Artificial Creativity: a Historical Computer Sciences View

Artificial Creativity — a Historical Computer Sciences View

Lectures by Maarten H. Lamers (Feb 2019)

Ever since people have considered intelligent machines, they have also considered whether machines could be creative. In fact, questions surrounding artificial creativity are closely linked to those of artificial intelligence and machine learning.

Can we model creativity and could it be the result of computation? Would we want creative computers, and why? Are artificial intelligence and artificial creativity really different? What does a dog-jellyfish look like?

With the current hype regarding machine learning, more and more creative machines are built. Computers are composing music, writing cooking recipes, painting "artworks", even writing jokes. The outcomes range from being impressive to utter crap.

Over time, many interesting thinkers have formulated their views on what creativity for machines could mean, or how it could be brought about. We take a tour along the ideas of Ada Lovelace in the 19th century, Alan Turing (1950), Marvin Minsky (1982) and Douglas Hofstadter (1982).

Photo: Margaret Boden (photo source)

Interesting as their ideas may be, it was Margaret Boden who in the 1990's attempted to bridge the gap between the bottom-up builders of creative machines and the top-down thinkers that history provides us with. We discuss her framework and use it to contextualize historically and contemporary relevant computer-generated output.

From this analysis we gain an overview of what computers can do, but more importantly what they cannot yet. Moreover, we uncover the importance of machine learning as being the "science of conceptual spaces", which links back strongly to the thoughts of Alan Turing and Douglas Hofstadter.

Machine learning, being the computational manipulation of concept spaces, offers the methods for computers to be creative.

We end by venturing into the first steps of transformational creativity, as defined by Margaret Boden, through machine learning. And surprisingly, this is where Google's "Deep Dreaming" plays an important role.

Photo: outcome of 50 Deep Dream iterations on a jellyfish photo, using a dog-recognizer (2015, photo source).

Lectures

"Artificial Creativity — a Historical Computer Sciences View" was taught by me on several occasions, for example at Leiden University, HOVO, Dutch Ministry of Infrastructure and Water management, Dutch Ministry of the Interior and Kingdom Relations, and the Royal Academy of Art (Den Haag). It comes with a small reader, and will be further developed in the near future. It does not (yet) come with exercises, but if you have suggestions then contact me. Let me know if you are interested in this topic or my teachings.

Listen to my appearance in a BNR Questcast episode on algorithmic creativity (March 1 2019). In Dutch only.

Header photo: output of Google's "Deep Dream" (2015, photo source)