4 ms·

>Trust in what we see is increasingly important in a world where image editing has become ubiquitous – fake content is a serious and increasingly pressing issue

>Trust in what we see is increasingly important in a world where image editing has become ubiquitous – fake content is a serious and increasingly pressing issue. I keep on reading such statements, but still don't feel like this is a real (increasing) problem. Pictures could be faked is a believable way 100 years ago already, and I cannot recall a single instance where new, fancy AI-driven fakes were any more of an issue than classic photoshopping. In my opinion, it's all just FUD.

tachyonbeam 7y ago
100 years ago you would have needed an expert, specialized equipment and several hours. 10 years ago, you still needed an expert and probably an hour at least. Soon, there may be tools anyone can use, feed in some photos of you and a description of what you want to see. It will be easy to generate fake profile pictures that look like you, for instance. Your voice will be easily cloned on the phone as well, and eventually, even video chats could be fake. Scammers could find out when you're on a trip from social media, reach out to your family, ask for emergency money because of a stolen wallet, etc. Still some effort for the scammers, but eventually, you might be able to automate the whole scam with a bot. The bot doesn't need to be that smart, it just needs to follow a script with a handful of variants, like a preplanned dialogue tree.
- blauditore 7y ago
  This is something that could be done before as well by just impersonating someone. The voice just needs to sound believably similar, which is not too hard with a sufficiently bad connection. And yet such attacks don't seem common at all.
  - e12e 7y ago
    Sure, you can impersonate someone (or even anyone). A botnet can impersonate everyone, and make robocalls to everyone. If you're doing a targeted attack, you might need a significant chance of success - if you can target 10 000 users, you might get away with a 1% success rate, and still make a significant amount (say 60 times 1 000 USD in "emergency" funds).
mc32 7y ago
I think the issue is scale. Keeping track of one doctored image with a missing Yezhov next to Stalin is easy. Keeping track of myriads and millions of doctored images is harder. Even if you have a hash database or thumbprint database to keep track of all of them, someone is going to think they’re miscategorized or attribute the classification to politics or soft censorship. etc.
michaelbuckbee 7y ago
A common way that bots are detected on twitter is that they re-use other users' avatars. This is pretty quickly being replaced by https://thispersondoesnotexist.com/ https://thispersondoesnotexist.com/ (and others).

tachyonbeam 7y ago

100 years ago you would have needed an expert, specialized equipment and several hours. 10 years ago, you still needed an expert and probably an hour at least. Soon, there may be tools anyone can use, feed in some photos of you and a description of what you want to see. It will be easy to generate fake profile pictures that look like you, for instance. Your voice will be easily cloned on the phone as well, and eventually, even video chats could be fake. Scammers could find out when you're on a trip from social media, reach out to your family, ask for emergency money because of a stolen wallet, etc. Still some effort for the scammers, but eventually, you might be able to automate the whole scam with a bot. The bot doesn't need to be that smart, it just needs to follow a script with a handful of variants, like a preplanned dialogue tree.

blauditore 7y ago
This is something that could be done before as well by just impersonating someone. The voice just needs to sound believably similar, which is not too hard with a sufficiently bad connection. And yet such attacks don't seem common at all.
- e12e 7y ago
  Sure, you can impersonate someone (or even anyone). A botnet can impersonate everyone, and make robocalls to everyone. If you're doing a targeted attack, you might need a significant chance of success - if you can target 10 000 users, you might get away with a 1% success rate, and still make a significant amount (say 60 times 1 000 USD in "emergency" funds).

mc32 7y ago

I think the issue is scale. Keeping track of one doctored image with a missing Yezhov next to Stalin is easy. Keeping track of myriads and millions of doctored images is harder. Even if you have a hash database or thumbprint database to keep track of all of them, someone is going to think they’re miscategorized or attribute the classification to politics or soft censorship. etc.

michaelbuckbee 7y ago

A common way that bots are detected on twitter is that they re-use other users' avatars. This is pretty quickly being replaced by https://thispersondoesnotexist.com/ https://thispersondoesnotexist.com/ (and others).