Thank you for initiating what promises to be a very interesting discussion, Bruce… and thank you for inviting me to be a respondent.
Before getting into more in depth discussion I thought I’d introduce myself and present a couple of opening thoughts as a way of easing in.
Between 2002 and 2007 I was working with AI loosely based on neural networks in projects like Small Work For Robot and Insects (https://www.hostprods.net/work#/small-work-for-robot-and-insects/), Fish Plant Rack (https://www.hostprods.net/work#/fish-plant-rack/) and Autoinducer_Ph-1 (https://www.hostprods.net/work#/autoinducer_ph1/). The AI softwares used in these works were developed by Brian Lee Yung Rowe and were very close to the pattern recognition algorithms which have found their main expression in the recommendation engines that power everything from online targeted advertising and social media feeds to Amazon and Spotify recommendations. The purpose of these AIs was to allow robotic/machinic systems to enter into some kind of dialog with natural systems (insects, fish, bacteria and plants). The AI served as a hub or interlocutor rather than to add any creative input of its own.
After hiatus of not working with AI at all, I am now working on new experiments with machine learning systems and mostly engaging with easily accessible online systems such as runwayML. The main project I am developing using this approach is called Catastrophe Jangled Hideously Out of Process (a name also conjured up by the algorithms behind junk emails over a decade ago) and generates images of disaster devoid of human agency or affect. (https://www.hostprods.net/work#/catjangled/)
These newer ML engines are highly attractive to artists seeking to create output without having to dirty our hands in complex coding (such as myself), the downside to that being that work produced using them has a very noticeable ML signature and can appear generic. If we are to follow the sampling analogy I would imagine that this signature will become less recognisable as technology progresses, but for now one of the bigger challenges seems to be to make original looking work.
So is current explosion of ML as an artistic tool similar to the early days of sampling? I'm not absolutely convinced that it is as far as the process is concerned. In techniques common to early music concrete and the early days of electronic sampling a discrete sound entity would have been manipulated in isolation and according to a series of human decisions and valuations. For me at least, using machine learning allows me to remove as much of the human value system as possible and let the machine take care of that, an approach which is highly appropriate for the project it is being used for. The more interesting discussions are to be had, as Bruce suggests, around notions of agency, ownership and deconstructive creativity.
I find the notion of training an algorithm to be very interesting in this context. Teaching a bunch of circuits to recognise a face, a vehicle, a word, a mountain or whatever leads us to a notion of the generic, recognising what things 'are' and removing what they 'could be'. Abusing this process by feeding non-coherent data sets into the algorithm can then produce a set of images (for example) of things that 'aren't' and 'could never be'. This is highly analogical to remix techniques and remix culture; a reimagining of the world based on information that the world has already produced. It provides opportunity to either consolidate the norms by which we recognise the world around us or to move into new modes of abstraction that all but erase the signature of the source material.
__ __ __ __ __ __ __ __ __ __ __ __ __
:: Andy Gracie ::
:: hostprods dot net ::
########################################################################
To unsubscribe from the NEW-MEDIA-CURATING list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=NEW-MEDIA-CURATING&A=1
This message was issued to members of www.jiscmail.ac.uk/NEW-MEDIA-CURATING, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
|