The Daily: How AI chatbots could change our lives and a possible AI agent revolution

Audio by Marcus Johnson, and Jacob Bourne | Oct 22, 2024

Episode Transcript:

Marcus Johnson (00:00):

This episode is brought to you by TikTok for Business. Vibe check, are you looking to slay your next campaign on TikTok? Of course, you are. Get ahead by measuring ahead to improve your campaign's performance. On TikTok, better measurement is the key to an improved ROI. Learn more from TikTok for Business at tiktok.com/business.

Jacob Bourne (00:23):

They found that people tended to ascribe human qualities to the chatbot. According to Harvard University economist David Deming, GenAI has had a faster adoption rate than the internet or personal computers. Now, that could include image generators too, but I think mostly we're talking about chatbots like ChatGPT.

Marcus Johnson (00:44):

Hey gang, it's Tuesday, October 22nd, Jacob and listeners, welcome to Behind the Numbers Daily, an eMarketer podcast made possible by TikTok. I'm Marcus. So I'm joined by our technology analyst based in California, it's Jacob Bourne.

Jacob Bourne (00:59):

Thanks for having me, Marcus.

Marcus Johnson (01:01):

Of course, [inaudible 00:01:02]. Thank you for being here. We start with the fact of the day, how many of each group of species are there? So this is according to the International Union for Conservation of Nature 2022 data. So Jacob, in the past, in past episodes, we've spoken about the fact that there are around two million identified species on the planet. Scientists estimate that there are close to nine million species on earth that are yet to be discovered, which is roughly 86% of land species and 91% of marine species yet to be discovered. We've talked about this one before, but what I want to know was what buckets do they fall into, these two million identified species. So about half, one million are insects. That's too many. We don't need that many is my point. And an extremely distant second are mollusks, so snails, slugs, squids, things like that. There are about 114,000 types of those. A fraction behind that, in third place, are arachnids with 110,000 different flavors of them. Does there need to be?

Jacob Bourne (02:18):

I'm assuming this is not taking microbes into consideration like bacteria, because I would think there'd be a lot-

Marcus Johnson (02:25):

Oh, good point.

Jacob Bourne (02:26):

... of those.

Marcus Johnson (02:27):

I don't think this does, no.

Jacob Bourne (02:28):

Yeah.

Marcus Johnson (02:29):

A hole in my research, Jacob. There are 80,000 types of crustaceans, 36,000 types of fish, 12,000 reptiles, 11,000 birds, 9,000 amphibians, 7,000 mammals, and 6,000 types of coral. Are there? You know what? I actually looked it up because I didn't trust it. There's one called lettuce leaf coral. What are we doing?

Jacob Bourne (02:53):

Yeah.

Marcus Johnson (02:54):

Too much time on our hands. Anyway, today's real topic-

Jacob Bourne (02:56):

Interesting numbers.

Marcus Johnson (02:57):

Thank you. Can AI chatbots revolutionize our lives? All right, Jacob, we're talking chatbots today. We start with a definition, what even is a chatbot?

Jacob Bourne (03:13):

Yeah, I think a pretty easy definition is it's a digital interface that's powered by, well, currently anyway, a generative AI model, but not just any generative AI model, a particular one called large language models that basically produce natural sounding conversational output based on user prompts and the prompt can be a typed prompt or audio prompt now. Interestingly, one of the first chatbots was called ELIZA, and actually that was developed back in the 1960s.

Marcus Johnson (03:45):

Oh, is this the ELIZA complex?

Jacob Bourne (03:48):

Yeah, the ELIZA Effect.

Marcus Johnson (03:50):

Effect, that's it.

Jacob Bourne (03:51):

Where they realized that people would, even though it was a very rudimentary chatbot, not powered by generative AI, they found that people tended to ascribe human qualities to the chatbot, even though it was pretty basic.

Marcus Johnson (04:03):

Right, that was it. So voice assistants, how are they different?

