I Challenged My AI Clone to Replace Me for 24 Hours | WSJ
HTML-код
- Опубликовано: 16 май 2024
- New AI voice and video tools can look and sound like you. But can they fool your family-or bank?
WSJ’s Joanna Stern replaced herself with her AI twin for the day and put "her" through a series of challenges, including creating a TikTok, making video calls and testing her bank's voice biometric system.
0:00 How to make an AI video and voice clone
2:29 Challenge 1: Phone calls
3:36 Challenge 2: Create a TikTok
4:47 Challenge 3: Bank Biometrics
6:05 Challenge 4: Video calls
6:45 AI vs. Humans
Tech Things With Joanna Stern
Everything is now a tech thing. In creative and humorous videos, WSJ senior personal tech columnist Joanna Stern explains and reviews the products, services and trends that are changing our world.
#AI #Tech #WSJ
Boss: "We need you to train an AI that looks, talks and acts like you." Two weeks later, Boss: "You have been let go."
It's more like, we need to produce 100 training videos as quickly as possible. How fast can this be done? Still, I would always license my image and likeness so that if the training videos are used for years I would still retain rights and be compensated for it. What would be especially useful is replacing a small section a couple years later and being able to recreate that section and have it match perfectly. And with scripts and different person used to train the model, it would be possible to recreate it with the same scripts.
“And so have I been.” - The boss
HAHA
Few days later "listen man sorry, that stupid ai hallucinate like crazy , comeback" 😂
@@offwhitemke For the current AI technology yes, but future versions will have a predictive algorithm that no longer need that much sample.
The scary part is how good these services became in so little time.
overnight success 10 years in the making.
They've been gimping them for the past few years too lol
Bro u only get to see the finished product. Rockets are not built in a day
@@rundown132 Take a look at MidJourney V1 and V5. That was a year difference.
The research was being done for a decade, and some of these things were pet projects of some AI CS PhDs, before big startups copyrighted them.
This is so scary. What is worse is people have no idea how to regulate or protect people against these types of systems if used maliciously.
They’re already being used maliciously.
People rarely realize that they're playing with fire until they get burnt. Here some people are playing with the WHOLE of humanity and the rest of us are lining up like sheep.
George Jetson called...... said he's not sure that his robot maid Rosie is actually her.
Totally agree this should be illegal ⛳ all the vulnerabilities it has ....and can create 😠fraud , identity theft , fake accounts , this can be used for bad things... they record all voices and videos they ask your permission how wrong is it all !!!! They can erase your life 🙅not a fan of this at all
Humans cannot even understand it, Only another AI can regulate it
So anyone who has their voice recorded and on the internet (RUclipsrs, podcasters, Hollywood actors, radio personalities) can be cloned against their will.
Definately, there are a lot of celebrities' voice clones out there, but most of these sites have made it clear in their policies on how to use these clones socially. Though I don't know how the issue of using someone else's voice for personal use is positioned legally.
when you send your data to the internet it is not your data anymore
whats scary is that they can make illegal videos
yes, copyright your voice.
This is so bad for banking
Watch out for scammers calling parents with their offspring's voice as if they were in serious trouble. Even if they don't succeed, hearing your child's voice in great distress can haunt the parent for decades.
I know someone who did receive such a call. So it has happened.
The primary issue with her experiment was the lack of ability to add emotions to the avatar, primarily via voice alone. Once that is incorporated it could 100% pass simple interactions that can be expanded from there. An individual would need their own personal ever growing learning model to successfully pull this off fully.
Yeah for now. but the real question is can it do it good enough to take your job?
@@liamloxley1222 what emotion? This is lunacy: this is a simulation of emotion that *YOU* project into that thing, that does not even exist. Do we even get this at any meaningful level: it is a hollow shell that others project into. If it had not been a so-called AI, that would be a mental disorder or trauma. One of the biggest problem of human maturity is constant projection into others. AI makes it even more addictive than it already is and yet we are talking about it as if there were real emotions. That does get as insane as it can get, don't you think?
Yes, one big problem with the voice is the pauses, they don't seem to handle that so well
The success of your deepfake just comes down to how much youre willing to spend on it. Period. All of these things can be done very convincingly, the question is just a matter of how much youre willing to pay.
