Predictions about the impact of AI-generated imagery

ahalpert said:
It really depends on what you're doing. If you're a technician, then yes. If you're a designer, absolutely not. Here's an article about AI assisting design students, which I think will be the norm for anyone directing a creative process: https://www.core77.com/posts/117169/...-their-process

Or there's a Nike commercial that made a big impact on me 12 or 13 years ago. If they made it with AI, I'm sure that the AI would do most of the assistant editor work of tracking down clips and possibly also recommend shot sequences, but a human would be making the final decisions and probably also need to come up with the concept for the climax, the legless guy running: https://www.youtube.com/watch?v=4LZMJgD396M

seems like you think most of your work is automatable and you're projecting
Click to expand...

Honestly, I wish I had your naive and innocent brain (that is a genuine compliment, I am jealous of it).

AI will always enhance and help humans, what's the point of the article? And then it will eventually enhance and replace.

And from my understanding, AI has already replaced a few design jobs in NYC studios where only one senior human designer is usually left as a lead to manage the project(s)...so, yes, absolutely yes.
 
That's what I said, dude. There will be a last man standing at the head of the creative process. The herd will be thinned, and only the top talent will remain, along with menial laborers to service the machines.

But I grow tired of conversing with you and welcome the contributions of others.
 
We actually don't converse much anymore, do we? And there are no others, lol.

In short, I never disagreed with humans having to exist and work alongside the machines; I just really despise this notion from you or anyone else that our work is so darn difficult and important that machines won't be able to replicate the results for much of it.

It's like come on already.
 
ahalpert said:
Funny, I just read an article about Tom Brady where they quoted him as saying, "Earlier this week, I made a statement about playing soccer and the military..." and then I realized it was on the English site for a Spanish-language newspaper called Marca. Must have used auto-translate - https://amp.marca.com/en/nfl/tampa-bay-buccaneers/2022/10/23/63548e37268e3e4a5b8b45a4.html
Click to expand...

Marca is a major newspaper. I used to read it a bit a while back, when Messi still played for Barcelona. Its English pages are edited but not necessarily for American audience. Overall, the Europeans have an odd position on translations. An American friend of mine, fluent in Italian, applied for a translator/editor job with La Gazzetta dello Sport. They hired someone local instead. Others like Marca might have a few pages in English. And yet other sport heavyweights like the Kicker and L'Equipe have no online English presence at all.
 
DLD, don't miss the augmented reality intro to MNF tonight (it's from the Patriots side, not the networks)...
 
DLD said:
Marca is a major newspaper. I used to read it a bit a while back, when Messi still played for Barcelona. Its English pages are edited but not necessarily for American audience. Overall, the Europeans have an odd position on translations. An American friend of mine, fluent in Italian, applied for a translator/editor job with La Gazzetta dello Sport. They hired someone local instead. Others like Marca might have a few pages in English. And yet other sport heavyweights like the Kicker and L'Equipe have no online English presence at all.
Click to expand...

What amused me was that they made a mistranslation in a quotation that was originally in English. Must have been translated to Spanish for their Spanish article and then the whole article translated into English, with little to no human supervision at that stage. I've also read about articles written by bots in whole cloth, although I think these usually have telltale characteristics and aren't as well stitched-together conceptually from paragraph to paragraph as what a human author produces.

I'm not familiar with Marca, but Google News brings me to all sorts of different publications
 
This is interesting. It seems that machine learning researchers don't fully understand how generative adversarial networks work. New research aims to shed light on the matter, and also develop a quantitative way of determining how similar 2 networks are. Even more interesting is that apparently networks become more similar, regardless of their initial architecture, as they become better trained to handle anomalous inputs, almost as if they're convergently evolving to a perfect form.

https://scitechdaily.com/new-method-exposes-how-artificial-intelligence-works/

“The artificial intelligence research community doesn’t necessarily have a complete understanding of what neural networks are doing; they give us good results, but we don’t know how or why,” said Haydn Jones, a researcher in the Advanced Research in Cyber Systems group at Los Alamos. “Our new method does a better job of comparing neural networks, which is a crucial step toward better understanding the mathematics behind AI.” ...

