“Dreams are a biological necessity … Dream logic seems to proceed on associations.” – William S. Burroughs.
Dreams serve a biological necessity in processing and exploring subconscious thoughts. Dreams are deeply personal, subjective experiences that people ascribe meaning to and in relating to one other. Dream Machine is a multi-player mobile application paired with a neural device that allows the user and their partner to select a theme and dream together.
How It Works?
1. The application collects and analyses your daily chats with people with whom you wish to share a collaborative dream.
2. Keywords from chat conversations are sent to a Machine Learning System which uses GPT-2 to perform ‘text to image’ and StyleGAN to create photorealistic images to mimic dream-like vision. For this exercise, the team trained their model on RunwayML using images of Colosseum and Italy. The video clip at the end is created by training StyleGAN model on RunwayML using images from Bing and Google Images.
3. This algorithm performs text to image conversion and creates a dream-like video collage. The video is pushed to the user and the partner’s brain through neural implants while they were sleeping.
4. This neural device tracks the user’s Rapid Eye Movement and dopamine levels whilst asleep and at peak levels takes a snapshot (postcard) of the visual output.
5. The user gets an option to share the postcard with the partner. The Dream Machine leverages the sleep state to unite faraway couples.
Like contemporary IoT devices, text capture which is sent to a cloud for processing raises the potential for exploitation of this intensely personal data.
Disturbance: Whilst input is passive, there is potential harm (ie. interrupting sleep cycles).
Accuracy: Machine Learning models often fail miserably in certain scenarios, impact analysis of their failure cases is crucial.