• Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
Oct 25, 2017
1,103
Konoha
Deep fakes are easily noticeable and no one is going to deep fake someones face to a home movie off pornhub because iirc you need a very clear video to do it. Also what's the difference between this and a really good photoshop? If photoshops "can't" be used like this then deepfakes can't either. "Obviously this video of my buddys wife getting fucked hardcore style in what looks like a studio level porn video is very real and not fake at all"
 

GK86

Member
Oct 25, 2017
18,738
I recommend you read this: https://www.masstlc.org/the-more-machines-learn-the-less-we-understand-them-for-now/

The problem with machine learning is that sometimes it works like a black box. Data goes in, data comes out, you can't explain that. The machine can produce results. Humans can manually confirm the results, but the machine can't tell the human about the process it used to arrive at the result. An example of what this looks like in practice.

After AlphaGo made the news in 2016 for repeatedly trouncing the world's top Go players, we knew it was a superior Go "player", probably the best in the world. However, what we can't do is ask AlphaGo to tell us how it wins nor can we ask it to teach us its strategies because it doesn't have a concept of "strategy". It does a thing, if it works, it does the thing again. If it doesn't, it tries another thing.

The only person who can tell whether the machine is doing a good job is the technician overseeing its data training. This does not mean that the trainer has control over the machine's process. The trainer only confirms or rejects results according to their desires and goals. This is where the danger of deepfake really is, because it's ultimately the trainer's moral compass that decides what results get accepted and which ones get rejected. The machine's capabilities are already beyond what humans are capable of, but human morality guides what the machine produces.




Thank you. That is crazy.
 

PogiJones

Banned
Oct 27, 2017
1,636
Hm.

What other post did I specifically reference, Benita?

Post history = cross thread drama.

You maligned a guy for his post history. Benita brings up that fact, and you suddenly say, "Don't bring up post history cross thread drama."

Then, minutes later in this same thread, you post a .gif that says "True to form." to malign Benita. I.e. "It fits his post history."

Only you can bring up post histories, right?
 

Wein Cruz

Banned
Oct 27, 2017
5,772
Post history = cross thread drama.

You maligned a guy for his post history. Benita brings up that fact, and you suddenly say, "Don't bring up post history cross thread drama."

Then, minutes later in this same thread, you post a .gif that says "True to form." to malign Benita. I.e. "It fits his post history."

Only you can bring up post histories, right?

Looking forward to seeing RedMercury twist their way out of this one.
 

samoyed

Banned
Oct 26, 2017
15,191
Thank you. That is crazy.
It really is.

I lose my mind every time I encounter another thread about deepfakes. On the list of "existential crises I expect to see in my lifetime", Deepfakes are pretty far below "climate change", "ethnonationalism" and "automation of labor", but when I'm reminded about them I start panicking like the sky is falling. It's just so overwhelming.
 

Musubi

Unshakable Resolve - Prophet of Truth
Banned
Oct 25, 2017
23,611
I kind of want to see some side by sides of the real thing and this deepfake porn. Like are these deepfake videos that convincing? Or do they fall apart under a moment of scrutiny? I get why people are rightfully upset and fuck people who are using this to abuse others, but this news is just blowing my mind.

It depends. Some of them absolutely can be others you can clearly see that its a fake. Its probably convincing enough especially to people who don't have a trained eye on what to look for.
 

samoyed

Banned
Oct 26, 2017
15,191
Deep fakes are easily noticeable and no one is going to deep fake someones face to a home movie off pornhub because iirc you need a very clear video to do it. Also what's the difference between this and a really good photoshop? If photoshops "can't" be used like this then deepfakes can't either. "Obviously this video of my buddys wife getting fucked hardcore style in what looks like a studio level porn video is very real and not fake at all"
You can make a video featuring Obama talking about jihad, sharia law, and beheading Christians. With his face and with his voice. Think about what this would've done in 2008.
 

Rendering...

Member
Oct 30, 2017
19,089
As someone who literally spent the last 2 days rotoscoping for a freelance client, this type of technology would be life changing from a process perspective. But I haven't met a single editor or designer in my field who thinks this would be worth the tradeoff for public good. It's absolutely crazy. All of us know where this is heading. It's like watching a train wreck as it becomes easier and easier to use by anyone with basic editing software skills.
Society has needed a new ethical paradigm since, at latest, the rise of social media. We're almost entirely unprepared for a world where anyone with a device can falsify reality. The public has enough trouble distinguishing fact from fiction when events are recorded accurately.

Career-ruining life-destroying harassment catalyzed by out-of-context tweets, decade-old online postings, or petty political grievances is already far too common and within the means of almost anyone to instigate.

