Dreaming of AI, While It Dreams of Us

C2C's Web Producer Lex's new article published on Medium goes on a journalistic odyssey exploring generative AI's connection to the dream state. Below is an excerpt:

Creating “photography” in the generative AI program DALL-E (in tandem with Bing) was a revelation to me, evoking the old-school thrill of a photo darkroom where a picture appears before your eyes. But instead of a film negative, it’s a written prompt that teases out the AI photo, as though plucked from the dream state itself and the hidden motherlode that underpins it.

OpenAI, the makers of DALL-E, “scraped” more than half a billion images off the Internet to form their database (other AI programs like Midjourney and Stable Diffusion use a similar process). While the commandeering of this material is not without controversy, the result is a deep well of protean possibilities. Primed with a surreal prompt, DALL-E seems more than happy to deliver a postcard straight out of the subconscious. The fact that it can be a “photograph” rather than an illustration adds an uncanny valley level of verisimilitude. Real, but not real…like our dreams.

Over on the Singularity subreddit, rdyazdi posted an intriguing comparison of generative AI to human dreams. Both can be “creative, unpredictable, and at times, quite abstract…[and] tap into a vast knowledge base, connecting seemingly unrelated concepts in fascinating ways,” they write. The limitations of AI and dreams also share similarities, as “they might not provide coherent or accurate information. In the same way that dreams can be influenced by our emotions, biases, and experiences, AI models are shaped by the data they were trained on,” rdyazdi added.

See the full article here.