Simon Colton Please follow me @SimonGColton on Twitter

0 downloads 108 Views 13MB Size Report
We aim to be the Deep Mind of. Computational Creativity. • We want to democratise videogame design through advanced AI
From #CreativeAI to Computational Creativity and Back Again Simon Colton MetaMakers Institute, Games Academy, Falmouth University Computational Creativity Group, Computing, Goldsmiths, UoL and Imaginative AI Ltd.

Please follow me @SimonGColton on Twitter

One Point of Philosophy •

I believe the notion of creativity is essentially contested (Gallie ’54) in that its proper usage involves endless debate about its proper usage •

If we’re not arguing about creativity, we’re not talking about creativity



We’ve agreed to disagree forever about what creativity is, as this is a driving force for society. This is scary, but ultimately a good thing

One Point of Reality

“The Oscars in-memorium section this year will show all the black guys who were shot by cops on their way to the cinema” (paraphrased)

“Just saw @Oprah's empowering & inspiring speech at last night’s #GoldenGlobes. Let’s all come together, women & men, & say #TIMESUP! #United”

One Point of Reality •

It’s obvious that background personality, motivations and process change the meaning behind messages, and much of art has message



By whom/what, how and why an artefact is created is very much taken into account when we judge the value of that artefact



Authorship, ownership and authenticity are particularly important in the arts - this is a good thing, as a celebration of humanity (as well as a way of being elitist, for sure) •

In particular, the notion of the death of the author ideal didn’t survive in the age of computational generation of works of art, music and literature



Computational Creativity: the art, science and engineering of computational systems, which, via taking on certain responsibilities, unbiased observers would deem to be creative •





Weak and strong aims for projects

Creative AI: Training and employing generative AI systems (e.g., deep neural networks) to make art. Seemingly with 2 maxims, which are not always emphasised in Computational Creativity •

Quality of the artefact: up to human++ levels



Mass adoption of human-centric techniques through consumer apps such as Prisma

Maybe this is what’s bugging me: Creative AI projects don’t seem to care much about the AI dream, whereas Computational Creativity research is very much about working towards this dream

If you are interested in the AI dream… •

Here is one journey from a generative system to a system which would enhance our understanding of Computational Creativity and Artificial Intelligence generally •



Based on The Painting Fool: 17 year project to build a system which is taken seriously as a creative artist in its own right one day (www.thepaintingfool.com)

Roadmap of seven levels of autonomous behaviour in the system •

Does not have to be in academic research

Generative Systems •

So, you’ve designed a generative system and are having fun making pictures with it. You play around with input data and parameter settings, and realise that the output is not only high quality but really varied



You write a little GUI which enables to you play around with the inputs/parameters, and this increases the fun and variety



It’s clear that the space of inputs/parameters is very big. You begin to suspect that the space of novel outputs is also vast



You’re at LEVEL ONE: an interesting generative system

Appreciative Systems •

Generating images becomes addictive and you gorge on the output. In your gluttony, you get a strong fear of missing out what if I miss the parameters for a really interesting picture?



You decide to systematically sample the space of outputs, but there are millions of images that can be produced



So, you somehow encode your aesthetic preferences into a fitness function and get the software to rank/display the best according to the fitness evaluation, perhaps tempered by a novelty measure which keeps things fresh



You’re at LEVEL TWO: an appreciative system with some autonomy (Harold Cohen never chose to go this far with AARON)

Emotion-driven Portraiture

More Portraits

Joint work with Maja Pantic and Michel Valstar

Artistic Systems •

At some stage, some humility sinks in, and you begin to think that maybe … just maybe … your particular aesthetic preferences aren’t the only ones which could be used to mine images



You give the software the ability to invent its own aesthetic fitness functions and use them to filter and rank the images that it generates



You’re at LEVEL THREE: The software itself has some potential to affect the world artistically

Cityscape Results • Each scene exhibits at least two patterns, and there are some surprises...

Persuasive Systems •

Some of the output is great - beautiful new images that you perhaps wouldn’t have found yourself. But some of the pictures are unpalatable and you can’t imagine why the software likes them



However, sometimes the images grow on you and you realise that your own aesthetic sensibilities are being changed by the software. This is weird. But fun.



You want to give the software the ability to influence you more easily, so you add a module which produces a little essay as a commentary on the aesthetic generation, the artefact generation and the style that the software has invented



You’re at LEVEL FOUR with software that can change your mind through explanations as well as high quality, surprising, output

Inventive Systems •

You begin to realise that you enjoy output partially because of what it looks like and partially because of the backstory to the generation of it and the aesthetic that it appeals to



You want to increase both aspects, by enabling the software to alter its own code, perhaps at process level, and by taking inspiration from outside sources like newspapers, twitter, art books, other artists, etc., so you have less control



And you add natural language generation to turn the commentary about the process/product into a little drama



You’re at LEVEL FIVE, where what your software does is as important, interesting and unpredictable as its output

FloWr

Authentic Systems •

You’re loving the commentaries/essays/texts about how and why your software has made a particular picture/aesthetic/style/series or invented a new technique, and the software pretty much has an artistic personality



However, sometimes it doesn’t really ring true, such as the series inspired by childbirth. You realise that you’re reading the output as if it were created by a person, which is a falsehood you don’t like



You give the software plausible and believable reasons to be creative by implementing models of intrinsic motivation, reflection, self-improvement, self-determination, empowerment and …?



You’re at LEVEL SIX where your software is believable, authentic and fully independent of you

Philosophical Systems •

Ultimately, you find it thrilling to be in the presence of such an interesting creator as your software - it’s completely independent of you, and it teaches you new things and inspires you and others



You realise that for the software to be taken seriously as an artist, it needs to join the debate about what creativity means, as an essentially contested concept and societal driving force



You implement philosophical reasoning based on its own creative endeavours and enable it to critique the thoughts of others. You add dialogue systems to propose, prove and disprove hypotheses about the nature of creativity and generally provoke discussion and thought around the topic



You’re finally at LEVEL SEVEN, where it’s difficult to argue that your software isn’t creative. We’ve advanced towards the AI dream

And Back Again

#CreativeAI uses in Computational Creativity •

The ease of use (in principle) of deep generative models for things like style transfer are very encouraging

Ghaith Abdul-Ahad

Ludando System Built for Imaginative AI Ltd. - Please ask for details

And Back Again •

The imaginative power of deep learning models to hallucinate things which don’t exist is remarkable •

Will be using #CreativeAI projects to discuss imagination

Nvidia

Old Hanoi Series

SpaceHopping TreeHugger in Hanoi!

Imaginative AI Ltd. •

Founded in 2017, UK based



We aim to be the Deep Mind of Computational Creativity



We want to democratise videogame design through advanced AI-support





So that it becomes as natural to communicate through games/play as through text/images/video



Instagram for games

We are looking for investment!

www.metamakersinstitute.com ccg.doc.gold.ac.uk www.thepaintingfool.com www.gamesbyangelina.org www.whim-project.eu www.computationalcreativity.net www.prosecco-network.eu www.wevvagame.com

Questions? @SimonGColton