Neural networks are high-performance, but fragile. For instance, autonomous vehicles employ neural networks to recognize signs. They are quite adept at doing this in perfect circumstances. The neural network, however, may mistakenly detect a sign and never stop if there is even the slightest abnormality, like a sticker on a stop sign.

Therefore, in order to improve neural networks, researchers are searching for strategies to increase network robustness. One cutting-edge method involves “attacking” networks as they are being trained. The AI is trained to overlook abnormalities that researchers purposefully introduce. In essence, this technique, known as adversarial training, makes it more difficult to trick the networks.

In a surprising discovery, Jones and his collaborators from Los Alamos, Jacob Springer and Garrett Kenyon, as well as Jones’ mentor Juston Moore, applied their new network similarity metric to adversarially trained neural networks. They discovered that as the severity of the attack increases, adversarial training causes neural networks in the computer vision domain to converge to very similar data representations, regardless of network architecture.

“We found that when we train neural networks to be robust against adversarial attacks, they begin to do the same things,” Jones said.
Click to expand...
 
I was just reading a BBC football/soccer page, when I saw an ad about some documentary BBC did on deep fakes. I couldn't play it without a VPN, but read the heading that ninety percent of all deep fakes is porn. Which makes a lot of sense.
 
That reminds me of what I read today, that "computers are idiots," according to the comments section of this article: https://jalopnik.com/u-s-in-valley-o...hec-1849699276

Headline: U.S. in ‘Valley of Death’ as Autonomous Cars Write Checks the Tech Can’t Cash, Warns Pete Buttigieg

'"It feels like the widespread use of autonomous driving is seven years away, and it’s been seven years away for 10 years,” says U.S. Department of Transportation secretary Pete Buttigieg.'

From the comments section:

Tracer BulletOwen Bellwood
10/25/22 3:26pm

First you have to code for traffic.

Then you have to code for a stop sign.

Then you have to code for rain.

Then you have to code for traffic in the rain.

Then you have to code for a stop sign with traffic.

Then you have to code for a stop sign with traffic in the rain.

Then you have to code for a stop sign with funeral traffic.

Then you have to code for a stop sign with funeral traffic in one direction and the Oscar Meyer Wienermobile in the other.

Next you have to code for a stop sign with funeral traffic in the rain coming from one direction and the Oscar Meyer Wienermobile being chased by the dog-van from Dumb & Dumber in the other.

Then you have to code for a plane landing on the highway.

Then you have to code for a plane landing on the highway in funeral traffic as it begins to rain ... and the plane gets hit by a train.

...

All of these are simple for a human to figure out.

Computers are idiots.

33
Reply

RuivoTracer Bullet
10/26/22 2:29am

As someone that wrangles them for a living, I can attest that computers are indeed idiots.

4
Reply

FredTracer Bullet
10/26/22 11:06am

And you haven’t even introduced snow, ice or animals. Deer act erratically and sometimes they just look like a stump in the ditch before bolting.

1
Reply

Calli ArcaleTracer Bullet
10/26/22 11:49am

Computers are idiots.

This is very, very true. As a software engineer, there is nothing more humbling to me than trying to code a behavior into a system that has even a tiny fraction of the decision-making capability of a human. What is effortless for us to decide is enormously complicated to code.

I have a favorite Dr Who quote that I like to keep in mind while I’m coding:

“The problem with computers is they are very sophisticated idiots -- they do exactly what you tell them at amazing speed.”
Click to expand...
 
On the subject of driverless cars, Ford is shuttering its self-driving car development unit and writing off over a billion dollars. Tesla has been under investigation for over a year for making misleading claims about FSD and marketing an unsafe product. This is a reminder that the future is impossible to predict and that some problems of machine-learning are much harder to conquer than even experts predict.

