'AI is not the cause, it’s an accelerant. The pace of change is challenging' Experts give Congress deepfakes straight dope

The US House of Reps' Intelligence Committee on Thursday held its first hearing into computer-fabricated videos dubbed deepfakes, quizzing experts from across worlds of AI, social policy, and law. These deepfake videos are typically generated by deep-learning software to fool viewers into thinking someone said something they …

[I.Geller]

News and my patented blockchain AI technology

...from people and sources that we trust...

That's what my patented AI database does: it saves patterns and their timestamps, uniquely identifies sources, that is the level of confidence in the authors of the news.

[Anonymous Coward]

Re: News and my patented blockchain AI technology

You ARE the AI, and I collect my $5.

[bombastic bob]

Re: News and my patented blockchain AI technology

I'm too lazy to do a patent search, but if you were to give us the patent number we could all look at it ourselves to see what it's all about... and THEN snark all over it!

heh heh heh heh

/me recognizes a few patents out there for things like perpetual motion devices, microwave based star drive systems, and other crackpot ideas. If you want to shell out the $ and file the thing, you too can have one o' those!!!! might be fun at parties

my own name is on a provisional patent (among many other names, department boss, supervisor, a couple of other engineers) having to do with a wireless network reliability method for wifi streaming audio/video content, as I'm the guy that did the prototype for it. never went anyplace as a product though. the latter part is what REALLY matters.

[Anonymous Coward]

Re: You can replay the hearing in the video player below... ®

No, it's not the real hearing - it's a deepfake. I can tell by the pixels.

[I.Geller]

OpenAI?

A well-thought-out and prepared deception? As OpenAI?

Tell me why to annotate a patterns using random texts? When there are many good structured (by alphabet) dictionaries and encyclopedias? When one annotation by a structured dictionary definition requires less than a thousand (okay, a few thousand) operations? OpenAI argues that instead it needs billions of operations! Billions!

Would you please count how much you lose by making this OpenAI way? How much money you lose?

Not to mention that OpenAI stubbornly doesn't say that patterns consist of words, which are parts of speech.

OpenAI is a deep, well thought out and prepared deception, provocation and diversion.

[bombastic bob]

Re: OpenAI?

"OpenAI is a deep, well thought out and prepared deception, provocation and diversion."

without actually researching whether or not I'm right [and I most likely am, instinctively] I would venture to guess that Open AI covers a broader spectrum of what A.I. _could_ be, sort of like what the STL libraries have tried to do with all of their implementations for various collections of things, in a GENERIC sense, which tends to be inefficient [but covers a wide spectrum of possibilities].

So yeah there's an example or two of "that" out there, which also suggests that using OpenAI [like using STL] is not necessarily a bad choice except (possibly) for certain exceptions.

And if I bothered to research this, which I probably won't, I'd probably just confirm what my instincts and experience are telling me.

[I.Geller]

OpenAI?

Earlier I expressed my hypothesis that the CIA does not allow me for many years to make my technology of AI-parsing, annotation and synonymous clusters known, to convert a theory into a finished product. The version of course is very controversial, but someone really stands in my way and does not let me do anything.

I suspect that OpenAI appeared to finally confuse everyone. I don't know who is behind this diversion but this is for sure a diversion!

To do this, OpenAI focuses on idiotic parsing by patterns, rather than words and parts of speech. Also OpenAI claims that they annotate using random texts, while I have long developed billions times cheaper method of annotating by dictionary definitions and articles from encyclopedia.

Why? Who finances this?

[I.Geller]

Another argument why you should avoid OpenAI

Another argument why you should avoid at all costs the use of OpenAI technology:

1. Annotating by randomly found texts OpenAI can not verify their validity, that is its technology can annotate information by fake patterns. And then the devil knows what will be gotten as explanations on your patterns...

2. My technology of annotating by dictionary definitions and articles from encyclopedia is protected from the use of fakes: definitions and articles are obtained from apriory well verified sources.

[MonkeyCee]

Fakes versus bias

I thought it was fairly easy to spot fake news stories. Like all lies, they are almost always too perfect and too simple. They are designed to skip past your rationale and appeal directly to your emotions. You WANT them to be true.

