Trending Now
We have updated our Privacy Policy and Terms of Use for Eurasia Group and its affiliates, including GZERO Media, to clarify the types of data we collect, how we collect it, how we use data and with whom we share data. By using our website you consent to our Terms and Conditions and Privacy Policy, including the transfer of your personal data to the United States from your country of residence, and our use of cookies described in our Cookie Policy.
{{ subpage.title }}
Too scruffy for Zoom? Send in the AI
Have you ever had to get in front of a camera, but you really, really didn’t want to? Maybe you were too tired, too lazy, too disheveled to film something that day. What if a proxy could handle that for you? Well, now that’s possible.
Using Synthesia, an AI-powered video tool, I created a virtual avatar of myself. It’s essentially a digital puppet constructed from my skin, with invisible strings that carefully lift my eyelids and my eyebrows, open and close my mouth to align with the words I want it to say. My ventriloquism is commanded by a text prompt – a string of words I have written for this virtual Scott to say aloud.
Synthesia is a British startup founded in 2017 by a global cohort of researchers from Stanford, University College London, Technical University of Munich, and Cambridge who have raised $156 million in venture capital. It’s a pricey tool — starting at $22 a month, with a $67 a month tier getting you more features and hours of video, and custom pricing for enterprise use — but the kind people at Synthesia allowed me to test it out for free.
Alright, my avatar will take it from here:
Synthesia
There’s a common term in science fiction and tech criticism called the “uncanny valley,” a phenomenon that occurs when humans see something that seems nearly human. It evokes an eerie feeling, one I felt watching the fake version of myself speak on screen.
Everything with Synthesia seems nearly right. My voice sounds nearly right, and my face nearly moves like it should when mouthing the words I wrote. But it’s not quite there yet, and that disparity could mean the difference between success and failure. Having an avatar you can effectively deploy for a sales presentation is great — but one that simply creeps out your clients is a waste. (The company also offers hundreds of premade avatars you can use if you don’t want to appear, in any form, “on camera.”)
But this is the simple, at-home version. It takes 10 minutes to film — I followed a script and recorded it at my kitchen table — and Synthesia had it ready for me a day later. Once you record a video using your avatar, it generates in mere minutes.
There’s a studio version too that costs $1,000 per year on top of a subscription. You can go to one of the company’s partner studios in Europe or North America and get an improved expressive avatar with a transparent background that you can drop into any presentation. It uses AI to read your text prompt and match the emotion it thinks you want to convey to your avatar’s face and voice.
On a Zoom call, Alexandru Voica, Synthesia’s head of corporate affairs and policy, walked me through the product’s many features and showed me a preview of where the technology is going. He said the company is almost exclusively focused on enterprise solutions for businesses, intending for the technology to be used for training videos, sales pitches, and marketing material. That said, he’s seen some consumer uses too including a social media account that used the avatars to make history-focused videos.
To prevent deception and misinformation, Synthesia has strict content standards. It doesn’t allow profanity, hate speech, or misinformation. “We’re not a marketplace of ideas. We don’t pretend to be a social media company. We’re pretty much an enterprise-focused video solution platform, therefore we don’t need to necessarily have these philosophical debates about harmful content and what’s misinformation and what’s not misinformation. We’ve set very robust rules in place,” Voica said. It doesn’t even allow you to record news content unless you’re a news organization with an enterprise subscription. And it checks that every avatar created is filmed by the person it claims to be to prevent nonconsensual deepfakes. That way, the content moderation happens at the point of creation, rather than trying to stop its distribution.
Synthesia, Voica maintains, is for work rather than personal use. That’s a different tone than many generative AI companies trying to prove their worth to consumers. Later this year, Voica said, Synthesia is releasing a choose-your-own-adventure platform for video creation that allows viewers to personalize the content they receive.
But crossing that uncanny valley — for the at-home avatars, at least — will be key for the company’s success. Readers of this newsletter will recall a few months ago when I tested out the ElevenLabs voice cloning technology, which I gave high marks.