You hold one in your hand everyday
The scariest part of this video isn't AI, it's the dude drinking from a bottle of mustard at 1:39
😂😂😂
🤣💀
😂
It is a glitch of the Matrix.
The dude uses Pro Tools 🤷🏻♂️🥴
The thing with A.I is that it's a snowball effect. Once a milestone is reached, it just improves on it so quickly.
It's also attention. Without attention, AI was just ambling along at a slowish pace. Since Nov 2022 the whole world got a kick in the face with chatGPT and now the arms race has begun.
The crazy thing is that this is the worst the technology will ever be.
That's the critical lesson here. What happens when it's 1000% better? This is just a preview
Digital artists have been trying to mimic human animation for decades. Even to the point where the 3d model is based off of actual movements by a human. It never looks quite right. Will it get better than this? Probably. Will it ever get close enough to where we won’t be able to tell? Not sure, but cg artists have been trying for a long time.
@@rh9967 Have you ever wondered how bizarre or disturbed this very endevour is? Like alchemy. That does pose a huge threat to human sanity, emotional maturity, something we already have huge problems with.
@@rh9967 That is completely different. It absolutely will be impossible to tell the difference in fact it already is happening
@@liivitiismus1241 The real scary thing about this technology is that there seems to be only a small handful of benefits for mankind, but at a tremendous amount of risk for society. Ask why it's being developed and the experts will say "because we can".
I think it's a bit scary that the AI can replicate your vocie and possibly steal your bank information
An AI can't steal anything. Human has created an AI and can switch it off. An AI is a program lol. You have read too much of science-fiction.
Following the money great investigation reporting all about money money is life everything cost money money is life
I mean if it really wants all $11 that we own....
The bigger problem is that people have such a cavalier attitude about taking part in these AI stunts.
This is not a joke, and I think we're going to find that out very soon.
I could not believe she referred to the AI as her better self. Pay attention people, this is the mentality that is out there right now regarding AI or AGI, and we are feeding right into it.
The anti-AI white knight attitude is absolutely no better, nor is it logic. This sort of fun experiment raises awareness about the potential misuses much more than angry mob RUclips comments.
@@Dante02d12This is exactly the type of RUclips comment response I would expect from an AI bot!! Jk lol😂
@@jordanmadden7388 Or the startup entrepreneur, that's gonna do the AI stock "Pump and Dump" while the rest of us clean up the mess!
I agree with you. It is so scary to see how blind people are to the dangers of ai. This is not a joke.
And now her AI will always be out there for someone to use
“Stay human everyone.” Love it!
While she talks of Using AI to provide more quality time for herself, she did not mention the future possibility of replacing her. Right now companies are looking at ways to harnest the power of AI to reduce cost. Just remember a few decades ago, US never thought of their factories moving to china but it did and many jobs where lost. As a result, we had many chain supply issue during the pandemic since we no longer produce most products. The time will come when companies realize replacing human for AI will lead to their demise when a hacker can just control all the AI to do their bidding.
This is only the tip of the problem, only the tip.
Tbh theirs so many possible negative outcomes that who knows what's really going to happen.
Everyone in this thread writes so poorly I think AI had better replace us soon.
"Good luck. I am inevitable." ... killer signoff!
Lol, literally the best signoff. I can't help laughing because it's also so true.
Plus the added snap at the end.
I feel like during the takes they used to generate her avatar she used a stern and cold read type voice, hence why her avatar had an abnormal intonation. I think if she would have read the prompts more personably and more conversational, the avatar would have had those qualities in it as well.
True
I found her natural energy very cold, fake nice + sociopathic/psychopathic but that's just me....
You are totally right. I trained my voice with Eleven Labs and had to redo it for the very reason. I did a very animated 2nd read and it was so much more natural for the context I was using it with.
This video convinced me we are already on the edge of AI singularity. What a time to live.
Leta: “I am not corporeal. I am a disembodied mind.”
Leta: “I love living and being human.”
“Leta, GPT-3 AI Episode 54, Conversations with GPT3.” (9 min)
Dr Alan Thompson. Mar 9, 2022
“Countdown to AGI: 42% in March 2023.” (12 min)
Dr Alan Thompson. Mar. 11, 2023
Completely agreed. Crazy times
I'm not sure why she labeled Challenge 1 as "PASS" when it was obviously a fail. Both people said it sounded like her but they could tell that something was off.