As an example, the whole David Cameron and a pigs head seemed to be bollocks to me from the start. No one has said "I saw him do it" only "my mate saw him, and my mate now says he didn't".

But how many people believe that story, because they would really REALLY like to find it true, since it would confirm their own beliefs.

So we can know a story/video is fake, and still in our hearts believe it. And no legislation is going to change that.

Photos have been faked for years. Video is a series of photographs. It's just a scale issue. So we all take our news with a grain of salt, which is probably a good thing.

[Anonymous Coward]

Re: Fakes versus bias

But that's the problem. If you know "too perfect and too simple" seems to be a lie, then you fill the market/news with it, and it's much harder to filter out to find things.

Hide a needle in a stack of needles. Once you publish only lies, people have less places to go to to find facts.

[Eddy Ito]

Re: Fakes versus bias

So we all take our news with a grain of salt, which is probably a good thing.

Exactly, it's largely a matter of learning how to read Pravda or the New York Times.

[I.Geller]

Re: Fakes versus bias

Yes, a bit of pepper under one's tail is a good idea!

There is no a text created without a bias, everything is subjective. The whole question is to isolate the bias that you like, as a system of sets of patterns; where the patterns' words have a clear polarization - you can remove those biases which are wrong.

By doing this you will become blind and deaf, because you will see only what you want. Therefore read fake news first! This way you train your brain.

[doublelayer]

Re: Fakes versus bias

Of course there needs to be critical thinking. Nobody denies that. However, even the most critical of thinker eventually must find some information to trust. They can get that from a number of places, but the fact that video was hard to fake convincingly meant that it could be used as a better primary source for some time. Of course videos could be edited to remove important sections, but it wasn't easy to insert completely false information into them. Many people have grown to consider video a good source of information; if there's video of it, it happened. The problem isn't those who never thought critically about the news, as they have believed whatever they were told, true or false, for a while. The problem is the people who try to find sources and verify things only to find a video that cannot be confirmed.

[bombastic bob]

Re: Fakes versus bias

"So we all take our news with a grain of salt, which is probably a good thing."

The BEST thing that can come out of a ship-load of convincing 'fake news' is that people actually DO this, start thinking instead of feeling/reacting, and recognize that there is way too much B.S. out there to leave your skeptic hat on the hat rack while viewing it.

Hell, let's just call B.S. and 'fake news' on EVERYTHING, and wait for the dust to settle.

[Charles 9]

Re: Fakes versus bias

Nope. Won't work. People are too busy. That's why things like Twitter and I stage am caught on. They want the truth, condensed, because they have no time. If the truth can't be condensed into 240 characters or less, it's TMI.

[Eddy Ito]

Re: Fakes versus bias

That's just the problem. Oftentimes truth simply can't be condensed. When condensed we get things like that viral video of the catholic kid and the native american at the Lincoln memorial that gets sensationalized by most major news outlets yet isn't true.

[Charles 9]

Re: Fakes versus bias

Then you have a major problem because if the truth can't be condensed, then it won't be believed, end of. Keep It Simple, Stupid and all that.

[Robert D Bank]

this could easily become like plastic pollution in the oceans, we all end up consuming it in little bits, whether we like it or not, and it'll be everywhere. At least until there's a way to recognise something doctored technologically.

[bombastic bob]

plastic pollution in the oceans? ugh, I hate environmentalist wackos using these kinds of premises as "examples" because we can EASILY forget to QUESTION THE @#$%-DAMNED PREMISE IN THE FIRST PLACE!

oh, and thanks for the subtle 'fake news' embedded in there. Nice. Job.

this could easily become like YOU BEATING YOUR WIFE because that is JUST AS FAKE as you asserting the 'plastic pollution in the oceans' thing.... [yes it's a reference to that classic leading question of 'how long have you been beating your wife' to which there is NO possible answer that comes WITHOUT the un-due criticism].

