Why artificial ‘intelligence’ is better than artificial ‘knowledge’

Today’s artificial intelligence (AI) services are remarkable things. They can write. They can draw. They can even speak. And thanks to their training on large volumes of data, they can produce answers to a wide variety of questions.

But intelligence involves more than getting knowledge. The more important component is knowing how to learn. And even better is knowing how that knowledge was acquired in the first place.

Understanding beats knowledge

Just learning everything almost feels like cheating. It gives the appearance of intelligence but without having that underlying understanding.

It’s like being able to swallow an encyclopaedia and then regurgitate it on demand. Think of it as the ultimate pub quiz team but without the ability to make sense of the answers this knowledge provides.

And yet it’s the ability to understand how knowledge is obtained that really helps with future decisions – and so is actually useful, as opposed to just merely interesting.

For example, autopilots know what should be done now. They don’t just replicate what was done before, even if they do use that historical knowledge to inform what they do.

How to get useful intelligence

Paradoxically, acquiring useful intelligence doesn’t need an advanced ability to perform precise extractions from a vast body of general knowledge. Instead, it needs a simpler skill: to work out answers from a much smaller amount of very specific knowledge.

In effect, it’s the difference between asking, “What do people have for breakfast?” and “What does Chris have for breakfast?” In other words, while it’s interesting to know what the habits of people in general are, it’s more useful to know Chris’s habits.

But getting such useful information isn’t that hard. (Although AI tools don’t know what I have for breakfast, it’s not really a secret or even that difficult to work out. Even without access to my movements, a few supermarket receipts would show what I eat.)

Artificial intelligence – not artificial knowledge

Establishing useful information isn’t about collecting as much data as possible, but about collecting the right data.

In turn, this means real artificial intelligence – as opposed to artificial knowledge – doesn’t have to be based on large, general data models, but on small ‘hyperlocal’ ones.

And this isn’t just another form of ‘cheating’, i.e. being able to assume the answers if the data is skewed in a particular direction. Instead, it’s a more accurate reflection of the way natural knowledge is obtained. So, my children first learnt what I have for breakfast by watching me. And their knowledge of how other parents eat came later when they visited friends’ homes.

Capturing the right data

For most organisations, this distinction means AI doesn’t have to be as abstract, disconnected or scary a concept as it might initially appear. Plus, the real data needed to make the interpretive techniques and natural language models work usefully is easily available to them.

To be really useful, this data also needs structure, because randomly capturing a small amount of everything isn’t enough.

Capturing behavioural (or operational) data – organised around interactions and their context – is worth far more than a mass of static reference data. That’s because the behavioural data incorporates the concepts of timeliness, precision, accuracy and comparability.

But even here, this requirement doesn’t have to be daunting.

The ways to capture that behavioural data aren’t specific to any particular domain. The principles of classifying, qualifying and quantifying data to produce good operational information apply regardless of the applications from which the data originates. Only the specific instances of classifiers, qualifiers and measurements change; the concepts remain the same.

Identifying and collecting behavioural data

Quantiv has long been an advocate of the benefits of good data. Our NumberWorks method helps identify the useful behavioural data that makes for good operational information. And our NumberCloud service makes collecting that data a natural part of any IT solution.

To learn more, call us on 0161 927 4000 or email: info@quantiv.com