Our Thoughts

AI and the Customer

19th July 2018

The Institute of Customer Service has just published an excellent report on AI and the future of customer experience. You can download the free exec summary here. We were lucky enough to work with the Institute on the research, and we managed to speak to a lot of very knowledgeable experts and practitioners.

Here are a few of my personal reflections:

AI is poorly defined and therefore overhyped
There is no clear definition of what AI is. One joke is that AI is whatever we haven't figured out how to do yet. Media coverage tends to drift into science-fiction, but the reality is that the vast majority of current AI consists of relatively simple machine learning systems....but these can be very useful.

"I'd say ML is both overhyped and underrated. People overestimate the intelligence & generalization power of ML systems (ML as a magic wand), but underestimate how much can be achieved with relatively crude systems, when applied systematically (ML as the steam power of our era)."
François Chollet, Google

How should we define AI? Is it synonymous with machine learning? Should we include rules-based systems? I think it makes sense to define it in terms of the kinds of problems AI can solve. As one person I interviewed put it...

"AI is necessary to deal with unstructured data, if it was structured AI wouldn’t be necessary."

Current AI is very simple
Artificial General Intelligence, i.e. the human-equivalent intelligence that media reports tend to focus on, is still decades away (c. 2040). The tools we have at the moment are comparatively simple. Beneath the marketing hyperbole and the profusion of algorithm names, machine learning can perform a handful of tasks:

  • Learn to attach a label to something based on input features
  • Learn to predict a numerical outcome based on input features
  • Group things together based on how similar they are

The art is in applying those simple tools effectively (i.e. finding the right data, identifying the right features), and combining them effectively with other tools to solve problems in the real world.

Start with the problem, not the solution
When it comes to application, people tend to want to adopt the latest trends. Do you really need deep learning, or are you using a trendy sledgehammer to crack a nut that could be opened just as well with the judicious use of Robotic Process Automation? Start by being clear on the problem you are trying to solve, then figure out how best to tackle it. Always use the simplest approach that will get the job done.

The future belongs to centaurs
Many people, naturally, are concerned about the social impact of AI. Will a robot take your job? In some ways the answer is bound to be yes - intelligently-applied AI will take over more and more tasks that humans currently do. But tasks are not the same thing as jobs.

One of the most optimistic views of AI is to be found in the idea of intelligence augmentation (IA), as discussed in this brilliant article. In chess competitions a "centaur" consisting of a team of human + AI has been shown to outperform either human or AI competitors working alone.

To thrive in the future we need to learn to work with AI to improve our own performance, rather than trying to compete against it.

01484 517575
Taylor Hill Mill, Huddersfield HD4 6JA
Twitter LinkedIn
...