- 1 week ago
- #ai
- #technology
- #deepfake
- #movies
- #future
Is the world turning into a playground for fakes? Dive into the wild world of deepfakes, AI, and the tech shaping our reality! From resurrected movie stars to mind-blowing neural networks, we cover the cool, the creepy, and the hilarious fails. Don't miss out—subscribe for more tech goodness and let us know in the comments what blew your mind most! #AI #technology #deepfake #movies #future
👉 This channel was created in collaboration with https://www.youtube.com/@kramola_online
0:00 - Introduction to Synthetic Realities
0:46 - Democracy and Media Manipulation
0:54 - Deepfakes in Entertainment
2:18 - Technologies Behind Deepfakes
3:55 - Corporate Race Toward AI Dominance
5:36 - AI Limitations and Security Risks
9:09 - Societal Impact and AI Responsibility
11:03 - Future of AI and Human Understanding
👉 This channel was created in collaboration with https://www.youtube.com/@kramola_online
0:00 - Introduction to Synthetic Realities
0:46 - Democracy and Media Manipulation
0:54 - Deepfakes in Entertainment
2:18 - Technologies Behind Deepfakes
3:55 - Corporate Race Toward AI Dominance
5:36 - AI Limitations and Security Risks
9:09 - Societal Impact and AI Responsibility
11:03 - Future of AI and Human Understanding
Category
🤖
TechTranscript
00:00It seems like our world is turning more and more into some kind of fake paradise.
00:05Or a paradise for fakes.
00:08Mona Lisa comes to life, Brad Pitt becomes the Terminator, or Tom Cruise, or Sylvester Stallone.
00:14Of course, it all started with adult content, but now they're even changing actors so that they speak other languages
00:21with matching mouth movements,
00:22meaning they use neural networks to synthesize lip and facial muscle movements.
00:32Of course, sometimes it seems like the lower part of the face has just moved away from the upper part
00:36to Voronezh and is living its own life with kids and a mortgage.
00:40But any technology is imperfect at first, and then gradually improves and can even become like this.
00:46America! You blame me for interfering with your democracy, but I don't have to. You are doing it to yourselves.
00:54This is nothing new for the film industry. In 2019, the ninth episode of Star Wars was released.
01:02The cast list included the late Carrie Fisher, whose character appeared in the film despite the actress's death.
01:08The audience received the movie rather coolly.
01:11An example of a deepfake that is considered more successful is the appearance of the late Paul Walker in Fast
01:17and Furious 7.
01:18The actor died in a car accident before filming was completed.
01:23His sudden death forced the script to be rewritten and the movie to end on a special note.
01:28Instead of Paul, the remaining scenes were played by his younger brother, Corey Walker.
01:33But if a program allows us to bring cinema legends back to life, do we even need new actors?
01:38And maybe now, people will be required to specify in their wills whether they allow or forbid the use of
01:44their likeness in the future.
01:46After all, concerts featuring already deceased performers are nothing new.
01:51Instead of Tupac Shakur, there was a hologram.
01:54They brought him back without permission, and they made a fortune from ticket sales.
01:58So who, in that case, has the right to control a celebrity's likeness?
02:01It's quite possible that after some time, celebrities will start passing the rights to their likeness to close relatives.
02:08Then there's at least some chance that, after death, they won't have to participate in projects they wouldn't have liked
02:14while alive.
02:18Tom Cruise has probably become one of the most popular subjects for such videos.
02:22Politicians also look real in deepfakes.
02:25Not only does their appearance change, but it's now common for their voices to be simulated as well.
02:30Our engineers have trained a neural network to speak in different voices.
02:34Now, absolutely any voice can be copied and used to say anything.
02:39It's actually amazing, although a little bit scary.
02:42In the U.S., deepfakes are considered a threat to national security.
02:47In California, some types of deepfakes have even been banned by law.
02:52AT giants are developing anti-deepfake technologies, training algorithms that could accurately detect fake videos.
03:00Recently, Facebook, which I'll talk about a bit later, organized the Deepfake Detection Challenge, a competition for the best deepfake
03:07detection program.
03:08The prize fund was no less than $10 million.
03:12First place, by the way, was taken by a programmer from Minsk, Salim Seferbikov.
03:17He received $500,000.
03:19Russia is also keeping up.
