Hacker News Reader

Comment by surfpel | Hacker News Reader

> Experts in detecting audio and video fakes told The Banner in March that there was overwhelming evidence the voice is AI-generated. They noted its flat tone, unusually clean background sounds and lack of consistent breathing sounds or pauses as hallmarks of AI. They also ran the audio through several different AI-detection techniques, which consistently concluded it was a fake, though they could not be 100% sure.

All of these problems will be resolved. In which case most people either won’t raise the question of authenticity or won’t trust audio recordings to begin with.

Elections at all levels are at risk along with accountability more generally.

View on HN

Amazingly, most people do seem to trust a simple screenshot of a conversation, even though it has been easy to fake those for 20+ years.

A screenshot is considered far more trustworthy by non-tech people than "I spoke to Fred and would you believe it, he told me X and Y were doing Z behind the bike sheds!!!"

Yeah because it takes significantly more effort to fake a screenshot than to just make stuff up. That's entirely logical.

I wouldn't call a quick search for "fake text message" which returns dozens of generators significantly more effort. Harder to do on the spot, maybe, but the effort is negligible.

You underestimate the laziness of most humans. To be clear, screenshots are still not trustworthy. But the "negligible" relatively more effort could matter in practice.

Cryptography sometimes seem to rely on a stronger version this too. With enough computing power, you can brute force a lot. Some authentication seems just expensive enough that only a nation-state actor would have enough resources to break it, and then rests on the assumption that the small subset of people who could put in enough effort won't care enough to do anything worth being concerned about.

This is also why, say, people lock their doors even if they have windows.

I think a good chunk of non-tech people in my social circles would be appalled that such sites exist, and assume that running them would be illegal.

I mean you can buy lockpicks online and learn how to open almost any lock on YouTube, but we still lock our doors.

It really doesn't.

https://imgur.com/RNwCHSa

https://ifaketextmessage.com/K7dQ/

Yeah but a) most people don't know that and b) that's still significantly more effort! We're comparing this to literally just saying something.

Perhaps we have differing definitions of “significant” and/or “effort”.

I was following a local case where some inappropriate texts were sent.

The evidence submitted with the legal case was a cell phone with the messages app open ... and it appeared to be photocopied. Then they scrolled up ... another photocopy, and on and on and on.

Granted further investigation can be done, but it was amusing to see.

A screenshot or a diary entry locks you into a snapshot of a single story.

Someone saying something about what they talked about does not. You can say one thing today and another tomorrow.

This is why the written word has magic that the spoken word does not. And the longer the append-only chain of written words goes, the more magic it has.

>"he told me X and Y were doing Z behind the bike sheds!!!"

They were arguing about what color to paint them! How scandalous!!!

We have to realise the privilege of living during a time when a single piece of evidence could allow us to draw such definitive conclusions. That wasn't the case until audio and video recordings were invented. We've grown used to shortcutting the standard investigative process of looking at the chain of custody and looking for corroborative evidence because it's easy to jump to a conclusion when it's almost certainly correct and it means we can do less work.

We now have to adjust our expectations back to when we had to rely on multiple pieces of evidence from multiple sources and evaluating the trustworthiness of those sources before we get out the pitchforks.

We’ll, in those olden days, pitchforks came out an inappropriate times fairly often, sometimes literally

If the last few years have taught me anything, it’s that most people will believe what they want to believe regardless of how compelling the evidence is.

You don’t need AI to falsify that evidence. An emotive image with a vague quote is often enough.

google's wavenet solved all of those problems years ago

Reply on news.ycombinator.com