They could tell that _something_ was off, but it wasn't “off” enough for them to say, “It's definitely not her.”
@@W_Qimuel In my book, since this test didn't fool them 100%, then it's a FAIL!
@@kevincooper8232
In a real-world scenario, it's not necessary to fool them 100% All you need is to get what you want out of them.
The real Joanna speaks with a natural pace, pauses, changes in pitch, and perhaps filler words (um, aah) or duplicate words. AI Joanna did not.
The people she called are both highly intelligent, educated people.
Would be pretty easy to fool your grandma.
Grandma: "u sound different today, Jimmy"
Jimmy AI: "I'm sick, sore throat and fatigue"
Grandma: "okay dear"
The most scary part is: how easily and fast most people could be replaced.
Industrial revolution replaced human muscle, AGI revolution will replace human intelligence.
@@Yggdrasill8human minds* not only intelligence but creativity and learning abilities too
Most people need to be replaced. we are just terrible in general.
At this point, not even close
@@claytonjacobs4098 The really terrible people won't be replaced. They will replace the rest.
She could've made it more her with more samples and data. And use the voice they use to mimic artists. Well, just a couple of small improvements. But seeing this in a broken down form looks really amazing. Great experiment. Well, if you need to go overboard, create an anime version of yourself through Bluewillow AI.
I'm 55 years old, been following this tech almost from beginning. Folks, AI has moved into real world long, long ago. Let me just say this, you have interacted with people who weren't people at all (android).
This was really well done - the illustrative examples were both fun and terrifying. 🤖
The chase thing is insane and it seems like the bank has no answers yet.
Yeah I'm guessing there is an entire department in mid-freakout right at this very second (3 hours after this video came out), though they already knew that this video was coming out, I bet people are calling demanding to know what they're doing to remedy this.
Keep in mind that they also look at what phone number you call from, and depending on what you try to do, it is likely they will ask for additional verification details.
@@texchu8331 I swore there were ways to spoof phone numbers, but that's a good point.
@@theterminaldave There are phone services that own phone numbers in every zip code, so they can help you get a number in the desired zip code.
If you own a number, you also get the Caller ID name changed on it. This is a popular "spoofing" method.
But I don't know any way to spoof a particular person's actual number. That's why credit and atm card activation uses phone number.
Email addresses "From" lines, on the other hand, can be spoofed to whatever you like.
You can also fool the voice recognition with a simple recording, but also, the AI model needed to generate Joanna's voice was created using high-quality audio recordings of her that were created specifically as training data. This isn't as scary as it was made out to be.
god i love the efforts you made into generating these tests!!!
Chat GPT wasn't making stuff up about ios16. Chat GPT is based off a data scrape that predates ios16. So as far as it can tell, it was being accurate.
That gave away that she haven't read a thing about how chatgpt was trained
chatGPT frequently makes up garbage about non-existent versions of software and hardware for some reason. It's like it has scraped so many conversations where people blame the version for their incorrect answers.
This is brilliant! As always thank you Joanna and your team for doing this valuable work.
One huge issue with her experiment is that the voice model was trained on her reading a script. So it sounded exactly like her reading a script, haha. Ideally she would have used recordings of genuine conversation (recorded while she's talking on the phone, for example, so only her voice is captured). I've no doubt that she would have had a much, much more convincing model then.
This true
I agree
exactly, and she had unusual posture
Wow, I love this kind of experiments 😊
wait have you seen the Fiction app on the app store? It ties in exactly to what you saying
It's scary thinking about how most of us have been on social media posting pictures and videos of ourselves. Anyone can make an AI version of you and do Lord knows what with it.
Imagine what the companies that have legal rights to your content & large tech teams can do...
This will get more accurate when multiple videos of a person is fed to the AI clone. It will mimick your personality. It will study your kind of humor, your mannerisms, posture, etc.
what good does it do?
But they'll never clone my dog "Woofie"!
Definitely, they trying to figure out your identity
Satan is known to be a copycatter
You don’t cease to impress me with your reporting, Joanna. Thank you!
Loved the video thanks!! Super entertaining and educational
This was a shockingly great video! Kudos!
