The Daily: AI in 2024—Multimodal chatbots, the copyright battlefield, and how AI might influence the election

On today's podcast episode, we discuss the impact of multimodal chatbots, the copyright case between The New York Times and OpenAI, and how AI could influence the election. "In Other News," we talk about how much Microsoft's new AI keyboard key might affect AI use. Tune in to the discussion with our analyst Yory Wurmser.

Subscribe to the “Behind the Numbers” podcast on Apple Podcasts, Spotify, Pandora, Stitcher, YouTube, Podbean or wherever you listen to podcasts. Follow us on Instagram

Episode Transcript:

Yory Wurmser:

... and then if you tack on media partnerships, content agreements, it really does tilt the business case for these companies in a way that's going to leave a lot of losers and maybe fewer winners into battle.

Marcus Johnson:

Hey, gang, it's Thursday, January 25th, Yory and listeners, welcome to the Behind the Numbers Daily: an eMarketer podcast. I'm Marcus. Today I'm joined by our principal analyst who heads up our media advertising and technology desks based in New Jersey. It's Yory Wormser.

Yory Wurmser:

Hey, Marcus, how are you?

Marcus Johnson:

Hey, fella. Very good. How are you?

Yory Wurmser:

I'm doing great.

Marcus Johnson:

Perfect. So today's fact. Bananas are a little bit radioactive. You knew this, didn't you? You've known this? You knew this?

Yory Wurmser:

I've been on a banana fix the last few weeks for some reason, and I did know it was radioactive.

Marcus Johnson:

You're feeling ill. Is that why you...

Yory Wurmser:

I just turned off the lights and checked if I was glowing and I wasn't, so I was good.

Marcus Johnson:

So luckily they're not too radioactive. According to the US Environmental Protection Agency, bananas have naturally high levels of potassium and a small fraction of all potassium is radioactive, but you would need to eat about a hundred bananas to receive the same amount of radiation exposure as you get each day in the US from natural radiation in the environment. So do we find Yory... How many are you eating?

Yory Wurmser:

Not that many.

Marcus Johnson:

A questionable amount? Okay.

Yory Wurmser:

Not more than a few dozen. No, it's just-

Marcus Johnson:

A day?

Yory Wurmser:

It's probably about two a day.

Marcus Johnson:

All right. You might want to see your doctor. Why are bananas curved, Yory? Do you know this one too?

Yory Wurmser:

No idea about that.

Marcus Johnson:

Okay. Because of a phenomenon called negative geotropism, instead of continuing to grow toward the ground, they start to turn towards the sun. The fruit grows against gravity, giving the banana its familiar curved shape. I think the listeners can tell we're running out of interesting things when you've reached why bananas are curved slightly. Let's get on with the show before listeners stop listening. Today's real topic, AI in 2024.

In today's episode, first in the lead, we'll cover the most important things to pay attention to when it comes to artificial intelligence in 2024. Then for another news, we'll discuss how much AI keys on keyboards we'll move the needle for AI. So, Yory, we're going to be talking about a number of different focal points expectations surrounding artificial intelligence in the year of 2024. We'll go through multimodal chatbots copyright issues, what's going on with OpenAI, who seems to be leading the charge for artificial intelligence, and a little bit about elections as well.

So we'll start at the beginning with multimodal chatbots. So Cade Metz of the New York Times notes that AI companies are building multimodal systems, meaning AI can handle multiple types of media. For example, OpenAI's latest version of ChatGPT can both generate images and text. The next stage of evolution would be systems inhaling diagrams, charts, sounds, and video to then produce their own text images and sounds. What are your thoughts on the impact of multimodal chatbots in 2024?

Yory Wurmser:

I think it's huge and I think they're going to become, especially towards the end of the year, become really integrated in our workflows. The reason they're really important is a lot of the way we deal with images right now and text is kind of one after the other. The program interprets it, takes the language, sends it to an image generator, and then image generator would send it back something an image or send it back to the text generator to create text.

When you start merging it all together, it just creates new use cases. Things like, I mean, I've seen this, you point your phone towards a parking sign to see whether it's okay to park, and it'll tell you whether it is or not, but much more sophisticated things like you mentioned, generating charts from data or doing an interpretation of a picture or things like that. It just, the use cases proliferate once you sort of put it all in one model.

Marcus Johnson:

Okay. Okay. So do you think most models will be multimodal or do you think a fair amount will, but you'll still see some single use cases?

