Artificial Intelligence

Artificial Intelligence

Following on from my visit to the artificial intelligence laboratory of the University of Zurich I thought I’d highlight some of the topics they’re dealing with:

The first question to answer is that of intelligence. And when you try to analyse that you will start to see a host of problems – how do you actually define intelligence? We all seem to know what it is when we see it. We talk about intelligent people, smart people and clever people. We all know the IQ test but does that mean, for example, that dolphins are not intelligent because they can’t do an IQ test. Gardner developed the theory of multiple intelligences in 1983 and this has become widely accepted, at least in popular literature. Yet we don’t really class emotional intelligence as intelligence per se do we?

Now the classic approach to artificial intelligence was to simply build number crunching machines and find ever complex algorithms. Yet that has failed to spit out something remotely “intelligent” in the form of human intelligence. Many argue this is doomed to failure (and I agree) because you cannot deal with human intelligence without taking the body into account. At first many may think this is illogical - our brain is after all the seat of intelligence so why not just replicate the brain’s processing ability. But here lies the key – our brain grows with our body and our body influences the brain. To understand the concept of seeing you need to see, same goes for any senses. Our language is rich in human emotions and body references (not to mention body language) and human concepts – even such simple words as “excited” refer to an emotional context which has a visceral response: we feel good. How do you explain the concept of “excited” to a computer with no body and no senses?

Indeed we can argue that intelligence is only present because we have a body. For if we were a one-celled organism and could live without moving there would be no need to do anything else. Only because we needed to move to get nutrition and / or mate and reproduce did we develop brain power to co-ordinate this and then develop ever more powerful strategies to eat, mate and protect ourselves form the dangers of the world.

The interesting thing when we’re looking at replicating human intelligence and functions is how complicated some factors are to programme into robots that are so simple for us. Walking, for example, is incredibly difficult, not to mention running - nobody has yet managed to construct a running robot. Picking up a cup – in fact even knowing that a cup can be picked up, in fact even knowing a cup is a cup (think of all the different forms and colours a cup can take) - is a complex task. So many of the things that we find incredibly easy to do raise challenges for producers of robots and intelligence that are today far from being solved.

What else can we see? We can actually see that because we have intelligent bodies it makes our life easier – having skin and soft touch fingertips means that we can easily pick up a glass – which is very difficult for a robot with metal fingers to do. Our walking is controlled not neurologically but often by the elastic reflexes in our muscles. Our muscles and bones can also repair and have a softness that means as children we can fall down a lot while learning to walk (robots break!). So we owe our intelligence and functionality to our bodies being made of intelligent materials.

These are all insights that the AI Lab in Zurich are dealing with and looking into and finding out what intelligence is in the process and new definitions and forms of intelligence on the other hand. Their work deals with all forms of embodied intelligence and it is showing us how simple some processes can be and how complicated what we consider simple can also be. And this discovery processes is inhibited along the way constantly by our frames of reference (seeing things from our human eyes) and anthropomorphization (adding human qualities where none exist).

Anthropomorphization is an interesting concept - and holds true in many parts of our life. Take the so-called “Zurich robots” produced at the AI Lab. These are simple robots with a simple circuit that drive around in a confined space and will push blocks of polystyrene into piles. What clever robots many people will explain yet their circuitry is exceptionally simple. Because we see them apparently clearing up the area we add a quality of behaviour to what they do when non exists (we also do this to cars and computers and pets). And of course as soon as we add a human-like body or face our identification with human type behaviour rockets. Yet this is in our mind not in the robot - and, of course, raises many fears (supported by many films such as iRobot, terminator and many more).

The AI Lab in Zurich has built an extremely lifelike running “dog” using very simple electronics but taking into account spring reactions of muscles and natural coordinative states. Showing that some complex movements can be quite simple - to try to programme all muscular and body movement in a robot would be incredibly complex but the approach they took was to use natural springs and coordinative states and ended up with the best and first running four-legged robot. Yet to build a two-legged running robot is so far out of the reach of the world’s laboratories (and remember this is just locomotion not even thought processes).

The same applies to tentacles of octopuses which look like complex movements yet can be simulated with a few “tendons” and simple movements. The goal of the AI lab is to build a lifelike octopus.

Another interesting element of modern artificial intelligence is the use of evolutionary programmes. These have been used to successfully design, for example, antennae for satellites that are more efficient than anything designed by humans - and it is difficult to see how we could have come up with anything better! Obviously evolutionary intelligence will be one of the ways forward - after all our intelligence is also grounded in our development.

So all very, very interesting stuff -and what about intelligent plants? Well that is another part of the research being carried out at the AI Lab. I greatly look forward to my next visit.