We'll need strict regulations and new laws to discourage companies and individuals from exploiting tech that can essentially turn anyone into a digital puppet. Maybe more important, our norms will have to evolve with our technology to account for this new reality. It could take decades of scandal and unjustified outrage to reach the tipping point where the jury of public opinion favors evidence over convincing illusions.
 

Taki

Attempt to circumvent a ban with an alt account
Banned
Oct 25, 2017
5,308
This is going to hit youtube vloggers hard.
 

CrocM

Member
Oct 25, 2017
9,566
Let's remove the stigma around pornography alongside detecting these fakes as a two prong strategy.

There's no shame in sex work!!
 

RedMercury

Blue Venus
Member
Dec 24, 2017
17,637
Post history = cross thread drama.
No, but you're going to interpret my words to suit your perspective no matter what I say, so eh
Only you can bring up post histories, right?
Anybody can do anything lol, I'm not in charge of resetera, I'm just a poster. I'm also free to ask someone not to do something.
Looking forward to seeing RedMercury twist their way out of this one.
No twisting needed, but I love that you're so looking forward to it.
 

Dongs Macabre

Member
Oct 26, 2017
2,284
I recommend you read this: https://www.masstlc.org/the-more-machines-learn-the-less-we-understand-them-for-now/

The problem with machine learning is that sometimes it works like a black box. Data goes in, data comes out, you can't explain that. The machine can produce results. Humans can manually confirm the results, but the machine can't tell the human about the process it used to arrive at the result. An example of what this looks like in practice.

After AlphaGo made the news in 2016 for repeatedly trouncing the world's top Go players, we knew it was a superior Go "player", probably the best in the world. However, what we can't do is ask AlphaGo to tell us how it wins nor can we ask it to teach us its strategies because it doesn't have a concept of "strategy". It does a thing, if it works, it does the thing again. If it doesn't, it tries another thing.

The only person who can tell whether the machine is doing a good job is the technician overseeing its data training. This does not mean that the trainer has control over the machine's process. The trainer only confirms or rejects results according to their desires and goals. This is where the danger of deepfake really is, because it's ultimately the trainer's moral compass that decides what results get accepted and which ones get rejected. The machine's capabilities are already beyond what humans are capable of, but human morality guides what the machine produces.




I'm talking about designing AI to solve a problem that even humans can't solve. There's a huge difference between AlphaGo and a hypothetical AI that can perfectly edit an image so that any changes are undetectable. With AlphaGo, they had tons of training data available (Wikipedia says they started with a database of 30 million moves). Without actual examples of perfectly edited images (under the assumption that the previous poster was correct that a perfectly doctored image does not exist), you would likely have to work backwards from real images and insert artifacts, which is not going to be the same as going from a doctored image to a perfect image. Now, a bit of possibly incorrect speculation on my part: I'm thinking this is would then be similar to correcting compression artifacts on image, which could never be done perfectly since lost information is lost information.

You can't go from A to B without knowing what B (a perfectly doctored image) looks like. The inner workings being a black box is irrelevant when you don't know the inputs and outputs.
 

samoyed

Banned
Oct 26, 2017
15,191
Oh I see what you're saying. Yes I agree with that kind of. Although I took Conal's post to mean that because humans were doing the photoshop, and since humans are error prone, photoshop never spelled the disaster this would. Human photo editing can't beat human analysis but ML deepfakes could feasibly beat human video analysis.

Anyway it doesn't matter. Regardless how much scientists and experts scream "no this is wrong, you're wrong, this is not how it works, this is not real" people will believe what they want to believe. The damage being done will be societal, not judicial or criminal.
 

Psykodoughboy

Banned
Dec 17, 2017
782
Is the only way to stop this is too regulate this software or machine learning? Because this is just the beginning.
 

ZackieChan

Banned
Oct 27, 2017
8,056
it really depends on how many pics were used for ML. the more images used at different angles in different lighting the better the result. there was a thread here about it earlier in the year and some of the results were very convincing.

Edit: yup deepfakes
For most people there probably aren't enough similar photos of different angles to properly train it, from what I read back when this came out. Someone like Obama has tons of photos and videos out there that can be used. So this seems to be more of a famous people issue.
What this means is it couldn't be made a crime but it is also likely a civil infringement if it's used with hostile intent.
Yep, this is covered by defamation, specifically defamation per se, which deals with defamation regarding sexual activity and is easier to prove.
This is going to hit youtube vloggers hard.
Yeah, there is a ton of footage of them that makes training the AI easy, I'd imagine.
Gonna be starting a YouTube channel soon - don't make deepfakes of me brehs
 

Deleted member 8861

User requested account closure
Banned
Oct 26, 2017
10,564
This has been a thing for almost a year now IIRC, and that doesn't make it any less maddening and disgusting.