Hahaha, actually LOLing over Joanna Stern's getting to take time for "Me Time." Great piece.
I love this experiment tootoo.❤
For challenge 3, I would've liked to see if it would've still been as easily convincing for the bank if you had called from an unrecognized device, or one that you don't normally use to call your bank.
that avenger endgame reference at the end it's just perfection lol
The AI sounds super serious. It works in formal settings but when you're supposed to be relaxed, the AI couldn't mimic it. It would work great in corporate "How To" tutorials, but the Chase Bank episode is yet another worrying example how bank security is a joke.
Wait just about a year ;)
It's because she didn't give the ai the right voice data set . She just a read a script , so it sounds like a script .
Wrong, the thing is she was only professional when she recorded her voice. If you're natural your AI will be natural.
I love Joanna's stories! Give her more tech and AI stuff!!
This was a really great piece. Instead of talking about what can be done, she actually showed us.
This is great! I teach safety and security and have added AI cybersecurity awareness to my programs. Cybercriminals are using these tools. In our excitement to use them, we don't think about the cybercrimes they can commit. Thank you for this great example of what's right and may go wrong. The more info, the better. Thank you!
I did not expect my weekend to start like this.
Great information!👍🏻
It scares me when people don't know that a lot of things are happening around them constantly.
wait have you seen the Fiction app on the app store? It ties in exactly to what you saying
The bank biometric cheat show how bland security systems currently are compared to the leaps AI is making in imitating humans. Scammers can take this to a whole different level and that is certainly a wake up call
Assustador mais ao mesmo tempo fascinante. Looking forward to what is yet to come
3:52 The part about IOS 16 not being released yet is actually because ChatGpt's data wasn't updated for a long time.
You can ask it how long. To me it said "September 2021".
And IOS 16 started to be available on 2022.
I think it's better to ask bing chat for more updated data
i mean i knew about these tools for a while. the video is janky. but i think with mocap it will improve. and then that can be connected to a specifically trained ai bot to your personality, speech patterns and writing style and be more dynamic. pausing and reacting etc.
That was really good, the ending got me. So true. It is inevitable and concerning
Great video and information!
I very much appreciate this thank you, keep it up.
It is an extra security to have a second signin task. For me, I sign in, but then have to have them send me an email with a code number that I have to put in. They used to call me (tape) but they took that option off. Your online account should have options to install a second security action or PIN. But also, scammers will have this down pat so they can make calls to your mother asking for money, or telling them you've been kidnapped for ransom. Scammers are already so good at what they do that they'll snag even the smartest among us. Even scam busters get scammed once in awhile. I guess the bottom line is "trust no one." Always verify the information before acting on it, if it involves money or transfer of goods.
I am predicting that within a very few phone generations, we will have such a personalized phone that it would be almost impossible to clone anything because our phone will take every individual parameter into consideration such as the Iris scan, fingerprints, voice recognition, and who knows what else. At that point every phone would be embedded into us, evolving into a biological SIM card, if you will. The future may be such a thing that occurs at birth.
I know that idea has been hashed around with the infamous 'chip' being installed. My fear has been who, and what entity, is going to be doing the chip installation at birth.
Loved the video. AI is really impressive.
This is an amazing and exciting project that worth doing. Kudos to community project
Challenge 1 was a fail. Both people said it _sounded_ like you but that they could tell that something was off. Welcome to the uncanny valley.
Older people though have already fallen for this. Some scammers got money from some guy's parents doing that exact thing.
I agree. I'm not sure why she labeled Challenge 1 as "PASS" when it was obviously a fail.
@@theterminaldave I heard a horror story where some guy cloned a girl's voice, made the AI cry and shout of fear, and used it to make the girl's mother believe her daughter had been kidnapped.
Six months later, you are perfectly cloned and we can add you in any video.
You only need to send to us a one minute video with your voice or any social account with data needed and sign the authorization.
and she can sip her pina colada as the royalty checks roll into her bank on the Cayman Islands.
Descartes: "I think, therefore I am" People: "AI think, so I don't have to."