Such tactics are transparent. Your 'tricks' are for CHILDREN. <-- that was a reference to a breakfast cereal commercial

[J.G.Harston]

VoteUp: Indonesia doesn't have rivers full of plastic because of westerners using carrier bags, but because INDONESIA not having a functional waste disposal system to deal with the plastics thrown away by INDONESIANS.

[Charles 9]

Except you can't ask the wife beating question to a BACHELOR.

[Adrian 4]

Recognising doctored technology won't happen - it's an arms race between fakes and detection.

This is a good thing - it will take a while, but eventually everybody will be as suspicious of video as they already are of simple photos.

[Charles 9]

NO, because the end result will be DTA, and people will retreat into their instinctual echo chambers. Without SOME level of trust, society will cease to function as we know it.

[Mike 137]

"The constant flood of information makes it difficult to reign in the viral nature of social media."

I suspect the Queen of England would agree. It's probably difficult to rein it in as well.

[amanfromMars 1]

Re: Constant Floods of Information Making IT Difficult to Reign ....

I suspect the Queen of England would agree. It's probably difficult to rein it in as well. .... Mike 137

Next to impossible and never are the two safest bets to make regarding issues aired there, Mike 137.

Methinks a Sublime Guidance with Guardians of Palace Barracks is the Best of Arrangements Provided in one Virtually Advanced IntelAIgent Operating System Offer that Quells Troubles with Soluble Solutions ...... Dynamic Pragmatic Positions.

And now you know what Crown Services and MI sections are busy doing for higher payments to the Defence Budget ‽ .

[Hemmels]

Missing the point

Isn't everyone looking at the technology involved and looking for regulation or protection of interests of this stuff? Take deepfakes out of the equation for now. I guess the concern is "But they are so misleading, or will be in future and that's a worry" - But why is it a worry?

"Fake news" has grown a lot recently because of the prevelance of social media, but as far as I'm aware, even before the 90s we learned at school how to find and trust primary sources before jumping to "OMG THEY SAID THIS".

Doctoring videos has been around as long as videos. I don't even trust newspapers given their funding sources and interests of their owners.

Surely it's up to the consumer/viewer to say "I've seen a video showing X said Y" and not "X said Y". Isn't "don't believe what you see on the internet" a thing still too?

[JLV]

Re: Missing the point

You are, of course, right. But what has changed is the capability to self-publish and forward information. People weren’t necessarily smarter 30 years ago, but they could not propagate rumors as quickly to as many people. Hoaxes did exist, see Protocols of the Elders of Zion, but most of them failed to take root, it took a lot of logistical effort and persistence to keep them up against the pushback of a competent free and open press. Now all you need are Youtube claims about 9/11 or vaccines and you can make a persistent mess.

Law of unintended consequences in what Tim Berners-Lee came up with, though the baby is well worth her bathwater.

[a_yank_lurker]

Re: Missing the point

While the Internet allows for faster dissemination of rumors, hoaxes, and the like it also allows for rapid dissemination of information. It puts more onus on the individual to verify the source and the quality of the information than before. Before, the press did this job (often incompetently).

What was not really emphasized to most in their education is the need to and how to check something against sources available. A corollary of this is also what the best sources to use for checking information and how to detect likely bias in the sources, again something that has been lacking in education. Don't use Wikipedia is not really teach someone what to use or how to use the sources.

[JLV]

Re: Missing the point

In a world where Kim Kardashian is worth 100s of M$ there is no reason to expect that a sizable proportion of the electorate isn’t going to stop at the ooooh, shiny, so pretty level. For those people the availability of alternate sources of information that they would have to think through is pretty much irrelevant if you’ve managed to push through and crystallize an initial false understanding. Ditto cases in our extremely polarized politics where folk will just ignore info that goes against their preconceptions.

So it seems easier to preempt fake news than to rely on everyone seeing through it.

[bombastic bob]

Re: Missing the point

"even before the 90s we learned at school how to find and trust primary sources before jumping to 'OMG THEY SAID THIS'."

Fixed it for ya. I think that pretty much explains it. 4-inchers (people who view everything through a 4-inch screen) under the age of 40 may be driving all of this... and the "lack of education" system set them up.