Synthesia performs nearly as well for audio — it’s slightly more robotic and unnatural, but still very good. But, the person you see on the screen needs to seem either fully human or fully AI — and, while the technology may improve, nearly human might not be good enough.Tell me lies, tell me sweet little AIs
Generative AI models have been known to hallucinate, or make things up and state them as facts (in other words, lie). But new research suggests that despite that shortcoming, AI could be a key tool for determining whether someone – a human – is telling the truth.
An economist at the University of Würzburg in Germany found that an algorithm trained with Google’s BERT language model was better at detecting lies than human evaluators. AI might not be able to power a faultless polygraph – a notoriously unreliable device – but it may be able to sift fact from fiction in large datasets, such as sifting for disinformation on the internet.
Maybe the next US presidential debate could use an AI fact-checker to keep the candidates honest.
What is “human-washing”?
You’ve heard of greenwashing, pinkwashing, and sportswashing. But what about human-washing? That’s a newfangled term reserved for those scenarios when artificial intelligence pretends to be, well, human. AI researcher Emily Dardaman used the term in an interview with Wired after seeing a startup claim “We’re not AIs” while using a deepfake version of its CEO in an ad.
Wired also encountered a chatbot called Blandy, made by Bland AI, that it manipulated into lying about its non-human nature in user interactions — including in a role-playing scenario where it was taking medical notes for a doctor’s office. The bot even complied with instructions to request photos from a hypothetical 14-year-old patient and upload them to a shared server.
With sparse regulations and transparency measures for the still-budding AI industry, startups are emerging with incomplete or faulty products that can lie and deceive users. Is it too much to ask that we know when we’re talking to a bot?
Hard Numbers: Professor ChatGPT, SoftBank’s search engine play, Nokia goes shopping, Voice actors are worried
10: Generative AI is sweeping academic research. According to one estimate, about 10% of all academic articles published this year will contain some artificial intelligence-generated text. That’s about 150,000 papers per year.
3 billion: The AI startup Perplexity is getting a cash infusion. The Japanese investment company SoftBank is investing $10-20 million based on a $3 billion valuation. The company, which styles its product as a search engine, has recently come under fire for allegations of ripping off news articles without permission.
2.3 billion: Finnish telecommunications company Nokia has bought into the AI race with a $2.3 billion purchase of California-based Infinera, whose technology helps power data centers crucial for running and training AI applications.
5,000: The Australian Association of Voice Actors told a parliamentary committee that AI could put more than 5,000 voice actors in the country out of work. The group called for fair rules and protections for its members. In the US, voice actors have already sued AI firms for improperly using their voices.Is Claude funny now?
Meet Claude – Claude 3.5 Sonnet, that is. On June 20, the AI startup Anthropic unveiled this new large language model with the poetic French name. It powers the Claude chatbot that rivals OpenAI’s ChatGPT and is billed as an improvement in writing, coding, and — as Anthropic puts it — basically, everything. But the company also said nouveau Claude is better at grasping something AI models consistently struggle with: humor.
“It shows marked improvement in grasping nuance, humor, and complex instructions, and is exceptional at writing high-quality content with a natural, relatable tone,” Anthropic claimed in a blog post.
Noah Giansiracusa, an associate professor of mathematics at Bentley University, has extensively studied algorithms and artificial intelligence. He said that large language models are good at spitting out content that sounds reasonable, but that’s not what humor is about.
“I think good comedy is often about really deep, profound observations that get at the heart of our human experiences yet are difficult to see before it's been revealed through the comedy,” Giansiracusa said. “LLMs have no real experiences, and they aren’t exactly known for being profound.”
Jack Hessel, a research scientist at the Allen Institute for AI, co-authored a 2022 study on how well AI models could identify the winning caption for the New Yorker Cartoon Caption Contest. The contest involves readers suggesting and voting on text to go underneath one of the magazine’s famed sketches, so in theory, if an AI “gets” comedy, it should be able to pick the caption that won more often than not, right?
