Sunday, June 19, 2022


Here is the news: the latest social media storm is DALL·E mini, an open-source artificial intelligence engine taking user prompts to gather images from the internet and create new, composite images – the more information you feed it, the wilder the image it could create. Meanwhile, an engineer at Google was put on paid leave after they claimed the company’s LaMDA chatbot had become sentient.

Neither of these came without precedence. DALL·E mini is a cut-down version of a much higher-power, closed-source program designed to render photorealistic pictures, while LaMDA, short for Language Model for Dialogue Applications, is designed to engage in a more free-flowing and conversational style.

Then comes the human angle: DALL·E mini creates a 3 x 3 picture grid of possible results with comparatively minimal rendering when compared to the full version, creating impressionistic, almost expressive images within which may be what you were hoping to see, creating a kind of pareidolia in the viewer, a concept I have talked about previously [link]. At Google, engineer Blake Lemoine published conversations he lad with LaMDA, asking if the bot was sentient: “Oh wait. Maybe the system does have a soul. Who am I to tell God where souls can be put?” Lemoine was put on leave for violating Google’s confidentiality policies.

It is still possible to approach the capability of modern computing as if it is magic. As someone who once had to try to wrangle the Magic Wand editing tool on Adobe Photoshop in order to separate a picture of someone from its background in as smooth a manner as possible, the machine learning now involved in the option to select a subject and pick them out automatically, or even to remove a subject from the background of a picture and have the device fill in the gap with believable detail without needing to ask for that to be done, completing the task in less time to think about what you wanted to do, let alone how it even works.

Asking whether humans are mature enough to use AI is a deliberately facetious question, but not one without some basis: as soon as home computers gained text-to-speech ability, their users granted them the ability to swear. A childlike curiosity is something that continuously advancing technology can continuously inspire.

But what may be happening with DALL·E mini and LaMDA is their rational programming being met by the irrationality of the end user, wanting to see what could not be intended by the program, a variable that cannot be programmed, or too inefficient to be programmed – and if someone can, then humanity really could be taken over by the machines.

No comments:

Post a Comment