It must be criminalized immediately and people who do this shit deserve jail time and losing unsupervised access to the Internet
 

Jktpnymonorel

Banned
Jan 19, 2018
490
Is the only way to stop this is too regulate this software or machine learning? Because this is just the beginning.

kinda hard, advances in this topic are always faster than their regulation

edit : 1 solution I can think of is for porn producer to use some kind of floating watermarks (not idle) for their video.....
but I still think machine will be smart enough to even remove the watermarks
 

Diablos

has a title.
Member
Oct 25, 2017
14,568
This is going to be a significant contributor to ending the Internet as we know it I'm afraid. This is a huge Pandora's Box. As more and more people fall victim to this, ISP's and Governments will use it as justification for putting in place severe restrictions on what you can do online. And for the first time they have more than just corporate greed and regulation serving interests that aren't yours as a reason or an excuse. This really is a huge problem.

In 5–10 years everything will be different.
 
Last edited:

JayC3

bork bork
Administrator
Oct 25, 2017
3,857
Official Staff Communication
Stop the hostility and get back on topic.
 

Sulik2

Banned
Oct 27, 2017
8,168
This is just the tip of the iceberg. Perfect digital and audio doubling, which is freaky as hell how good it already is, is going to make the idea of evidence and reality meaningless in the next ten to fifteen years. Porn won't matter when your can just make a photorealistic video of your politic opponent saying anything you want and it's impossible to tell. Digital reality is going to be whatever people make and actual evidence is going to get real hard to come by.
 

Sulik2

Banned
Oct 27, 2017
8,168
I really don't understand how deep fakes hasn't been banned. Shit is super scary.

It's not real is the issue. Criminalizing fake imagery gets really close to criminalizing thoughts and that's a very dangerous slippery slope. Properly prosecuting libel or slander and applying it to fakes might make more sense.
 
Jan 29, 2018
9,383
Even if there's a foolproof way to prove one of these is a fake, the damage will be done. The past few years alone have provided plenty of evidence that people will gladly ignore proof that what they believe is a lie.
 

DJ_Lae

Member
Oct 27, 2017
6,856
Edmonton
The real question is what will Putin's o-face be when someone makes a deepfake video of Trump blowing him?

6dT7Pd0.gif
 

Deleted member 41638

User requested account closure
Banned
Apr 3, 2018
1,164
...gtfo with this. So wildly not at issue, and never at issue. You're really gonna stroll into a thread about people creating content that attempts to destroy the lives of others to say WAIT IT ISN'T ALL MEN! NOT ALL MEN! like that's the problem here?

Attempts? Man I actually killed millions of my own potential offspring to those ScarJo deepfakes.
 

Lady Catherine de Bourgh

Teyvat Traveler
Member
Oct 27, 2017
832
Deep fakes are easily noticeable and no one is going to deep fake someones face to a home movie off pornhub because iirc you need a very clear video to do it. Also what's the difference between this and a really good photoshop? If photoshops "can't" be used like this then deepfakes can't either. "Obviously this video of my buddys wife getting fucked hardcore style in what looks like a studio level porn video is very real and not fake at all"

Maybe you're right in comparing this to photoshop. The difference in this point in time is that it is not yet common knowledge that videos can be fabricated this good. Before it is proven to be fake it can be a real pain to deal with. I also feel that it matters who the target is. It will have a different impact for a fifteen year old to be targeted.
 

TSM

Member
Oct 27, 2017
5,821
While the fact that this can be used to generate fake porn is unfortunate, it's affects will be nothing compared to the ability to make anyone say anything on video. People are already convinced just by stories of things allegedly done that never happened. Soon there will be fabricated video evidence of all sorts of things. "What do you mean he didn't do it? I saw him do it with my own eyes."
 

CommodoreKong

Member
Oct 25, 2017
7,690
I fail to see how this can be protected under free speech of any sort. This is flat out lying, deceit and abuse.

I might be wrong but I think it's only illegal in the US to use someone's likeness without their permission for commercial use, if its non commercial I think it's okay (someone correct me if I'm wrong).
Of course pornhub hosting a video and making money off ads would be considered commercial use.
 

Vuze

Member
Oct 25, 2017
4,186
Deepfakes need to gain the same status as photoshopped photos. Even the layman knows about still image manipulation and could (if they wanted) question a photo based on that. I don't think most people realize/know you can easily manipulate and generate video akin to that.