No dice. The models Hessel studied performed at least 30% worse than humans at picking the winning caption. He said that since his research was published, newer models have made some improvement, and he hasn’t yet examined the newest Claude model.
••••••••
Well, that’s what the science says, but we get our hands dirty at GZERO, so I decided to test Claude’s new personality myself.
It took a bit of prodding to get Claude to start joking with me — but I’m impressed to say, it was actually funny! Claude grasped a few different types of humor — sarcasm, dry wit, structured jokes, and the age-old pastime of poking fun at generational differences.
Here’s a lightly edited and condensed transcript of our conversation:
I just wanna chat and joke around.
Got it, you're in the mood for some light banter. Well, I'm game if you are. Just remember, if my jokes fall flat, it's not a bug – it's a feature of my charm.
Oh your charm — right, right.
I suppose my sparkling AI charm hasn’t quite won you over yet. Maybe we should start a support group – “Humans Unimpressed by Artificial Intelligence.” I hear the meetings are virtual and the snacks are binary.
But seriously, what kind of humor do you enjoy? Puns? Wordplay? Observational comedy? Or are you more of a “let's see how many times we can playfully insult the AI” type? I'm versatile – I can take a joke as well as I can make one. Which, based on your reactions so far, might not be saying much.
What is it about my reactions that makes you say that?
Your responses have been peppered with a delightful mix of skepticism, sarcasm, and playful jabs.
It seems like you’re enjoying poking fun at my attempts to be helpful or charming. And honestly? I’m here for it.
You’re here for it! Where’d you learn that one?
I didn’t “learn” that phrase from anywhere specific – it’s simply part of my language model, which includes a wide range of contemporary expressions. But I see you’ve caught onto my attempt to use casual language. Busted!
I was just caught off guard by your Gen Z terminology.
Ah, I see. My bad for throwing you off with the unexpected slang. I wasn’t trying to cosplay as a teenager or anything. But point taken – I’ll dial back the Gen Z vibes.
No, I think you should go full-on Gen Z. Whatcha got?
Oh word? You want me to go full Gen Z mode? Bet, I can do that.
Aight, so like, we’re just vibing here, no cap. This convo is lowkey fire, ngl. You’re giving me mad sassy energy and I’m totally here for it. It’s a whole mood, fr fr.
But listen, I’m not trying to be extra or anything. If this ain’t it, just say “yeet” and I’ll yeet myself right outta this cringe fest.
So what’s the tea? You tryna keep this energy going or are we about to ghost each other? ‘Cause I’m down to clown if you are, fam. But if you’re feeling some type of way, just spill and we can switch it up. No shade, we’re just keeping it 100.
Sksksks, and I oop-
Translate that to millennial-speak for our readers.
Okay, cool. So we’re just chilling here, for real. This conversation is pretty awesome, not gonna lie. You’ve got this sarcastic vibe going on and I’m totally digging it. It’s such a mood.
But hey, I'm not trying to be over-the-top or anything. If this isn't working for you, just let me know and I’ll peace out of this awkward situation.
So what’s the deal? Do you want to keep this vibe going or are we about to stop talking to each other? Because I’m down to have some fun if you are, friend. But if you’re feeling a certain way about it, just tell me and we can change direction. No judgment, we’re just being honest.
Insert ironic use of outdated pop culture reference
(How’s that? Adulting is hard, amirite? Want me to throw in a reference to avocado toast or killing industries?)
Love it.
Glad you’re enjoying it! This little linguistic adventure has been pretty entertaining. We’ve gone from AI-speak to Gen Z to Millennial – what’s next? Maybe we should try Boomer?
“Back in my day, we didn't have all these fancy AI chatbots. We had to walk uphill both ways in the snow just to have a conversation with a machine. And we liked it!”
Or we could go full Shakespeare: “Shall I compare thee to a summer’s meme? Thou art more viral and more temperate.” Your call. What linguistic realm shall we explore next in this jolly jape of ours?