03:20So, in February 2021, the Russian Ministry of Internal Affairs announced a competition to create an effective detector.
03:27The initial contract amount is 4.8 million rubles.
03:31The system's codename is Mirror Camel, so that no one would guess.
03:36A similar algorithm was developed by scientists from the University of Buffalo.
03:40It supposedly has an accuracy of 94.
03:42It detects deepfakes by analyzing the reflections in the eyes.
03:47But that won't last long.
03:49Sooner or later, we'll have realistic reflections on the cornea, too, since the field of machine learning is developing very
03:54rapidly.
03:55For example, the Big Gun algorithm from Google and NVIDIA not only allows you to replace faces, but also to
04:01create faces of non-existent people that are impossible to distinguish from real ones.
04:05Most often, deepfakes are created using a so-called Generative Adversarial Network, or GAN.
04:11It consists of two systems, a generator and a discriminator.
04:15They work like a pair of students in a university study group.
04:19One suggests ideas, the other criticizes them.
04:22The generator creates an image, and the discriminator, trained on real photos, indicates what needs to be fixed.
04:31Usually after this, people comment,
04:33Very interesting, but nothing is clear.
04:36But in fact, you can definitely master it if you really want to.
04:40Not so long ago, Zuckerberg set a goal for his research centers to create artificial intelligence by 2025.
04:46That would surpass humans in vision, hearing, speech, and general cognitive abilities.
04:52In other words, essentially, Zuckerberg wants to create a strong AI machine that will surpass humans.
04:59Another tech figure, Shane Legge, the chief researcher at Google DeepMind, predicted that by the mid-2020s,
05:05artificial intelligence will finally surpass humans.
05:08And it might seem like all of this is just another set of fairy tales about the great AI,
05:13which are regularly broadcast by strange people called futurists.
05:17But it's one thing when this kind of fortune-telling come from individual oddballs,
05:22and quite another when megacorporations with tens of thousands of employees and thousands of scientists working for them,
05:28with billion-dollar research budgets,
05:30put in their business plans to create a machine that surpasses humans in every way.
05:36In fact, Facebook is already practically such a machine.
05:40The collective mind of this company has created a model that extracts enormous resources from humanity to achieve its goal.
05:47And this goal, of course, is not to promote the flourishing of culture and science,
05:52not to benefit nations.
05:54In the mid-2010s, a race began among the largest global technology companies.
05:59There are no more than a dozen of them for the right to
06:02to become the company, the only corporation that could soon dominate the entire world.
06:08Strong AI is just a means.
06:11However, even the most ambitious entrepreneurs have not yet learned to change the laws of nature.
06:17And it seems that this task may turn out to be much more difficult than they think.
06:21Today, programs lack understanding of meaning or consciousness.
06:25That is, what is the essence of the human mind.
06:28It's interesting whether AI will ever be able to overcome the semantic barrier.
06:32Said mathematician and philosopher Giancarlo Rota once.
06:35There's a vivid example.
06:37A sausage roll.
06:38A deep learning or machine learning system would understand this phrase as a meat product
06:43inserted into a relative with the social status of a father-in-law in a certain place.
06:48The journal Computer Science published a study in August 2018, ironically titled
06:53The Elephant in the Room.
06:57The scientists showed that if you insert a small image unrelated to the subject somewhere
07:02in the corner of a large image, for example, an elephant in a picture of a living room,
07:06this will affect artificial neural networks and deep learning algorithms so strangely that
07:12the AI will end up classifying many objects in the picture into the wrong categories.
07:18Here's another example.
07:20Google created a drawing tool that allows people to draw imaginary animals from their fantasies.
07:26It's called Chimera Painter.
07:29But if you confuse the neural network, you don't just get beautiful dragons like these,
07:33but also something like this.
07:36Programs that confidently beat humans in computer or video games would start losing badly as
07:42soon as you changed something in the game, like the background color or slightly move the
07:46player's starting positions.
07:48Overall, these examples show that even the best AI can fail, but not all mistakes are the same.
07:53It's one thing to lose in Counter-Strike, but it's a whole different matter if an AI at
07:58the airport doesn't let you board and instead calls the police, mistaking you for a criminal.
08:03Or if an AI trained on fake roads in a self-driving taxi doesn't notice you crossing the street
08:08right in front of it because of unusual lighting.
08:11Yes, that's right.
