“First they ignore you, then they laugh at you, then they fight you, then you win.”
We have passed the laughing stage and moved into the fighting stage. The side of technological process will win, in the end. If AI is outlawed, then there will be many, many outlaws.
…nobody is demanding that AI (which, of course, really isn’t AI) be outlawed. Simply that it respects intellectual property rights as a baseline.
This somehow seems appropriate:
It’s not clear what ‘respecting property rights’ means in this instance. Is an AI reading simething that has been put online for the public to read, a violation of someone’s rights? Assuming it doesn’t make verbatim copies of the work, that is.
What I sort of wonder (but not very much, because I assume (? hope) the answer is “no”) is whether there are going to be unexpected consequences in the form of a resurgence of destructive collection of material organisms among naturalists and nature hobbyists, for example.
If an automated system can generate a completely believable “photograph” of any creature in any locality, how will anybody know which “attested” sightings are faked? I hope this won’t incentivize people to resume collecting actual rare eggs, feathers, skins, stones, etc., as a more prestigious replacement for now-infinitely-fakeable nature photography.
They can’t. They can only generate things that they have seen enough examples of to well understand. So if you want to generate an image of an Ivory-peckered Woodbill, you will need plenty of good photos of them in the training set. Same for every other creature and concept. And, since the AIs are learning concepts, not simply compositing images, the AI in question has to have come to the right conclusion about the data. For instance, I have found that Stable Diffusion 1.5 has reached a pretty clear idea of what a Teletubby is–a colorful, blobby alien biped. It draws them like that pretty consistently. Excapt none of them have televisions built into them, because the neural net didn’t grasp that concept and therefore draws them without them.
Right, but many such images do exist for many creatures. I’m not asking whether I can get an AI-generated photo that will convince scientists that I’ve found a Lost World-type survivor of a long-extinct species. I’m asking whether I can, say, win the Birders’ World annual photo contest with an AI-generated image that looks like a superb photograph of a well-documented creature in a well-documented place.
If the answer is “yes”, then as I said, I wonder if that could potentially shift the focus of hobbyist competition back towards more destructive forms of collection of material objects that are harder to fake.
Theoretically, some day? Yes. Current systems? None are good enough to be flawless.
Isn’t that largely illegal (at least in the US) anyway? I doubt many contests are going to accept evidence that violates the Migratory Bird Act. Even in general, saying “I have a sasquatch hide” doesn’t prove that you found it in the Ohio timberlands.
The “best” defense against AI pseudo-photography is some sort of digital watermark in the image, readable with the correct software. Of course, if someone can make the images with that watermark, someone else could potentially make them without with a cracked/edited version of the software.
…its respecting intellectual property rights. And if you want to know what it means, please feel free to refer to the various lawsuits and arguments made in all of the other AI threads you’ve participated in.
Its the people who built/and/or are using the “AI” who have used other peoples intellectual property as part of the training dataset without permission or compensation who are using that “AI” to independently generate wealth.
There are bogus facts in books, too.
Japanese PG-rated cheesecake models can give a (deep, heaving) sigh of relief for now.
There are hours and hours of slide shows of similar AI models on YouTube. Other than being preternaturally cute and a bit pneumatic, they’re far from uncanny.
I have to assume that more absent-minded AIs who forgot to get dressed that day are plentiful from other sites too.
A few comparisons between Firefly and Dall-E.
I’ve been playing with Photoshop’s generative fill. It’s very good, and very useful. Integration with Photoshop makes for a much better experience as well.
I just spent a solid week using Generative Fill, and have it down pretty well now. Already use it for work, as well as play. It’s not without bugs (they have the “not permitted” error set too sensitive for one thing, and most of the errors make no sense. Even if you go near selecting a breast or groin area to retouch, it gives you a warning and stops generating…very frustrating).
But Generation Fill is only as good as you are with correct prompts, and just as importantly, good selections for the In-Painting touch-ups you want it to do. It can take many re-tries to get the results you want. The learning curve is worth it if you’re into graphics (as I am). Overall, it’s pretty amazing and I can’t wait for the updates.
Even more amazing is MidJourney (similar to Dall-e, but much better, IMHO). I’m spending this week learning MJ. I’m pretty gobsmacked at its potential already. I feel sorry for professional photographers, graphic designers, stock-photo companies, etc.—I believe their jobs will soon be in jeopardy. However, those that use it as a tool to enhance their talents, instead of using it to create a final product, should be OK, at least for a while.
What I’d like to see in the future is AI being used to tutor kids (and some adults). But, not just a tutor; a childhood friend, learning coach, and talent groomer, too:
Problem: Many humans don’t realize their true talents until it’s too late, if ever. For example, would folks like Mozart and Beethoven have become great if their talent wasn’t recognized at an early age, and been groomed from that point forward? How many “Mozarts” are there that never happened because they weren’t introduced to music and groomed from an early age? Many, I suspect.
Solution: Conscious AI (coming soon, I believe) tutors that are available to everyone (open-source). It should be available (and fine-tuned) for everyone from ~age 5 onward. It would be like the kid’s smart friend who teaches and monitors, steering the child in the directions he/she shows interest and talent.
AI should not take the place of parents, or human teachers and friends, because kids need social interaction with other humans, lest they become recluses. The AI should just be an addendum to their upbringing, not a replacement.
Even better would be having the AI tutor incorporated into Boston Dynamics-type robots. The robot will grow (in both mental capacity and physical stature) as the kid grows (upgrade to the next age-appropriate model, as needed).
To guard against only the rich being able to afford an AI robot for their kids, they should be government subsidized, the cost based on family income (like the .gov health insurance marketplace). Otherwise, the have-and-have-not divide will only worsen.
Let’s not allow AI to outperform us in virtually every category of job description (something it will have no problem accomplishing soon). Let’s instead use it as a tool to make us better.
Have you visited the long thread here on digital art? Started in the long-ago primitive days of a year ago. Early parts are gushing over things that are laughable now.
Yes, it’s come a long way…baby.
Pre-version 4 of MidJourney was impressive, but v.5.1 is entering the pro leagues. It has a lot of functionality I still want to see come to fruition (and, I’m confident that it’s coming down the road), but the output today is stunning.
Who needs flying cars and food synthesizers (à la Jestons), when you can create real art in minutes, with little talent (though talent helps)?
My current tangent is with the Dall-E 3 (or whatever they end up calling it) found in Microsoft Bing. You can make some truly weird-ass shit there. (I created this one a couple of minutes ago.)
(Of course it goes without saying this is just a rip-off copy of a real artist’s image of…um…that and not something that never has existed before and probably never should…)
To make real art that stands out from the millions of generated images will still require talent and artistic vision.
When photography came along, I’m sure realist painters that spent a lifetime trying to capture reality were pissed that someone else could just press a button and create a portrait. But it turns out that great photography requires artistic talent as well, just of a different kind.
What we should be celebrating is the democratization of artistic vision. People who have an artistic vision no longer have to spend years perfecting mundane actions in order to bring it to life. People with a story to tell no longer have to find intermediaries to sell the idea to. They don’t need teams and production budgets and all the rest. A vision and an AI are enough.
Of course the current people on top of these industries and activities will dislike it. Cel artists were probably pretty annoyed at the computer graphics guys putting put ten times the animation content per day, too. But raising their productivity didn’t put artists out of work; it gave them the ability to create greater visions.