Thursday 28 February 2019

Artificial Intelligence: Learning to see

Memo Akten: Learning to see

One of the most interesting aspects of computer aided drawing at the moment is the way that AI is being used to open up ways to think about how we think and why we do the things we do. Artificial Intelligence is inseparable from learning software and what it does as it learns, tells us a lot about why we think in the way we do. The artist Memo Akten is looking at how to visualise one of the fundamental problems we have as human beings. The problem of not being able to see another person's point of view. Akten's video of his computer simulation at work makes it clear why we have problems. On the right side of the screen we have images made by a computer program that has been 'fed' with images of the sky, sea and breaking waves. This is the world that this program has 'learnt'. It expects therefore for everything to conform to the behaviour of rocks and sea and sky. It looks for waves splashing when they meet obstacles like rocks, it expects the swell and movement of water to be contrasted with the hard immovability of rocks. It expects the moving areas to reflect light, (the sea and sky) much more readily than the fixed points, (the dark rocks). Edges between moving surfaces and static ones are seen to be where waves break up and produce foam and splashes. The rhythm of sea movement is recorded over and over again, so that the software understands the nature of these rhythms and how they work.




The way the sky gradually changes tonality is understood as a particular curve in space. After being fed thousands of images of the sea and rocks, the software can generate images that are a typical, average view of sea and rocks. This is its world and it knows it well enough to reproduce it. But what happens therefore when its camera or its information feed is trained on something different? In this case a yellow duster, hands and a piece of blue fabric, some wires, a plug and other bits and pieces.



The program can't see the dark mass as a plug, it can only see the experience as a rock. The duster is read as sea, a fold in the duster as sea breaking over a rock, the edge between one material and another the edge between the sea and the sky, a moving hand becomes a sheet of water.

The implication is that if you as a human being grew up in one particular culture/environment everything you would see after that would be 'coloured' by your initial experience. In fact it would be impossible to 'see' the world as it really is, because the world is not a series of objective things, it is a set of relationships, that are read in as many ways as there are readers. The person calling for Brexit is not going to be convinced by another's argument that we should remain, because their world view is so different. Of course gradually the camera feed would educate the right hand world and if the same amount of hours input were given relating to this new environment, dusters would become to be seen as dusters. However because viewpoints can take a lifetime to be constructed, many of us can't take that time to immerse ourselves in another person's point of view, therefore we will forever see the duster as a sea and the plug as a rock.

See more about Memo Akten's 'Learning to See' here

More on computer generated art
Making a start in machine learning





No comments:

Post a Comment