08:12If you didn't know, arrays of fake data are used for training in many areas, in retail,
08:18robotics, and of course, in autonomous vehicles.
08:20Actually, they're not called fake, but synthetic data.
08:24It sounds more respectable that way.
08:26After all, no one is going to train an autopilot by crashing hundreds of real cars off bridges
08:31or cliffs until it figures out how to recognize them properly.
08:35As a result, developers create entire virtual cities where models drive billions of imaginary
08:40kilometers.
08:41And then they fine-tune everything on real roads where the now-experienced autopilot racks up
08:47the final mileage.
08:48On top of that, hackers can interfere with the algorithms, which aren't very reliable
08:53to begin with.
08:54A hacker can introduce those tiny distortions into an image or text, which can lead to major
08:59errors, and those in turn can have catastrophic consequences.
09:03For example, in power management systems or in the traffic control of a metropolis.
09:08Imagine an audio signal that blends in with the background music in someone's home, but
09:13their AI assistant, Alexa or Siri, picks up this signal and interprets it as a command.
09:18For example, to turn off all the lights in the house, lock it, or unlock a locked door
09:23at a certain time.
09:24As Pedro Domingos noted in his book The Master Algorithm, people worry that computers will
09:30become too smart and take over the world.
09:32But the real problem is that they're too dumb and have already taken it over.
09:37Can I even overcome the semantic barrier or the so-called barrier of understanding?
09:42Undoubtedly, to do this, we need to study more deeply how human perception and consciousness
09:46work.
09:48Human history has shown that we adapt quite well to a wide variety of situations.
09:53Some things we do industrially, some intuitively, some instinctively, some according to common
09:59sense, and some in spite of it.
10:02A human is a complex being, but without this complexity and unpredictability, there probably
10:08wouldn't be consciousness.
10:09For decades, researchers have been trying to develop methods for teaching AI generalization,
10:14common sense, and even things like intuition and imagination.
10:18But while they've made some progress in generalizations, with all the caveats, there hasn't been much
10:24progress in other areas yet.
10:26All of this might not concern us, ordinary people who are far removed from the challenges
10:30of building AI.
10:31Let the scientists do their job, and we'll do ours, right?
10:36No, not quite.
10:37Or rather, not at all.
10:39All these poorly functioning PC systems and artificial intelligence agents are being used more and
10:45more everywhere.
10:46And they're making decisions that people's lives depend on.
10:49There are huge investments behind AI, with massive fortunes and the careers of talented
10:54industry leaders at stake, ambitious scientists, and bureaucrats who dream of advancing without
11:00doing anything or taking any responsibility, or rather by shifting their responsibility onto AI.
11:07Too many people are interested in possibly spreading AI more widely and perhaps accelerating
11:12its development.
11:14The problem right now isn't that a superintelligence or strong AI that can easily pass the Turing
11:19test is about to appear.
11:21The real problem is that we consider these algorithms to be more advanced than they actually
11:26are, and we're ready to hand over entire areas of human activity to them.
11:30Yes, we are sinful and fallible people, but they want to replace us with machines that make
11:35mistakes just as often.
11:36Machines that don't know sin simply because they don't understand what it is.
11:39But it would be a mistake to think that artificial intelligence is just another overhyped deepfake,
11:45a slightly smarter relative of the calculator, or simply the fantasy and unattainable idea
11:50of techno-optimists.
11:52Tens of thousands of scientists around the world go to work every day to achieve a breakthrough
11:57in creating strong AI.
11:59They are motivated every day to make AI work better and better.
12:03First on narrow tasks, then on broader ones.
12:06Neural network algorithms have become incredibly complex, but most likely, in order to overcome
12:12the semantic barrier, developers and AI will have to take a step back.
12:16Forget about creating ever larger neural networks and databases and look for some interesting
12:21processes from a field of science that is still unknown.
12:25Scientists will definitely have to take a closer look at humans and their characteristics,
12:29at their cognitive abilities and shortcomings, which could also lead to discoveries.
12:34In general, letting the genie out of the bottle isn't that hard, but will we be able to put
12:38him back in?
12:39That's a good question, isn't it?
12:41Let's discuss it in the comments.
12:43And while you're thinking about your deep philosophical answer, here are a few more.
13:15Let's discuss it in the comments.
13:38Let's discuss it in the comments.