regardless of the minute variations that may occur as a result of different handwriting.
Machine learning allows computers to function in a similar manner to the human brain. If we give a computer a large number of handwritten letters, we can build a system that takes the examples given (known as the training data), to automatically infer a series of rules for recognising letters. As the system makes decisions, it adds examples to its training data, informing future decisions and improving its accuracy. The computer recognises that the letter “a” requires curved lines in certain places, and thereby distinguishes it from other vowels.
A more sophisticated example would be creating a system to distinguish between tweets to determine who sent them. If we were to examine tweets made by Presidents Trump and Obama, it would be clear that there are certain key terms that occur at a higher rate of incidence in the tweets of one than in those of the other. For example, President Trump may be more likely to tweet about “fake news”, while President Obama may tweet more about “affordable care”. By giving the system in our last example a sufficient number of their tweets, we could teach it to distinguish for itself between tweets from both Presidents – not only that, but we could extend the system to make predictions concerning their future tweets.
This is why Resolver believes the application of ML is crucial to the future of complaints-handling. With sufficient data, businesses can build systems that afford them greater insight into their own processes and their customers’ behaviour.
Having worked with consumers and businesses to resolve over two million cases, Resolver has generated a vast amount of clean data around complaints.