Yory Wurmser:

Yeah, I mean, you're not going to see the single use case disappear, but I think when you're talking about chatbots, I think that the flagship chatbots out there are going to be multimodal just because the things you can do will become a lot more powerful that way. But I think there still will be some single use case models that are purely language. For instance, if you have something that's optimized for customer service that might just make sense to make it as powerful and as quick as possible.

Marcus Johnson:

Right. Good point.

Yory Wurmser:

Just because the smaller the model is, the faster it works, the less compute power, the cheaper it is to use. So you'll see both, but definitely for the top line, I think they're mostly going multimodal.

Marcus Johnson:

Okay. A second trend or thing to watch, we should probably say, we wanted to talk about was copyrights and questioning whether copyright will be artificial intelligence's 2024 battlefield because Morgan Morone of Axio certainly thinks so. She points to the New York Times is filing a lawsuit against OpenAI and Microsoft in December claiming their AI systems wide scale copying constitutes copyright infringement. OpenAI in response says, "Mass scraping of the internet falls under fair use rules." Yory, do you see copyright being AI's 2024 battlefield?

Yory Wurmser:

Yeah, for sure. And two reasons. One is what the New York Times is doing is they have all this content that has been scraped. They want to get a share of it. I think a lot of other big media companies are going to do the same thing. Either that or use it as leverage to get a better agreement with OpenAI, Google, and other large AI companies for using their data. So it'll either be used as leverage or as the basis for a suit, so absolutely, I think it's going to be a big battleground.

Marcus Johnson:

Yeah, it's such an important case because if the New York Times wins, yeah, it could embolden others. Our senior director of briefings, Jeremy Goldman, noting that the Times lawsuit can inspire others, other actions from musicians, authors, comedians against AI setting industry-wide precedents for copyright. It's going to be interesting to see where folks land on this, Yory, and a big part of this could be how this case shakes out because which side of the debate will most publishers land?

The Associated Press and our parent company, Axel Springer, are they going to land on that side? That's the side that have struck individual deals with AI companies saying you can pay us for our data and information, or are they going to fall on the New York Times side basically saying, no, you can't and we're going to sue you if you try?

Yory Wurmser:

Yeah, and I think you're also going to see the AI companies try to preempt these suits with links or these deals with these companies. And I read somewhere, I'm not sure where, but that the big AI companies are going to have an inherent advantage because they have the deep pockets to both fight the lawsuits and to come up with these deals with the content companies. So there is a danger that it's just going to lock in some of the big players at the expense of some of the more scrappy startups.

Marcus Johnson:

Yeah, Yeah. University of Miami, professor of law, Andres Sawicki of Axios saying, "If the courts determined that training the systems constitutes infringement, they thought that it was going to have a really significant impact," and this lawsuit could set OpenAI back quite a bit. The information was pointing out OpenAI's 2024 revenue expected to be about $5 billion and this lawsuit could jeopardize that. Right?

Because The Times they've not asked for a specific amount in damages, but the lawsuit says, "It seeks to hold them responsible," The New York Times, "seeks to hold OpenAI responsible for billions of dollars in statutory and actual damages." So we don't know exactly what that means in terms of how much money, but if they do have to pay billions of dollars in fines, that could be pretty devastating.

Yory Wurmser:

Yeah, I mean it's such a capital intensive business already, AI is. You need a ton of compute power to both train and to run these programs, and then if you tack on media partnerships, content agreements, it just, it really does tilt the business case for these companies in a way that's going to leave a lot of losers and maybe fewer winners in the battle.

Marcus Johnson:

I was talking to someone about this episodes and what we were going to be talking about in the show today, and they asked me, Yory, they said, "Who wins this?" And a lot of the time you can roughly guess whether who's going to win on a privacy case or an antitrust. Do you have a sense of how things are going to shake out? Do you have any idea of New York Times going up against OpenAI for copyright infringement? Any sense of how this plays out?

Yory Wurmser:

Well, I mean, I'll preface this by saying I'm not a legal analyst, but I mean it's a really tricky question whether this stuff is fair use, this content is fair use. So if New York Times is breaking news, it's news that's happening out there. Is what President Biden says fair use? Yeah, of course. If you're scraping that from a New York Times article, is that fair use? Probably. What is the language on top of that? That's fair use. And that's the tricky case and I don't know where the courts are going to fall.

Marcus Johnson:

Doing this on a mass scale feels like it falls outside the realm of fair use. I'm fascinated to learn more about fair use, and I'm sure more will come out in this case is the lawyers kind of dig through what that actually means. But using five seconds of a song on a video that you're putting together to make it a better piece of content seems different from scraping the entire internet and hoovering up as much information as possible to then create a product that you can then make money from.

