Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Facebook Discovers Fake AI-Generated Profiles

nytimes.com

Post reply on HN
Print view View original

31–40 of 219 posts

Re: Facebook Discovers Fake AI-Generated Profiles

#31

Post by pushpop »

Barrin92 wrote: post #23
MuffinFlavored wrote: post #15

Earlier quoted context omitted.

Require manual profile validation for new Facebook accounts? Probably would get downvoted if I said SSN + a credit check. Phone number required + needs to be validated + can not be reused + can not be Twilio. Prove your address? Prove your ISP cable bill? I don't know. Everything I guess, people here are going to tell me it's an invasion of privacy and they don't trust Facebook to do it. But they'll let CreditKarma h…

or simply charge a few bucks for account creation.

You then run the risk of excluding developing economies. Unless you adjust the payment to be region specific, in which case you’d just get foreign powers (be that corporations or state sponsored) using their wealth to fund hordes of new registrations from said developing economies; thus negating the benefit of a sign up fee.

Re: Facebook Discovers Fake AI-Generated Profiles

#32

Post by scarejunba »

The most fascinating thing to watch is the democratization of propaganda. Now you no longer need the printing press. You can speak with a million voices without being a Getty or Rockefeller.

The NYT no longer has a monopoly on convincing lies.

These are interesting times. I think we're close to the emergence of trustworthy celebrity journalists.

Re: Facebook Discovers Fake AI-Generated Profiles

#34

Post by scarejunba »

joeblubaugh wrote: post #3

“Discovers” People have been talking about this group for months and Facebook hemmed and hawed and delayed and delayed

People have also been talking about every other group. The sensitivity of "people talking" may be 100% but the specificity is near zero.

Re: Facebook Discovers Fake AI-Generated Profiles

#35

Post by Barrin92 »

pushpop wrote: post #31
Barrin92 wrote: post #23

Earlier quoted context omitted.

or simply charge a few bucks for account creation.

You then run the risk of excluding developing economies. Unless you adjust the payment to be region specific, in which case you’d just get foreign powers (be that corporations or state sponsored) using their wealth to fund hordes of new registrations from said developing economies; thus negating the benefit of a sign up fee.

a low single-digit dollar amount of money isn't really prohibitive any more in almost all regions on the planet for a one-off payment (networked regions, at least). That's an argument facebook itself likes to make but if you look at the WhatsApp growth which used to have a model of 1$/year it seems completely fabricated.

And arguably a small payment amplifies anti-abuse measures because you have a hard financial trail. If a foreign entity buys thousands or more of accounts, not only are you putting a cost on it, it should be trivial to see the anomaly in the system.

Re: Facebook Discovers Fake AI-Generated Profiles

#36

Post by mr_toad »

rvz wrote: post #5

And so it appears that the inevitable has 'finally happened'. Mark Zuckerberg's website is now a place not to be trusted as a source of news, finding friends and meeting others if there's a possibility of AI generated profiles being proliferated on the website. I've heard that meeting real humans outdoors makes it less likely to be deceived. Their authenticity is vouched via word of mouth and experience by other huma…

> a place not to be trusted as a source of news, finding friends and meeting others

Sounds like Usenet in the 90’s. The more things change...

Re: Facebook Discovers Fake AI-Generated Profiles

#37

Post by prvc »

jedberg wrote: post #4

It's ironic that as we move into a world where anyone can be a "journalist" and anyone can have a voice, we need professional journalists more than ever. Now a journalist will not only need training in writing well, conveying a story with facts, and investigating, but they will also need to become experts in identifying fake source material. In a world where anyone can make a video or photo of anyone doing anything,…

Professional journalists have abdicated that role, and can no longer said to be "reputable". Plausible deniability that they were working in good faith was irrevocably lost when they failed to adapt in response to the valid criticisms they were receiving on social media, not because "fake news" trolls took over. Twitter did indeed kill the newspaper star, just not in the way they think.

Re: Facebook Discovers Fake AI-Generated Profiles

#38

Post by mmagic »

If anyone's interested in neural networks and how they work, you should check out victoR. VictoR is a powerful tool which utilizes file storage size to modify specific neuron mapping components and optimize processing power. It's still quite experimental, but here is a download link for an installation wizard if you want to check it out: https://bit.ly/35mtZZj

Re: Facebook Discovers Fake AI-Generated Profiles

#39

Post by avivo »

FYI, I actually just published a guide on how you can help ensure that ("deepfake") tools that you make are less likely to end up misused like this: https://www.technologyreview.com/s/614912/ethical-deepfake-t ...

(Also at /newest; 3rd page now, and the other submission. Ideally those topics at least would be merged?)

Summary of mitigations:

# Limiting users or use cases

- User Vetting: Carefully vet those who can use a tool (Example: Synthesia only working with vetted enterprise clients.)

- Usage Constraints: Limit the synthesis possibilities. (Example: Humen is providing only a limited set of movements for generated videos.)

# Discouraging malicious use of flexible and public tools

- Clear disclosure: Ensuring and requiring that manipulated outputs are clearly labeled (including via metadata).

- Consent protection: Requiring the consent of those being impersonated.

- Detection friendliness: Supporting those working on detection and not seeking to fool detection systems.

- Hidden watermarks: Embedding data through robust watermarks.

- Usage logs: Storing information about usage and media outputs.

- Use restrictions: Enforcing contracts that prohibit forms of violative or malicious use.

Not all apply to every system of course. The article has more detail!

Re: Facebook Discovers Fake AI-Generated Profiles

#40

Post by phkahler »

nsomaru wrote: post #7
dv_dt wrote: post #6

Earlier quoted context omitted.

In part, imho, the distrust comes from the the standards of the highest tiers of professional journalists being on a low right now too.

Journalists are following clicks instead of the other way around

[flagged]
Post reply on HN