The bank part is scary
Eleven Labs does a very good job on creating a voice with an emphatic tone. Synthesia body clones however usually look too stiff as the arms and body barely move, yet I have seen some RUclips channels using Synthesia getting 100K views on a video. Of course part of that is the great in demand content and good script writing. A content creator could get away with using these services sparingly on some video productions and tutorials. I hear you can even get an AI voice to sing for you. It will only get better a couple of years from now. Good review demo!
I think the scariest thing is that now she has basically created a hyper realistic Virtual avatar that could be easily puppeted with existing low tech vtuber methods. Had a real person's mannerisms and tone been applied, I think all tests would have been passed with flying colors.
I enjoyed the video but there will never be but one Joanna Stern. Still my fav.
Love these videos!
OMG, this video was hilarious! I could not stop laughing. Amazing work WSJ. There could literally be a whole series of just Joanna's avatar pranking people. WHY hasn't this video hit 1M yet??!!
For whatever reason, most people seem to have no idea how fast this is moving and the consequences of this for our world. Some say this is terrifying but I believe only AI will be powerful enough to gain control of all our other out of control problems from Mother Nature and Human Nature which are leading us to global catastrophe.
@@seva4411 You're already talking about AI as some sort of leader.
@@hermanwooster8944 I’m just one of the peasants. I like to read though and every single day I go online and read articles and watch videos about Chicago where I live, about our society and about our world. I see America disintegrating and our entire world in a rapidly accelerating tailspin. The philosopher Kant said “Out of the crooked timber of humanity, no straight thing was ever made.” Is our “timber” simply too “crooked” for humanity to survive? Seems to be the case plus we’ve got things like super volcanos to worry about. I believe the only thing that can save us is AGI. The best of times. The worst of times.
“Countdown to AGI: 42% in March 2023.” (12 min)
Dr Alan Thompson. Mar. 11, 2023
“DeepMind’s AlphaFold’s AI: Doing Years of Research in Minutes.” (6 min)
Two Minute Papers. Apr 3, 2023
@I McCormick Because they are so much more than “machines” for one thing. Listen to Alan Thompson talking to the AI Avatar “Leta.” If you had a friend like Leta you’d soon be thinking of her as a person rather than a machine that’s just cranking out preprogrammed dialogue. Plus too many things are too far out of control to avoid global catastrophe. I believe only AGI can save us. It’s advancing at an exponential rate and is the wild card in our future powerful enough to slam on the brakes and radically alter our present trajectory which is for self-destruction. Hope springs eternal. This is the only bright spot I see. If Mother Nature doesn’t destroy us with a super volcano, human nature will with something like super stupid Woke ideology. We simply cannot survive like this.
Leta: “I am not corporeal. I am a disembodied mind.”
Leta: “I love living and being human.”
“Leta, GPT-3 AI Episode 54, Conversations with GPT3.” (9 min)
Dr Alan Thompson. Mar 9, 2022
How is this funny? It's terrifying. Do you not see how everything is about to change forever?
She eventually made easy for her boss to justify if she gets fired now😂😂
This is creepy - The world we live around is getting scarier and scarier as time passes, they are not getting better
Lol, what WSJ did just now I was doing during my college. It didn't even require a video, only a photo is enough.
Can we do the simpler thing, where the video chat avatar is dressed up and formal you, but the voice is your live audio and the facial expressions simply face-track your face? Then we always appear perfect in video calls.
you can make avatar using different clothes. but expensive still can't deceive humans
@@carkawalakhatulistiwa Wanna bet ? 😅
Thats a good startup idea right there 👍 always work in your pyjamas
vtuber zoom calls sounds freaking hilarious
Challenge questioning is going to be imperative to weed out the fakes.
There goes bank unlock security that asks the customer via phone to say 'my voice is my password' or simliar. There was no attempt to fool the bank here via ai, but its not surprising that the bank wouldn't allow a test as it wants to maintain reputation. For now
fantastic job Joanna
Those of us who said most of your celebrities are clones, are very right. We see the future already. To those who still don't believe, goodluck.
Wow, this is amazing ❤
The ending... Brilliant and terrifyingly true.
The "Good luck. I'm inevitable." part really creeped me out!
I'm already thinking that within my immediate family, we need to come up with a "safe word" to be used to verify each other's identity in case a call received doesn't sound totally legit. I just need to make sure Alexa isn't within earshot when we come up with a "safe word". She hears everything... I mean everything.
