Neural Networks & all that. The new Artificial Intelligence

A challenge

Frieder Nake

Summer 2019, Hochschule für Künste Bremen

Mondays 10 – 14, HfK 4.15.070

Starts 15 April 2019

Look at this quote taken from an announcement of Artificial Intelligence (AI) components on the market:

Although some so-called AI applications aren’t actually cognitive, there are technologies capable of achieving human- or superhuman-level intelligence on given tasks. (from an announcement in O'Reilly's publication "The New Artificial Intelligence market")

Are we supposed to swallow this? Do you do it? I mean, you are studying at an academic institution, so you may supposedly be scientifically- or artistically-minded. So, maybe, you ...

Here is another question: Do you believe that machines can learn? Before you answer, think about how it was when you learned how to ride a bicycle, or how to bind your shoe laces, or when you learned to speak. And how was it when you had to do calculations? How do you do 26 x 73, now, that you are grown up? Do you do it in your head?

A study is claimed to have shown that under certain circumstances human beings may be willing to sacrifice other humans' lives in order to save a robot. It seems to be enough to have the robot look a bit like a human.

But you also read this kind of news: "Organizations looking to benefit from the artificial intelligence (AI) revolution should be cautious about putting all their eggs in one basket, a study from the University of Waterloo has found." (Science Daily, Jan. 17, 2019) There seem to be problems in measuring success when machines are supposed to learn. Does this come as a surprise to you?

The really big current hype is, of course, called CNN. Now, as you will know, these three letters do no longer stand for Cable News Network, they rather mean Convolutional Neural Network. You know what a network is. That's somehow trivial. You know the word "neural" from talks about the brain. Do you know anything about "convolu¬tions"? Maybe only vaguely. But you may, as many of your friends, currently be doing something with those CNNs. Should we try to get behind them, perhaps even try to understand what they are? I suggest, we should do this.

Therefore, I offer this course. It will be divided up into three parts:

• a short look into the early history of what then came to be called "Artificial Intelligence",
• the main part on neural networks, including a bit of mathematics because, without that, we cannot understand what they are,
• and an outlook on aspects of what is now considered to be, or at least is subsumed under, the name of "Artificial Intelligence": machine learning, big data, "4.0".

Our approach will be critical and skeptical, historical and systematic. We should not jump onto the band waggon of digitization, rather try to learn what it means to turn activities, that humans do intuitively and skillfully, into computable functions. A main goal of our efforts should be to keep our great human capability of mistrusting the great promises of advanced technology (or of the people behind that), and rather acknowledge the fact that our lives are finite. Human life always happens against the horizon of death. That's also the horizon of human intelligence which is always historically and culturally determined.