Encrypt your images - adversarial

colak · 2021-01-05 15:13:01

For those up[loading family photos on the web, adversarial is a webapp that alters images in order to make them machine-unreadable, while leaving them visually almost indistinguishable from the original for the human eye. I know that it is not really an encryption, but a step towards the right direction anyway.

Bloke · 2021-01-05 15:52:27

Clever!

colak · 2021-01-05 16:39:23

Bloke wrote #328049:

Clever!

Indeed. Could this become a txp plugin some time not in the near future or is it too complex?

marios2 · 2021-02-20 18:35:16

It is basically deleting the meta data from the exif file info of the image, if that is a jpeg.

rgrds , marios

colak · 2021-02-21 06:55:20

marios2 wrote #328994:

It is basically deleting the meta data from the exif file info of the image, if that is a jpeg.

rgrds , marios

Adversarial does more than that. It confuses the AI into not recognising the content of the photo.

gaekwad · 2021-02-21 10:10:22

colak wrote #328051:

Indeed. Could this become a txp plugin some time not in the near future or is it too complex?

As I understand it – and I am calling for anyone with corrections and additions to jump in – image processing on PHP-based web applications like Textpattern need to have either PHP-powered code to do the work or use system utilities to do the same.

The dev team are intending to implement better image handling using Image Intervention, which fits into the “PHP-powered code” option above. Image processing can sometimes be very intensive work for a processor, and it’s arguably important to not be a “noisy neighbour” on web hosting with excessive resource use.

My gut feeling would be this is the job of a local application until the tech evolves / matures, rather than a plugin…whether or not it becomes sufficiently widespread enough to be adopted by web apps, only time will tell!

Edit: I submitted Adversarial to Hacker News for discussion and it gave me a few hundred fake internet points while I was asleep, which was a nice surprise. Good find, Yiannis!

Last edited by gaekwad (2021-02-21 10:11:44)

colak · 2021-02-21 15:49:15

gaekwad wrote #329000:

I submitted Adversarial to Hacker News for discussion and it gave me a few hundred fake internet points while I was asleep, which was a nice surprise. Good find, Yiannis!

I since spoke with one of the developers who admitted that they created the app as a proof of concept and since added to their site that it only works against Google Inception V.3. Our discussion also revealed that their aim was not to find a solution against machine recognition but to bring that to the forefront of discussion. The actual solution would be to make such machine learning software illegal or at least limited in their scope.

Your discussion in Hacker news is indeed very interesting. I’ll send it to him, if it is OK by you.

>Edited to add: Further to the discussion above, here’s a short Twitter chat

Last edited by colak (2021-02-21 16:50:42)

michaelkpate · 2021-02-22 01:01:01

colak wrote #329001:

The actual solution would be to make such machine learning software illegal or at least limited in their scope.

If machine learning software is outlawed, only outlaws will have machine learning software.

Destry · 2021-02-22 08:45:36

This reminds me of the ‘fight climate change caused by technology with more technology’ downward spiral (because capitalism), instead of not doing the problematic things to begin with.

In this case, don’t put photographs of your loved ones online. It’s a no-brainer. AI recognition is the least of the worries here.

Send print photos of the family to grandma by snail mail instead. Strangers do not need to see photos of your loved ones, period, and only the sickos will want to anyway.

Sounds like the creator is on the right track, though, to show that policy is needed against such blatant use of AI ability.

But, as always, the real measure of protection comes from the individual not putting themselves and others at risk to begin with.

gaekwad · 2021-02-22 09:07:33

colak wrote #329001:

Your discussion in Hacker news is indeed very interesting. I’ll send it to him, if it is OK by you.

Of course – I simply submitted for discussion.

colak · 2021-02-22 15:39:30

gaekwad wrote #329015:

Of course – I simply submitted for discussion.

From Francis (This is for Pete)

… thanks a lot for channeling it to ycombinator. We’ve got a bunch of offers for participating in the project through this!

gaekwad · 2021-02-22 15:59:24

Ah, the power of the network! Fantastic.

colak · 2021-02-22 17:37:13

Destry wrote #329014:

… But, as always, the real measure of protection comes from the individual not putting themselves and others at risk to begin with.

michaelkpate wrote #329009:

If machine learning software is outlawed, only outlaws will have machine learning software.

The problem of course is not just with facial recognition. Many Captchas we have to solve feed the AIs which are developed for military purposes such as non-human operated weapons. The time is very near when drones will no longer have their ‘pilots,’ just their missions. The ethical questions about this are enormous. Two of the key operational legal and human restrictions during war are those of choice and responsibility. Machines cannot be held responsible for any collateral victims, nor will they have the awareness not to carry out their mission because of a matter of conscience.

As such I am wondering about how the legality of such systems will help anyone, except maybe, some people whose jobs will become obsolete and those who currently have them will be able to sleep better at night. Having said that, I believe that the ethical and critical use of AI, in medicine, disaster prevention, and the humanities could be of a great help to us.

Destry · 2021-03-02 15:04:02

‘Finally! A way to make old family photos even creepier’ Deep Nostalgia

Eek!

Yipes!

Reminds me of the pictures on the walls in Hogwarts. Kind of cool… but could also be abused badly.

Also reminds me of a audio technology that was announced a few years ago, where it was possible to take a recording of someone talking, say a political speech, and edit the words in the audio to make it sound like the person was saying something completely different.

Maybe these things can be combined. Say a person plucks your photo off the web, animates it, makes some damaging words come out of your mouth, links the image to your website. Probably already being done.

Come to think of it, there’s a French comedian doing this now on French TV. Canteloup is his name. He takes political figures like Macron, animates their faces using some AI tech (his tech crew does this, I suppose, not him exactly), and literally speaks through them like avatars with his satire. It’s really weird. Not that old trick of cutting out the mouth on static image and inserting comedian’s lips, either, but real video-looking mimicry. I can only image how the Chinese president would react to that.

michaelkpate · 2021-03-02 18:22:15

Destry wrote #329085:

‘Finally! A way to make old family photos even creepier’ Deep Nostalgia

I decided to give it a try. It seems to work better with posed studio photos so I had this one of my grandmother.

It looks very robotic – kind of a Walt Disney World Hall of the Presidents.

Textpattern CMS support forum

#1 2021-01-05 15:13:01