Yory Wurmser:

For sure. And some things are cleared and others if an artist has a specific style and then you ask an AI program to create another artwork of that style-

Marcus Johnson:

Yes.

Yory Wurmser:

I think that probably is clearly not fair use, that you're infringing on copyrights there, but I think the questions that we're talking about are a lot more grayer.

Marcus Johnson:

Yep, yep. Let's move to our third thing to watch for artificial intelligence in 2024, and we're talking about OpenAI, this company that does seem to be at the forefront of artificial intelligence development. ChatGPT maker OpenAI, they weathered an epic leadership crisis nearly two months ago, but enter 2024 beset with uncertainties, writes Scott Rosenberg of Axios.

Two of those are one, as we kind of talked about making money, a former employee of OpenAI once describing the company as a money incinerator. And two, the remaining question about what spooked board members back in November so much that they ousted for a week, CEO, Sam Altman, the story there being that the board thought Sam Altman wasn't being transparent enough and they were potentially concerned about this advancement that OpenAI had discovered and they were a bit nervous about the pace of development, and so that's why you had this whole debacle around leadership at OpenAI. We still don't know the actual truth. There are independent investigations going on, but given that that happened pretty recently, Yory, what do you think's next for Sam Altman and OpenAI?

Yory Wurmser:

Yeah, it's an open question what led to that, that board revolt. I think the bigger question is how much of an unfettered business is OpenAI? It was originally formed as a nonprofit and then it's converted to a capped profit corporation. I think that's gone now too, but it still needs to figure out not only how hard it wants to go into monetization, and I think as you mentioned, it does burn through cash.

So it's essentially forced as a company to raise huge amounts of money and to sell a lot of different products tapped on top of it. So it has to decide that, but also this whole board revolt gave competitors a big in. Companies no longer want to rely solely on OpenAI, and that's a question that, that's a problem now that OpenAI has to face that they've opened a door for competitors to get in.

Marcus Johnson:

Yes. And because of that, are we certain OpenAI will keep its lead? Mr. Rosenberg thinks OpenAI that kicked off and still leads today's AI boom remains an enigma. It could be the next Google or the next Netscape. He recounts the Netscape 1995 IPO ushered in the internet era, but by 1998, 3 years later, the browser pioneer was on the ropes. The next year in 1999 it got bought by AOL, which crashed and burned, leaving Google and Facebook to harvest the web's profits. So even because it's the forefront right now, doesn't mean that it's going to be the front-runner forever.

Final thing we want to talk about, Yory, in the lead with regards to expectations for AI in 2024 was the election or the elections coming up. Sarah Fisher of Axios noting that ChatGPT maker OpenAI says, "It's rolling out new policies and tools meant to combat misinformation and abuse ahead of the 2024 elections in the US and worldwide."

Some of those policies and tools, they wanted real-time news reporting including attribution and links. They're working with non-partisan National Association of Secretaries of State to direct ChatGPT users to canIvote.org for authoritative information on US voting and elections. And then other things like not letting folks build applications for political campaigning and lobbying amongst some other things. What do you make of OpenAI's election safeguards?

Yory Wurmser:

First of all, they're incredibly important. You already saw it a week ago or so with the fake robocall using Biden's voice in New Hampshire. These created, these synthesized media, could be huge sources of disinformation, so that's one thing.

Marcus Johnson:

Especially at the wrong time. There was an Atlantic article from Ross Anderson and in the article they were basically saying, imagine if there is a candidate, there's two people left to run for president and one of those candidate's voice is cloned and there's audio of that person criticizing a demographic of people and that's days before the election? That can swing things so quickly, especially when we've seen elections hinge on hundreds of thousands of votes.

Yory Wurmser:

100%. So there's that angle of the synthesized incorrect sort of manipulated information and then there's just these inherent biases within these programs, and I think that's one of the big focuses or foci, I don't know what the word is, for OpenAI, is to make sure that there aren't these biases towards one side or the other, relying on ways, on third party sources to verify information.

Marcus Johnson:

Yep.

Yory Wurmser:

So they have to be really careful with that. And I know they also have some program giving grants for democratic oversight, the democratization of building AI models, which is really important as well.

