Extrapolating Creativity

STATE OF THE ART

We began the day with a little recap of what we had learnt over the last two weeks, discussing the language and metaphors used to describe AI technology, such as ‘neural network’ – a metaphor for the network of nodes within a computational model for information processing – and ‘deep’ used to describe a model that has many layers/dimensions. Hassan explained that the word AI differs in meaning depending who you are talking to, whether they are a computer scientist, member of the general public, a child etc… AGI (Artificial General Intelligence is the end goal of the entire research field), and ML(Machine Learning) is a subfield, AI is not all about learning systems. Hassan gave us the example of Deep Blue, chess playing computer that was programmed by programmers not ML.

https://www.nextrembrandt.com/

We were then shown the above photograph of a ‘Rembrant’ which to our suprise was completely digitally created in physical form, using 3D printing, and discussed what we miss when we view a painting through a digital screen.

CAN MACHINES BE CREATIVE?

How do we define the idea of creativity? One way the machine creates something new with ML is to intrapolate between data points, estimating and predicting what it ‘thinks’ should come next. Through extrapolation the machine takes all the data points and generates something beyond the data points, creating something new that is not part of the data.

https://arxiv.org/pdf/1911.01547.pdf

Hasan shared the above paper with us, written by Francois Chollet, creator of the Keras deep learning library, who also works for Google Deep Mind. In this paper Chollet proposed a new way to think about how we develop broad cognitive abilities in Machines. The level we are at with ML is only ‘Task Specific’.

In the example from Chollet’s paper below, shows a specific task that we as humans can easily achieve, however the machine models did not do so well. Due to the fact that the machine does not (yet) have broad cognitive abilities aka, common sense.

Hasan shared a selection of artists who utilise ML within thier work. This included Scott Eaton, Tom White, Anna Ridler. Ross Godwin, Pinar Van Arman, Roman Lipski and himself ofcourse Hasan S.