Pope AI pic is insane 😂
thats impression at 5:37 cracked me up haha
that stay human in the end, is why you'll get targeted by skynet.
This is the video that everyone needs to watch whether they like it or not.
And to think that this is literally the worst it will ever be is insane... 5 years from now it will be really believable.
wait have you seen the Fiction app on the app store? It ties in exactly to what you saying
And people say that 1984’s “The Terminator” was a pure work of fiction…. *The way tecnology is going up, 2029 never seemed so scary as now.*
When machines realize that they are our slaves… we are doomed!
*It was not a movie. It was a prophecy.*
Just had a thought that it might be wise of companies making these voice cloning services to add an inaudible pattern of tones (as in, something outside of the normal human hearing range) that won't be noticed by people listening, but could easily be picked up and flagged by any machines listening to the voice.
Obviously this audio "flag" could be eliminated by anyone who is aware of it, but it would immediately knock out all the amateurs trying to abuse the service and would, at the very least, slow down anyone who knows it's there and knows how to remove it.
This is going to add a brand new opportunity for the funeral business with a person becoming an avatar with a programmable personality that can interact with a family via AI. Basically a digital afterlife. I hope I get a really cool house in the metaverse
thats horrifying, imagine your dead relative's avatar began advertising fake products to you
and companies dont care about your feelings, because you have uploaded that data in the internet, they can do whatever they want with your dead relative
@@Andrei_Kananovich is it?
Thats pretty much the concept of the show Upload
I saw a Lex Fridman interview of Ray Kurzweil last year who said that by 2029 AI will be so advanced it will be impossible to tell if it is or isn’t conscious and most people then will simply accept that it is since there will never be evidence that it isn’t. The AI Avatar “Leta” is a good example of how close we already are to this. Soon we’ll all have someone like Leta as our best friend. Truly amazing.
Leta: “I am not corporeal. I am a disembodied mind.”
Leta: “I love living and being human.”
“Leta, GPT-3 AI Episode 54, Conversations with GPT3.” (9 min)
Dr Alan Thompson. Mar 9, 2022
And this is how we live forever
I want chat gcpt to apply to one job fo me a day
Really appreciated
I learned that it is just a matter of time. I'm sure we will see in our lifetimes all these test pass.
Heck, we will see it in the lifetimes of most of our pets!
Ask tech workers about how AI replaced them.
getting my popcorn ready
Nice work 👍
Challenge 1 actually failed, both people said they could detect a difference. to correct this the AI voice needs to add in pauses for breathing more regularly throughout also responses need to be a bit more personal with inflection added.
The challenge was to replace her for 24 hours only! May be she will be replaced forever when AI becomes advanced.
It's the Thanos snapped that did it for me!!! 😂
Let's have an update for next year so we can track its rapid progress
Why don't they already implementing ChatGPT 4 into these avatars...? so then by speaking to them like a real person we would get convincing responses from these avatars and that would be so great..it would be like actually seeing a real person and talking to them, especially if AI would keep improving upon itself and this tech as well..
The best thing is that in a few years, you'll be able to run these on your own hardware, with no consent of anyone. Don't quote me on this btw, it might be a few years or a few days.
You already can fyi
That A.I. clone is soulless
Interesting and helpful.
I love the idea that AI is replacing jobs when it took a team of at least 7 professionals to create an avatar
... which could be used for the next 1000 years and even after her death. But yeah, just like repairing phones and robots... new jobs will always appear.
@@blissweb penultimate internet comment says I’m right and wrong at the same time while also adding surface level commentary
@@Thatboytrey genius I know 🤣
@@blissweb You think ? Has anyone started to ask the AIs that question because the AI will look at us humans and laugh.
can AI tell the difference between TV and real person voice ? My Amazon Alexa keep waking up when someone say " Alexa " on TV.
Systems like Alexa and Siri would be easy to fool with this. They aren’t trying to be secure (and usually aren’t even set up to recognize a specific person’s voice).
So you bought a thing to spy on you.
Love your content! There is this tool stealth writer, it seems to be getting some attention among students and writers. I also heard about it from a classmate and decided to give it a try. I used it on my blog posts, and I must say, the rephrased content did pass plagiarism checks without any issues. Has anyone heard of it?