@hugFMP work-in-progress exhibition
We held a work-in-progress exhibition in the studio where students and staff from IID and other affiliate courses such as IDA and MA UI/UX where invited to see our progress. It had to be a prototype, experiment or manifestation of our concept – rather than a presentation of your research.
For the exhibition we had to explain our concept in a small paragraph, the paragraph I finalised for the exhibit was this;
We are increasingly presented with software that imitates human behaviour or appearance. AI generated imagery and conversational agents are more difficult to identify as non-human than ever. In the dystopian fiction Blade Runner human ‘replicants’ are subjected to a ‘Voight-Kampff’ machine, which helps an operator identify the synthetic humans. What would a ‘Voight-Kampff’ machine of today be and how could it be used?
The information tries to give some further information about what deep fakes are and how you can identify them. To help explain how in 5 or maybe 10 years time these have the means to become difficult to differentiate from a image of a real life human being.
I found that most people couldn’t identify the imagery when presented with a mixed set of photos with real people and deepfakes.
Only until presented with the hints or pointing out some of the issues and flaws with these ‘Deepfakes’ were then people able to identify and differentiate between them.
For example the image second from the top left – one noticeable weird generation is the hood merging into the background. This can also be seen in the bottom left, which people can use as a clue to start pointing out other deepfakes making it easier. In contrast to the bottom right image, looks like a real person, however, one noticeable issue with it is a blur in the hair on the right. Although this could be noted as a print image – as I had originally printed them out onto paper for demonstration.
I had some suggestions about perhaps ‘gamify’ the experience so people could learn how to recognise and identify deepfakes.