Artificial intelligence

AI: How do we create truly intelligent systems?

28 March 2017 by Christian Møller
As we move to ever more intelligent and complex systems, computers need to figure out by themselves the value of the data they have collected. This in turn enables them to make good decisions.

Nothing is worse than steep bushwacking in the dark, in deep snow with a heavy pack on your back.

The worst possible scenario for my patrol in the military was moving from A to B in the dark, in an unknown area with fog and no GPS. We had nothing but a compass and a map, and to avoid exposing our position, we could only use a flashlight if we covered ourselves with tarp. Navigation was extremely challenging.

Follow Kongsberg Digital and Kognifai on Linkedin

As we slowly added to what little information we had, it became increasingly easier to navigate as we could identify features in the environment such as mountains, rivers, and pylons. Still, it didn’t take me long to realize that what appeared to be the fastest route on the map might not actually be so. Steep hills, dense vegetation, enemy lookouts, and bad weather would slow our progress. But these parameters cannot be found on a map.

Reinforcement is fundamental to the learning process – and to AI

You start making efficient decisions based on the available information through learning, and each decision you make, good or bad, reinforces this learning. If you have ever experienced being soaking wet with a soaked pack in sub-zero temperatures, you’ll know what I mean and avoid it at all costs.

Read also: CEO, Hege Skryseth's blog on Kognifai - one open ecosystem

I think a system that takes action based on its environment falls into the category of artificial intelligence. To make decisions, it needs information. As the amount and quality of the information it receives increase, the system will make better decisions and thus become more advanced. By applying reinforcement, the system will keep learning and constantly be rewarded for good and bad decisions, just like kids (and, in fact, most people) are rewarded for their good or bad behavior.

To build an intelligent system, you need to feed it with data.

Digitalization is, to my mind, the effort of quantizing in a digital format real world phenomena such as events, processes, and objects. Only when such a digital transformation has taken place, can we start creating genuinely intelligent systems that can act as humans would.

Current intelligent systems are highly influenced by the humans that build them, and we typically only provide them with the information we need to solve a specific problem. If for instance, you are teaching a computer to learn a language, weather data is arguably not necessary, and therefore not included.

A truly intelligent system must figure out the importance of the data it collects by itself

We are moving to ever more intelligent and complex systems where computers need to figure out by themselves the value of the data they have available, or more accurately the importance of the data, towards the decisions they are making. Today, huge steps are being made towards fully autonomous systems, some of which are fairly complex. As the world becomes ever more digitalized, we are building a fundamental prerequisite towards a world where autonomous systems and artificial intelligence become ubiquitous.

Artificial intelligence
Machine learning
About the writer
Christian Møller
Møller has worked for KONGSBERG since September 2016 and holds the position of Chief Technology Officer for Kongsberg Digital. He previously worked in research for Microsoft helping to develop a wide range of products, as well as evaluating new technology and technology companies. In Møller’s current role, he plays a central part in further developing the company’s efforts towards existing customers and segments, and also works with products and solutions for new segments.