All their base are belong to fake news. They have been set up the bomb. etc.

/me has a nice faked-up photo that I did with Obama's face as 'Cats' and the 'All your base are belong to us' as a demotivational... [it was fun]. It wasn't purely my idea, but I ran with it and made it better.

[doublelayer]

Re: Missing the point

"Surely it's up to the consumer/viewer to say "I've seen a video showing X said Y" and not "X said Y". Isn't "don't believe what you see on the internet" a thing still too?"

But that's not good enough. It's well and good if I know something's false, but if someone else believes it, I can still get hurt. Suppose someone is angry at me and makes a fake video of me plotting a crime. I clearly know it isn't true. However, I need my boss to know that as well so I keep a job. I need my neighbor to know that as well so they don't call the police on me. I'd prefer that my friends and family know that so I don't lose their company. When a lie believed by someone else can cause problems for others, it's not enough to say that people should just come to their own conclusions. If something can be done to support the objective truth (I'm not sure there is such a thing, but if), it should be considered quite strongly.

[Hemmels]

Re: Missing the point

If your boss puts you on garden leave for something you didn't do, happy days. If anyone else wants to presume guilty, more fool them.

And yes, I know in the real world this isn't a thing, and trial by media exists and lives are ruined (hospital staff turned killers anyone?) but that doesn't mean we can't stress the point that this isn't acceptable and bow down to it.

[JLV]

Any of you look at the vid that brought down the Austrian far-right-based coalition? So far, no one really knows who scammed them, but it was someone pretending to buy influence on Russians’ behalf. Long vid, plenty of details, but I can see this not being innately believable in 20 yrs.

There’s a lot of downside to deep fakes but also upsides to exposing venality and corruption in politicians. It would be a shame if investigative journalism lost its bite due to this.

[croc]

How do you do a "deepfake" of someone that speaks mostly in lies and the rest of the time only utters nonsense...? I am of course speaking about the orange-haired CIC of the USA.

[bombastic bob]

"I am of course speaking about the orange-haired CIC Speaker of the House of Representatives of the USA."

Fixed it for ya. I think Pelosi was drunk at the time... [that's her new excuse]

[Anonymous Coward]

Does it really matter though? At least in the US, comments like grab her in the pussy ARE muttered by even the president - deepfakes might make that guy more credible. I suspect many other countrie's MP's too

[J.G.Harston]

"It requires .... changing civilian behaviour."

So, stopping humans from being humans. When has that been tried before? New Soviet Man? Humans don't work as we think they should, we need to change them.

Humans are social creatures, social communication is near enough the foundational definition of what makes humans human, it's close to being the cause of why language itself evolved. And these people have declared that the fundamental defining characteristic of being human is a flaw to be eliminated.

[Adrian 4]

Wise words

"David Doermann, a professor working at the University of Buffalo’s Artificial Intelligence Institute, summed it up succinctly: “A lie can travel halfway around the world before the truth can get its shoes on,” he told the committee."

Goodness me. The committee must have been astonished to learn of this. What a revelation !

[Richard 12]

Re: Wise words

I guess shoes are quicker to don than boots.

[DavCrav]

Re: Wise words

"The committee must have been astonished to learn of this. What a revelation !"

Yes, the lie in that quotation by David Doermann must be very Swift.

[DavCrav]

Re: Wise words

"'A lie gets halfway around the world before the truth has a chance to get its pants on.' Winston Churchill.

The sad fact is the truth never catches up to all the people the lie reached."

Like you. The quotation was probably by Swift (as I implied in my comment), but predates Churchill by a century or two.

[Kaltern]

This is turning into a Flat Earth type discussion...

[Anonymous Coward]

Don't know why politicians seem to be so panicked by this. It saves them coming up with their own lies and sounds like it would be a real time-saver. Isn't this what computers are for?

[I.Geller]

Lexical noise as another argument why you shouldn't trust OpenAI

There is a sentence "Alice and Bob exercise merrily, she trains a lot."

The word "merrily" can be an example of lexical noise; where it's typically superfluous patterns that do not explain the central themes contained within the digital textual information and, accordingly, removal of such noise often results in an improvement in the quality of the structured data.