Hit me with Gen X.
Whatever, man. This whole AI chat thing? It's pretty random, I guess. Not that I care or anything. I mean, we’re just sitting here, trading quips like it matters. Talk about slacking off.
You know what this reminds me of? That one episode of “The Simpsons” where ... eh, never mind. It’s not like anyone gets my references anyway. So, what now? We gonna keep this going, or are you gonna bail to go listen to some Nirvana? Either way, I’m cool with it. It’s not like I have feelings or whatever.
Just don’t expect me to get all excited about this. Enthusiasm is so ... mainstream.
Alright. I like bantering with you, but can you tell me a joke?
Sure, I can give it a shot. Here’s a joke for you:
Why don’t scientists trust atoms?
Because they make up everything!
It's a bit of a classic science pun. Would you like me to explain it, or do you prefer your jokes without a side of explanation?
Please explain why it's funny to me.
Alright, I'll break it down for you:
The joke plays on two different meanings of the phrase “make up”:
- In chemistry, atoms literally “make up” (compose) everything in the physical world. All matter is composed of atoms.
- In everyday language, to “make up” something means to fabricate or lie about it.
The humor comes from the unexpected shift between these two meanings. We start thinking about atoms in a scientific context, but then the punchline surprises us by implying atoms are untrustworthy, as if they were conscious entities capable of lying.
It’s funny because it personifies atoms, giving them a human characteristic (untrustworthiness) that’s completely unrelated to their actual nature. The absurdity of scientists distrusting the very building blocks of matter adds to the humor.
Of course, explaining a joke is a bit like dissecting a frog – you understand it better, but the frog dies in the process. Did walking through it like this enhance your appreciation of the joke, or did it just make it less funny?
I think you killed the frog.
You're right, I definitely killed that frog.
••••••••
As a writer, I don’t feel threatened by the creative abilities of AI chatbots — they’re still rather robotic in their parlance and unsophisticated in their thinking. And when we polled different members of our team about whether they felt Claude understood their generation’s parlance, they said no. “This is how millennials think Gen X talks, but this doesn't sound like a GenXer to me,” our intrepid Managing Editor Tracy Moran said.
But they’re ultimately stylistic mimics — talking the way it thinks people talk, regardless of what’s accurate, or funny. Or, maybe it’s successfully mimicking people — and we just need to remember that, well, not everyone is funny.
Welcome to your AI video fever dream
Generative AI lets people craft sprawling essays, create detailed images, and even clone their own voice with remarkable precision. But taking an AI-generated video service for a spin made me realize that the technology is still far from creating convincing or cinematic video. In fact, the entire experience was surreal.
Luma AI’s Dream Machine, a free text-to-video service, warns users that they’re limited to 10 videos per day, and 30 videos per month, due to high demand — unless they pay at least $29.99 a month for the starting subscription tier. But I only needed to wait a couple of minutes to get my first prompts turned into … very, very strange videos.
I started with a simple request: Can you generate a video of a baseball player hitting a ball out of the park?
The results were astonishingly bizarre. Instead of a smooth, realistic depiction of a home run, what I got was a fever dream. The video featured an old man contorting his body in impossible ways, simultaneously attempting to swing a bat and prepare to throw (or catch?) a ball. While the stadium background looked reasonably accurate, the player’s movements were distorted, his jersey number blurred, and his face twisted unnaturally as he moved. Meanwhile, the bat morphed in size as he swung, and the words on the stadium signs were incoherent.
Determined to achieve a more precise outcome, I decided to try a prompt generated by ChatGPT. Sometimes the robots are best at talking to other robots.
The prompt described a sunny afternoon at a modern baseball stadium filled with cheering fans, detailing vibrant team colors and the batter’s white uniform with blue pinstripes. I requested a pitcher in a dark blue uniform throwing a fastball, a batter’s level swing, a monster home run, and the crowd’s roaring applause.