Marcus Johnson:

Yep. A huge election year, Cole Yu of Time was pointing out, "Globally there are more voters than ever in history that will head to the polls as at least 64 countries, plus European Union representing a combined population of about close to half of the people in the world meant to hold national elections." So you've got US, India, Indonesia, Mexico, et cetera. Now not all of those elections are free and fair, you could argue, but of the ones that are, you've got close to 2 billion people voting this year. So it's not just US elections, it's elections around the world. More people voting have the potential to vote than ever before, and this is the first election cycle where GenAI tools are widely available to folks.

I'll close the lead by saying a quick piece from the economist, Yory, because I think it sums things up quite well, talking about how AI development will be both fast and slow. The article was arguing that many economists believe GenAI is about to transform the global economy. It was a paper published last year by Edgar [inaudible 00:15:57] and Tamay [inaudible 00:20:07] of research company Epoch arguing that, "Explosive growth with GDP zooming upwards is plausible with AI capable of broadly substituting for human labor."

But professor at Stanford University saying that he expects AI to power a productivity boom in the coming years, basically saying the change is going to take time. We won't realize everything in 2024. There was a Goldman Sachs analysis suggesting 5% of chief executives expect AI to have a significant impact on their business within one to two years, just 5%, but 65%, two thirds think it will have an impact in the next three to five, AI likely to change the economy even if it will not do so immediately.

Yory Wurmser:

Yeah, different parts of the economy are going to be hit earlier rather than later.

Marcus Johnson:

Yes.

Yory Wurmser:

Translation services, customer service, I think you're going to see big disruptions there on the earlier side, more creative roles and things like that. Right now they're more productivity tools rather than replacement for humans. I'm not even sure eventually that they will become replacement for jobs, but we'll see. I agree that it's going to come more gradually. The impact on the workforce is going to come more gradually, but the productivity increases are going to come pretty quickly.

Marcus Johnson:

Yeah, I almost want to throw out the year of AI and just talk about it in waves. It's going to be from now into the future and it's going to come in waves and this year we're going to deal with different wave than we will next year. That's all we've got time for, the lead. Time now for the second half of the show. Today in other news, just one story. Microsoft's new AI keyboard key.

Speaker 3:

Power back one.

Marcus Johnson:

Story one. Microsoft's got a new AI keyboard key. They've announced the biggest change to its keyboards in 30 years with the introduction of an artificial intelligence or AI key on new products from February that lets users access Copilot, which is Microsoft's AI tool on Windows 11 PCs. As Imran Rahman-Jones of the BBC notes, "Copilot helps users with functions like search, writing emails, summarizing meetings, and creating images. Copilot is based on OpenAI's GPT-4 large language model for context. Windows 11 users can already access Copilot by pressing the Windows key plus the C key, but this change gives them a dedicated button to access AI." Yory, how much does this new dedicated AI key move the needle?

Yory Wurmser:

I don't think it moves it much at all. I think it's more of a-

Marcus Johnson:

Really?

Yory Wurmser:

... gimmick.

Marcus Johnson:

Wow.

Yory Wurmser:

I mean, I actually think Copilot is, and similar productivity tools are going to be pretty transformative, but adding it to the keyboard, I don't see making it much of a difference. I mean, one of the advantages of AI is there are many ways to launch it. Maybe you can just say, use your voice or something like that. I think that this is a very short-lived or gimmicky type of introduction.

Marcus Johnson:

Interesting. Would you compare it to the Netflix key on a remote control? Does that really make people use Netflix more? Is it more seen as marketing tool because I'm reminded of Netflix, I'm going to watch Netflix, I'm going to watch it regardless of how I get there.

Yory Wurmser:

So Netflix I think is a little different in that it, I mean, it does simplify remote a little bit more than this, but yeah, I don't think it's a big deal.

Marcus Johnson:

Okay. Well, I mean to that point, there are other helpful dedicated buttons already that exist. Mr. Raman Jones of the BBC pointing out that rival Apple has included a Siri button or option on its touch bar in MacBooks for a few years. And yeah, I don't see personally a ton of people using it. Maybe they haven't had a reason to use it until now, but it doesn't seem like it's moved the needle much thus far, at least. That is all we have time for this episode. Thank you so much, Yory, for hanging out today.

Yory Wurmser:

Yeah, pleasure as always.

Marcus Johnson:

Yes, sir. Pleasure's all mine. Thank you to Victoria who edits the show, James Stewart and Sophie, who also help make the podcast possible, and thanks to everyone for listening in. We hope to see you tomorrow for the Behind the Numbers weekly listen and eMarketer video podcast. If you do want to watch that episode, which you can now do, you can head to YouTube at InsiderIntelligence, one word, slash podcasts, or you can listen to us the normal way on the normal places.