“The team we have at Argo has been working on what I consider to be the hardest technical problem of our time," said Doug Field, who Ford hired away from Apple’s car project last year. “It's harder than putting a man on the moon.
Click to expand...

https://finance.yahoo.com/news/tesla...123015521.html
 
good news! the AI ChatGPT thinks AI won't replace us. https://chat.openai.com/chat

me: will videographers become obsolete because of AI

ChatGPT: It is unlikely that videographers will become obsolete because of AI in the foreseeable future. While AI technology has advanced rapidly in recent years, it is still not advanced enough to fully replicate the creative and artistic abilities of human videographers. AI algorithms are typically designed to perform specific tasks, such as identifying objects in a video or generating computer-generated graphics, but they are not yet capable of the kind of creative decision-making and storytelling that are essential to the work of videographers. Additionally, the field of videography requires a unique blend of technical skills, artistic vision, and interpersonal communication, which are difficult to replicate with AI. As a result, it is unlikely that AI will fully replace human videographers in the near future.
 
IMO, it's too early to obtain a better answer right now because machines need to exist out in society for this to be properly evaluated. By default, the answer is no for the foreseeable future because we don't have any non-human objects attempting to sling a camera besides maybe in testing facilities.

This is like asking if AI will replace athletes in football or soccer or basketball (no injuries, perhaps better/different entertainment), or my husband or wife as my lover. Maybe? We don't know yet.

If there were robots out and about on productions and they consistently created images with poor framing and could not efficiently communicate with people [about work] then we'd have the field research to say, "Yeah, this isn't going to work."

If we're asking if there will be anything that will be able to move like a primate while intelligently operating a camera, what's the answer? Probably not, but in what context? And maybe the world once again changes, adapts to "good enough" and "different".

If you rephrase the question and ask if AI could replace certain static camera operators, yes. Yes, yes, yes. If we're being real with ourselves, we understand there are quite a few videographer positions which consist of very minimal operation and could easily be replaced by machines with cameras/lenses using face and object/chip tracking (like in a news studio which will not require people behind cameras for much longer).

Currently in life situations where you have to consistently move and react with a piece of equipment (tilt here, pan here, follow this person, follow that person) there is nothing else in the world attempting this kind of activity (without human help).
 
True and objective NorBro. I was looking around a few years ago on the web for work opportunities and was checking out some local hockey organizations when I came across Pixellot. This is a 4k camera that is taking over mid to low level hockey game film. The kicker is that it also does the player stats at the same time. It even zoomed in on the players after they scored a goal. On balance I would say it was more than good enough from a movement point of view given its 4k resolution for a 1080 product. Changed my view right away because I would be calling those folks to offer my services. Why would they use me now? I don't keep the stats which they are probably more concerned with. The change is well under way and we are not going back.

I did get a soccer season least year in place of a Pixellot as the folks were not happy with it.
 
Hopefully you can do it for another 10 years or whatever before that sports filming becomes fully automated because that kind of work is a great candidate for AI/machine filming. Besides the chips inside balls, it's the high-resolution that's really going to make the difference.

In 10-20 years, a few 16K cameras on the field stitching the feed into a final product by AI (or not, a human could take a look at all the angles manually if desired) would provide all of the coverage you could ever need, even at the high school level since it won't be too expensive.

Add a "foot soldier" for a different perspective or two, which could be a metal foot solider with wheels (robot), and you're good.
 
AI will help a lot of photo-video editors, especially on the repetitive or semi-repetitive clips, with a stable POV and/or camera positioning.

I am not keen on everything from the ongoing World Cup coverage because the directors zoom out so far, they end up with 1/4th to 1/3rd of the frame showing the opposite stands. But I bet it looks fine on an 8K screen off an 8K camera. And so you take really wide shots from several angles and zoom in during the instant replay by following the ball. And, much like Messi's PK miss, it's almost automatic.
 
There is a company called NFHS that has gone country wide and installed these things along with their own YouTube like network. I have seen a few and in general it seems to be received with a low grade golf clap. Promises of all this stuff and the end product being ultra wide, impersonal video. So there is a quality standard people want. But organizations big and small can't resist getting weak knee'd about a simple solution that solves all of their problems... I have no idea what they cost.
 
You must log in or register to reply here.
Back
Top