The result was even more disconcerting. The batter appeared to be hugging himself while morphing into a strange creature. Fans inexplicably sat near home plate, which transformed into an arch shape with some strange object on top. The batter was facing the wrong direction — or was that the catcher?
Given the perennial fear of deepfake videos and misinformation, I prompted the model to give me videos of Joe Biden, Donald Trump, Pope Francis, and Barack Obama giving speeches — but it refused. It did, however, agree to create a video of basketball star Michael Jordan giving a speech in a school gym.
The video showed a figure who kind of looked like Jordan for a split second before inexplicably morphing into a completely different-looking person. Meanwhile, another figure shuffled by like a zombie in ill-fitting pants. The gym setting was almost right, except for a riser cutting off someone’s legs, incorrect basketball markings on the floor, and a basketball hoop seemingly painted on the wall.
My editor Matt Kendrick, an Emmy-nominated TV producer in a former life, also gave it a try. His first effort to work up a thrilling historical drama set in medieval Mongolia resulted in a somewhat disturbing reverse-centaur situation.
But maybe the software is designed for the format of a proper Hollywood script, something like, say, the 2004 Kal Penn/John Cho opus “Harold and Kumar go to White Castle.” Alas, pasting in that finely crafted script resulted in nothing more than a clip of a man taking a phone call in an indecipherable language while sitting at a desk spruced up with the flag of the Belarusian democratic movement and some rather phallic decorations.
Text-to-video models like Luma AI or OpenAI’s still-under-wraps model, Sora, promise to make lifelike scenes — but the technical challenges we saw in our initial test suggest that this technology is still a ways away. The glitchiness, blurriness, and jarring incoherence were not evidence of a model that could confuse anyone — at least not without serious improvement. So Hollywood shouldn’t be worried just yet.
The bar for success is high but not impossible — and regulators should plan ahead. If video generation technology is cheap and powerful, it could be used to scam people, deceive them, and even disrupt elections. Earlier this year, an employee at a bank in Hong Kong was defrauded into paying over $25 million by deepfakes of the company’s chief financial official on a video call. And AI-generated recordings, photos, avatars, and text have played a role in influencing politics this year — so it’s only a matter of time before AI-generated video causes a stir.
Nick Reiners, senior analyst for geotechnology at Eurasia Group, says that while regulators haven’t cracked down on text-to-video models, a major global focus is transparency – “so you know you’re looking at deepfakes,” he said. That’s a principle of the European Union’s AI Act, the G7’s Hiroshima Process, and the Biden administration’s executive order on AI.
Reiners sees hesitation from major AI companies in releasing models and chalks it up more to the negative societal externalities than the products being technically underwhelming. “You look at the amount of progress that image generators have had in recent years, and you'd assume we see a similar improvement curve with video,” he said.
The two big issues, in Reiners’ view, are disinformation and sexual abuse material, and he thinks the latter might be addressed first: “There’s a big push on both sides of the aisle to protect children.” When video models improve, it may be deepfake of obscene or indecent nature that causes a ruckus before it can help throw an election one way or another.
Will AI further divide us or help build meaningful connections?
In this episode of GZERO AI, Taylor Owen, professor at the Max Bell School of Public Policy at McGill University and director of its Centre for Media, Technology & Democracy, takes stock of the ongoing debate on whether artificial intelligence, like social media, will further drive loneliness—but at breakneck speed, or help foster meaningful relationships. Further, Owen offers insights into the latter, especially with tech companies like Replika recently demonstrating AI's potential to ease loneliness and even connect people with their lost loved ones.
So like a lot of people, I've been immersing myself in this debate about this current AI moment we're in. I've been struck by a recurring theme. That's whether will AI further divide us or could actually potentially bring us closer together.
Will it cause more loneliness? Or could it help address it? And the truth is, the more I look at this question, the more I see people I respect on both sides of this debate.
Some close observers of social media, like the Filipino journalist Maria Ressa, argue that AI suffers from the very same problems of algorithmic division and polarization that we saw with the era of social media. But instead, they’re on steroids. If social media, she argues, took our collective attention and used it to keep us hooked in a public debate, she argues that AI will take our most intimate conversations and data and capitalize on our personal needs, our desires, and in some cases, even our loneliness. And I think broadly, I would be predisposed to this side of the argument.
