From boom to burst, the AI bubble is only heading in one direction (The Guardian)

Good commentary on this from John Naughton (see https://en.wikipedia.org/wiki/John_Naughton ):

https://www.theguardian.com/commentisfree/2024/apr/13/from-boom-to-burst-the-ai-bubble-is-only-heading-in-one-direction

The 5th stage is panic when the bubble bursts.

Naughton asks at the end of the final paragraph: "Are we caught in an AI bubble? Is the pope a Catholic?"


My thoughts:

1) The sooner this bubble bursts, the better. Generative AI isn't just causing harm. It's diverted way too much capital, talent, and resources to flawed tech which is unlikely ever to be reliable enough to deliver on promises made by AI proponents.

2) As a society, we're going to need a 6th phase - to clean up after generative AI.

Because so much AI trash is already out there.

AI-created websites pumping out nearly endless amounts of misinformation.

AI-generated scientific papers, sometimes "peer-reviewed" by AI, are adding more and more errors to scientific literature every day.

Image platforms are rapidly being overwhelmed by AI-generated fake photos and art. Yesterday I saw a tweet from someone who wanted to use stock images in a book he's writing, but does not want AI images, who was told by all the stock image platforms he checked with that none of them could guarantee him that images he'd pick wouldn't be AI. I also read that a large fraction of the AI images on Adobe - and nearly half of the fantasy images - are already AI. Many kids are also giving up on learning to create art themselves because they feel they can't compete with a tsunami of AI-generated images.

Publishing platforms have been overwhelmed by AI-generated books.

Music platforms are being swamped by AI-generated music that allows people to pretend they're creative by giving brief text prompts to AI that's obviously illegally using vast amounts of copyrighted material because with music (as with images) you can get near-exact replicas of copyrighted work just using adjectives that have been used to describe them rather than the artists' names.

Teachers are having a harder and harder time discovering if their students are actually learning anything because cheating via AI has become much more widespread than cheating without AI was in the past. One estimate, from a Scottish university if I recall correctly, was that it's 4 times worse now, and taking up more and more teacher and administrative staff time to deal with.

We don't yet have such unhelpful medical AI as chatbots using GenAI, which hallucinates, and which will be marketed as $9/hr replacements for nurses, but those are in the works.

The companies pushing GenAI on us have been well aware that their AI tools are unreliable. Many of them have crafted TOS designed to make individual AI users responsible for any harm done by their tools, including violations of copyright. You don't expect search engines and business tools to give inaccurate/imaginary answers? The companies tell you that it's on you to catch all their flawed tools' failures...and oh, you should also view the inaccurate results as "creative" rather than wrong. When it first became obvious Google's Bard got things wrong, it was relabeled as a "creative companion" and marketed as that. Marketing problem solved, or so some of the top people there thought, but a lot of Google employees were reportedly unhappy with their company's willingness to peddle a flawed product.

Basically, GenAI companies have slimed the world.

Yesterday I saw a person who's an AI enthusiast admit that we'll have to assume now that anything created since a couple of years ago has been polluted by AI. I saw AI critics referring to pollution by AI already over a year ago, but it was a shock to see an enthusiast talk about its effect as harmful.

And unless we want to live in a world where more and more of what we see and hear is untrustworthy, we're going to have to clean up after the mess AI has already made. The AI companies didn't give us tools to identify what's produced by AI, and apparently never intended to. The intent was always to fool people.

Yesterday Microsoft announced a new AI to create video from a single image of a face. I guess they thought we didn't have enough deepfakes yet.

It can do this:

Link to tweet

it can also do this:

Link to tweet