LEARNING FROM MACHINE LEARNING:
TASKING PEOPLE WITH AMAZON'S
MECHANICAL TURK PLATFORM
How can we create a creative computer? The lead question of this research explores the act of creativity – what it takes to create or to be creative.
In order to programme an A.I. capable of creative decision-making, the scope and nature of creative processes first have to be understood in order to "define an undefined output". This artistic research project asks how and if inspiration can be staged. The structure of creative intelligence has always been an enigma of civilisation and is fundamentally unresolved. Not just from a standpoint of A.I. programming or following the myth of "every person has the potential to be an artist" – the question of creativity could be the most decorate How of our future : How can we teach creativity, what parameters can be used to test or assess creative intelligence to computers, to animals, to ourselves? How much creativity is necessary for a society to strive, or for a single individual to feel accomplished? Would a machine desire to be creative? How predetermined should the rules of play, a framework or a substrate be to provide a fertile ground for the departure into unknown creative territory?
> How would you describe creativity? Would you call yourself a creative person? How would you react to a task, what parameters would you prefer?
The general notion of creativity reaches from mastering a craft to patterns which differ in some way from norms and are borderline risky, such as beeing drunk or dizzy, tuberculosis, fever dreaming, dyslexia, drugs etc. can lead to somewhat creative results. These circumstances have lead to a cliché about those who work professionally in the creative fields : The artist as an outsider or even an addict, who unlearned certain codes of society, more or less deliberately. To much worries and stress halt creative processes, so in order to reach a creative state and to perform under pressure, time and time again, catalyzers to overcome mental blockades may be necessary.
> How and why is worrying an enemy of creativity? Do we need to programme creative A.I. to be constantly happy and mega confident?
> How are emotions linked to creativity? Are they essential to creativity? Could creativity be controlled, more constant and less cyclic?
This is more common knowledge than exact science, but at least the tendency of said phenomena must be taken into account: To much creativity can drive people crazy as it elevates the creator or the creatress into the intangible spheres of a hyper-highway of ideas. The mind can be lost in the realm of possibilities and infinite correlations – like a flurry of flashes in a storm cloud. Despite the supposed originality, such levels of creativity can not be followed by a common audience with a structured way of thinking. Creativity without direction or product may be forever lost in obscurity.
Resulting or resolving
In general, creativity stands in opposition to “the usual“, to the already known, existing structures. But at the same time, it can only exist within these structures, it needs guidelines or a frame. It can not be something entirely new or unknown; rather it must be based in this world in order to be relatable to other people. If creativity is an unforced, unforseeable neurological link, it must follow the intended direction close long enough only to switch path in the last junction, almost at random. The level of abstraction stands in contrast to the applicability of an idea.
> To what degree can the principle behind creativity be described as a detour within the logic of "the norm"?
> Why is creating easier when the options are limited? Is there an ideal amount of rules to enhance creative results? How are educational and political programmes handle this?
> Programming a computer, computing a programme: How is the programmer´s agenda linked to the agenda of the A.I.?
> Is creativity the quantification of creation? How is knowledge and awareness of a situation, a problem, position and context linked to creative thought?
The platform (www.mturk.com)
On this website run by Amazon, anyone can access the “golbal, on-demand, 24/7 workforce“ to offer so-called “Human Intelligence Tasks“ to workers worldwide. As information has become a valuable resource, the idea behind this platform is to reduce programming costs by essentially outsourcing software tasks to people in front of computers. Simply speaking: It is more cost-friendly to ask 2000 workers for 1 cent each to select a favourite image out of a stack, compared to programme a software that can analyse and value specific visual qualities. The URL itself is a reference to the mechanical turk („Schachtürke“) by the hungarian inventor Wolfgang von Kempelen, a chess-playing, fake-robot from anno 1769, a very impressive feat for that time.
The use of this or any similar platform opens up another field: Labour and identity in this globalised period of time. However, the project so far focused on a different topic, but this will certainly become a consideration at a later stage.
For the record, I want to state that the participating workers were treated with respect by being honest, in direct communication and paying as fair as possible – at least matching local standards. As payment was per task completed, this also depended on the workers approach to be fast or put more effort into it). Unfortunately, this was still way more compared to other requestors, despite only offering a decent level of salary.
> Why do we struggle to programme the concept of beauty and morals, emotions, understanding cynicism or reading handwriting? Why can we not rely on statistical data from archives or lexica when it comes to these issues? How can culture be programmed?
Video (first assessment)
As mobile media are spread all across the world, the possibilities to ask for visual replies is higher than ever. This circumstances serve as starting ground for the visual research: Specifically unspecific tasks were published online on the above mentioned platform. These so called human intelligence tasks required some sort of visual data, for example:
"Reinterpret Chips: Our company is looking for a fresh, widely favorable design. Open a bag or container of some sort of chips and make a pile. Now, select 5 chips you particularly like and present them close to the camera. Turn it around and highlight its qualities. Criteria: The chips must be visible, bright and in focus. Landscape format. (.....)“
The workers were informed about what the videos are produced for, and that they decide on what is audible or visible besides the chips, in this case. Showing one´s own face was neither forbidden or wished for, as well as other efforts. Showing other people or sensitive content on the other hand could lead to a rejection of the task. Some early results (processed by roughly 40 paid workers) were put together in a short, titled «learning from machine learning» for the exhibition AESTHETICS OF CHANGE in the Museum für Angewandte Kunst, Vienna. The exhibition tackled utopian questions about the future of learning and working at Universities.
»Learning from machine learning«
Video compilation, with audio
Answers to "select chips"
and "guess chess" tasks
(click to enlarge)