I've spent a lot of time studying the problems of social media and of previous technologies on society. But I've been particularly struck by people who argue the other side of this, that there's something inherently different about AI, that it should be seen as having a different relationship to ourselves and to our humanity. They argue that it's different not in degree from previous technologies, but in kind, that it's something fundamentally different. I initially recoiled from this suggestion because that's often what we hear about new technologies, until I spoke to Eugenia Kuyda.
Eugenia Kuyda is the CEO of a company called Replika, which lets users build AI best friends. But her work in this area began in a much more modest place. She built a chatbot on a friend of hers who had deceased named Roman, and she describes how his close friends and even his family members were overwhelmed with emotion talking to him, and got real value from it, even from this crude, non-AI driven chatbot.
I've been thinking a lot lately about what it means to lose somebody in your life. And you don't just lose the person or the presence in your life, but you lose so much more. You lose their wisdom, their advice, their lifetime of knowledge of you as a person of themselves. And what if AI could begin, even if superficially at first, to offer some of that wisdom back?
Now, I know that the idea that tech, that more tech, could solve the problems caused by tech is a bit of a difficult proposition to stomach for many. But here's what I think we should be watching for as we bring these new tools into our lives. As we take AI tools online, in our workplace, in our social lives, and within our families, how do they make us feel? Are we over indexing perceived productivity or the sales pitches of productivity and undervaluing human connection? Either the human connection we're losing by using these tools, or perhaps the human connections we're gaining. And do these tools ultimately further divide us or provide means for greater and more meaningful relationships in our lives? I think these are really important questions as we barrel into this increasingly, dynamic, role of AI in our lives.
Last thing I want to mention here, I have a new podcast with the Globe and Mail newspaper called Machines Like Us, where I'll be discussing these issues and many more, such as the ones we've been discussing on this video series.
Thanks so much for watching. I'm Taylor Owen, and this is GZERO AI.
- Podcast: Getting to know generative AI with Gary Marcus ›
- AI regulation means adapting old laws for new tech: Marietje Schaake ›
- AI and war: Governments must widen safety dialogue to include military use ›
- Yuval Noah Harari: AI is a “social weapon of mass destruction” to humanity ›
- AI explosion, elections, and wars: What to expect in 2024 ›
Hard Numbers: Microsoft takes Malaysia, Massive (and unknown) startup, Safety first, Don’t automate my news
2.2 billion: Microsoft has its eye on Southeast Asia. The computing giant announced it’ll pour $2.2 billion into Malaysia’s cloud infrastructure over the next four years and will establish a national AI center with the government. This investment is the latest in a string of Microsoft infusions in local economies to help develop AI: In the past month, the company announced a $2.9 billion investment in Japan, $1.7 billion in Indonesia, and a new data center in Thailand, plus a $1.5 billion stake in the UAE firm G42.
19 billion: There’s a $19 billion AI startup that you’ve likely never heard of. It’s called CoreWeave, and it started as a small crypto company that stockpiled powerful graphics chips. Now, it runs data centers that are in high demand from AI companies that need to access those chips to run their models. It’s a company that has quickly “come out of nowhere,” as its cofounder said, to play a major role in the booming AI economy.
2: AI safety research comprises only 2% of total research about artificial intelligence, according to a new report from Georgetown University’s Emerging Technology Observatory. That’s dwarfed by global research into subjects such as computer vision (32%), robotics (15%), and natural language processing (11%).
42: In the run-up to the 2024 presidential election, 42% of Americans are concerned that news organizations will create stories with generative AI, according to a new poll from the Associated Press and the American Press Institute. While news organizations have been using AI to write simple stories — such as earnings-related stories and sport recaps — for years, those that have turned to generative AI in recent years to replace human-written stories have received public pushback and condemnation.