When to deploy AI

In the world of artificial intelligence, there is a tendency to consider the ultimate goal as creating something that humans can interact with in a natural way.

Today, smart speakers like Alexa and the Siri “intelligent” voice assistant on the iPhone, are far from natural. How often do people refer to the name at the start of a sentence to get someone’s attention? Yes it does happen, but talking to a smart speaker is a long way from having a natural conversation.

Nevertheless, it has its uses, from controlling smart home devices via voice, to finding out something useful and online shopping.

The thing is, human beings have adapted pretty well to interacting with the rather dumb smart speaker or intelligent voice assistant. And this is not a massive revelation. From the very beginning, early man adapted to use tools to enable people to achieve things they could not have done otherwise. In modern times, the automobile presents an unnatural user interface. No one would argue that it is not natural to use your right foot to accelerate or brake, yet we have adapted to the point that it makes sense to us.

What about software? In the days of green screen terminals, operators needed to learn and type in obscure codes to input relevant data into an electronic form. Today, the graphical user interface means that these codes have been mapped onto meaningful, human readable words and phrases that the user can pick from a drop down menu or use a button to click on “yes” or “no” or “cancel”.

The written word

For thousands of years people used the written word to distribute information. These days, those words are written on a computer keyboard. Some people may choose to use AI-powered speech-to-text, but people write in a very different way to how they would speak. The information they want to impart may actually become watered down and lost in the verbosity of natural language.

The question AI pioneers need to ask is whether the front-end user interface of a software application can be improved by using AI. It may make sense to deploy AI processing to provide insights at the back-end, but is the user interface a good use of powerful AI resources? In a word processor, for instance, the spell checker is simply a pattern matching algorithm. But grammar checking requires an understanding of context, something AI is notoriously bad at doing. AI pioneers need to focus on the things that truly add the most value to society and understand that humans are pretty good at adapting to work within the limitations of the tools they use.