Relax – we're not going to 'accidentally' create killer robots. Niklas Andersson, AI fan, BI consultant and data scientist in TietoEvry speaks AI.
A few weeks ago I attended a big event arranged by one of our partners. My colleagues and I were focused on gaining a better understanding of how business analysts, BI developments and data scientist could work together on creating advanced analysis and machine learning solutions. But there was another issue that couldn't be avoided, one that cropped up in every discussion, if only on a more general level. I'm referring to the difference between AI and AI. And yes, there are two types: the one we use today and the one that belongs in a distant future. Since then, I've been asking myself the same questions over and over: how can large companies like ours discuss AI in the same language as our customers? How can we reach consensus and strike a balance between expectations and reality?
I thought I could start off with the type of AI that actually exists today, the one which is popularly known as AI but which actually should go by its proper name: narrow artificial intelligence (NAI). This type of AI is data driven, and is developed to resolve specific tasks with a view to removing humans from the equation. NAI can be applied in an infinite number of areas: everything from process automation to minimising or excluding the need for manual input from humans or creating smart assistants. The change occurs with the help of the data-driven methodology that enables NAI, drawing on the potential that lies in underlying data concerning the task to be resolved. We simply let our models replicate human abilities and perform the work by feeding them examples of how we want them to act until we are satisfied with their performance.
Saying that you worked on AI got you ridiculed at a cocktail party [under AI vintern på 80- och 90-talet]. Today you don’t get invited to a party unless you work on AI. – Oliver Schabenberger.
The AI that exists in reality is the narrow type I mentioned above. It differs distinctly from the type of AI that is considered all over the world to be a potential threat to humanity, namely artificial general intelligence (AGI) or general AI. Unlike NAI, which adapts algorithms to data with no 'mind', AGI is a type of AI which not only is capable of learning how to resolve a specific task – it is also capable of doing everything we humans can do; AI that can think and act autonomously. Or to put it another way: think Terminator and Westworld.
When AI terminology is left unchecked and used to describe everything from highly technical and impressive solutions to simple implementations of basic "if this, then that" logic, there are bound to be problems. Especially for those of us who are expected to deliver advanced analytical solutions to our customers. The fact that AI ≠ AI means that expectations of what AI is capable of do not always match reality. But there is a way of avoiding the problem. First, we should all express ourselves clearly and use the term narrow AI when we want to refer to the type of data-driven solutions which, with the help of analysis and models, simplify everyday life. The general public may not be very impressed, but at least we can assure everyone who thinks that we might accidentally create something which will pave the way for future world domination by robots; as already clarified, we can leave that to AGI. We need to clear up misconceptions and reach consensus, both with our customers and with society at large, in order to keep discussions on the right level. As you may have noticed, AI and related subjects happen to be among my favourite topics of discussion.
/Niklas Andersson, AI fan, BI consultant and data scientist in TietoEvry