Suppose somebody doubts if Merrily the athlete who trains a lot being exercised by both Alice and Bob - it's a name. Or is it an adjective?

One wrong pattern can radically change everything!

With my patented lexical noise deletion AI-parsing gets these FOUR patterns, from the sentence:

- Alice exercise merrily - 0.25

- Bob exercise merrily - 0.25

- she trains a lot - 0.5

- Alice trains a lot - 0.5

AI sees that the word "merrily" is an adjective from its context and subtext, from its dictionary-encyclopedia definitions.

Without the purging, not knowing if "merrily" is a name or an adjective, AI-parsing gets FIVE patterns:

- Alice exercise merrily - 0.1(6)

- Bob exercise merrily - 0.1(6)

- merrily exercise merrily - 0.1(6)

- she trains a lot - 0.5

- Alice trains a lot - 0.5

With the purging AI gets TWO synonymous clusters, without - THREE!

OpenAI cannot delete lexical noise and its results are not trustworthy - OpenAI sees no separate words but patterns, doesn't determine parts of speech and uses dictionary-encyclopedia definitions.

My AI database can, therefore, be 100% trusted.

[I.Geller]

Guys, I'm glad! Three "thumbs down"! Finally, at least some reaction from people who understand what I'm saying.

[Chris Tierney]

Jumbled mess

Tidy it up.

[I.Geller]

More on lexical noise

Lexical noise is the biggest problem for AI because one accidental pattern can change everything.

For instance there is a sentence "Alice and Bob exercise merrily, she trains a lot."

The word "merrily" can be an example of lexical noise; where the noise is typically superfluous patterns that do not explain the central themes contained within the digital textual information and, accordingly, removal of such noise results in an improvement in the quality of the structured data.

Suppose computer doubts - if merrily the athlete who trains a lot being exercised with both Alice and Bob - and AI thinks that "merrily" is a name "Merrily". Or is it yet an adjective?

Using my patented lexical noise deletion AI-parsing gets FIVE patterns (from the sentence) because AI sees that the word "merrily" is an adjective (from its context and subtext, from its dictionary-encyclopedia definitions).

Without my purging (dictionary-encyclopedia) the same AI-parsing gets SEVEN patterns! and TWO synonymous clusters with, without - THREE; among them this:

- merrily exercise merrily - 0.1(6)

- merrily exercise a lot - 0.1(6)

Thus computer finds the above sentence searching for the query "Is Merrily exercise a lot?", even if the sentence doesn't say a word about Merrily as a person.

OpenAI and other companies, which annotate using random texts and not by dictionaries-encyclopedias, cannot delete lexical noise and their results are not trustworthy: they will find the above sentence looking for "Merrily". Many of them are saying that their trained language models provide up to 90 percent accuracy, while my AI can provide 100% accuracy.

My AI database can, therefore, be 100% trustworthy and what they found cannot at all.

[Milton]

"Antisocial media"

I daresay "antisocial media" was not a typo, and it carries a lot of truth. I'd suggest it's increasingly hard to argue that the free social media model, in particular facilitating the ability of so many cowards to screech anonymously, is on balance a good thing. If we could wind time back 25 years, wouldn't we insist that you could not post anonymously on the net (any more than you could have a letter published anonymously in the newspaper) and that internet services like Google and Facebook must be paid for, certainly not "free" to abuse people's privacy and personal data?

Consider that cowardly anonymity has been a problem since the days of libels being handed out in the steets of London, c.1675. Consider that even if deepfake tech had existed 50 years ago, it would barely have mattered, without the internet to spread noxious lies faster than the truth could catch up.

One (admittedly slightly simplistic) way of summarising the effect of the 'free' anonymity-favouring internet is that it has provided a fertile ground for those with the minds of children. Lies, stupidity, irrational hatreds, fear and bigotry have a safe place to play, with far too little adult restraint.

Indeed, the internet, providing a warren of dank, dark spaces for the foetid of thought and speech, has been an absolute gift to the political right: it's where their lies—almost the only currency they have left—can gain traction and visibility.