Jacob Bourne (04:07):

Well, voice assistants predate generative AI. So I think-

Marcus Johnson (04:10):

Right. Like Alexa, right? [inaudible 00:04:12]

Jacob Bourne (04:12):

Yeah. I mean, when we talk about chatbots now, I feel like we're often talking about ChatGPT and things that are powered by generative AI. Siri Alexa, Google Assistant, those have been around for years prior to the generative AI explosion. And those are AI powered, but didn't have the same kind of natural conversational abilities that chatbots have now.

(04:36):

Now what's interesting is that we're seeing kind of a blurring of the lines between historical voice assistants and generative AI chatbots and the reason why is because a natural next step for companies is to update their voice assistants with generative AI. And so we're seeing Google Assistant, for example, is sort of merging with Gemini, and a similar thing is underway with Siri and Alexa. They're finding that it's challenging though, because what you're doing is you have this old tech stack and you're trying to add generative AI to it versus developing it from the ground up, like with ChatGPT, which also has voice mode, interestingly. A lot of these generative AI models that were developed as such are getting voice mode. So there's a real blurring of the lines that we're seeing going on between the two types of technologies.

Marcus Johnson (05:23):

Okay. An alien lands on Earth. You are the first person to greet it, and it says to you, "All right, I've heard of these chatbot things." Paint the picture, what are the most popular chatbots or most impressive currently in the market?

Jacob Bourne (05:36):

Yeah. Well, I mean, as far as number of users or daily visits, I mean, we're still looking at OpenAI's ChatGPT as leading the marketplace according to current data. You also have other leading ones like Google's Gemini and Anthropic's Claude. Both of those have gained ground, not only that, but actually some recent research shows that they're toe to toe in terms of a lot of these benchmark tests. There might be some nuances between how they perform on certain tasks, but they're pretty comparable models, even though ChatGPT is really still getting that lion's share of consumer usage.

(06:19):

And of course there's a plethora of chatbots out on the market now in addition to those, and they all have their strengths and weaknesses, but they tend to perform similarly. And the benchmarks are helpful, but generative Ai is such a creative technology that I think that ultimately some of the performance is a bit subjective and of course varies depending on the use case.

(06:41):

The one thing I think is also important to note when we're looking at the leading chatbots is there's a lot of open-source chatbots on the market too and some of them are powered by Meta's models or Databricks, but there's a lot of other companies too that are supplying open-source models.

(06:57):

Now, what's happening with these open-source models is, well, a lot's happening. There's a lot of chatbot interfaces out there now that are powered by these open-source models, which can be a good and bad thing depending on what they're being deployed for. One use case that is really rising in popularity is this so-called relationship chatbot where people use the chatbots for companionship. And one danger here is if you're using an open-source model from a shadow company, there might not be any privacy protections at all and there's a concern that people's data is getting sold to whomever, it could be bad actors, for example. So in terms of performance, you could probably find a very high performing chatbot from a lesser known company, but you do want to take the time to look at the data privacy protections that are involved.

Marcus Johnson (07:48):

Yeah, there are a lot of options out there. To your point though, there was that data from World Bank Group, which we cited on a previous episode looking at the most popular generative AI tools and ChatGPT for OpenAI, 2.3 billion visits in March of this year, which is just multiples higher than Gemini in second place with 133 million. And then you've got smaller ones, Poe, Perplexity, Claude, DeepAI, Copilot, and then Image ones, Midjourney, Prezi, et cetera, et cetera.

Jacob Bourne (08:21):

[inaudible 00:08:21], images, yeah. Yeah, Poe is actually like a clearinghouse of AI models. So you can get Claude, you get ChatGPT image generators, which I think shows that people like a variety because not all chatbots perform equally well on every task. But I think the thing also is just to remember, ChatGPT has been on the market for longer. Gemini and Claude have gained ground in terms of users, and I think there's a lot more market share to be had out there.

Marcus Johnson (08:48):

Yeah, because a lot of this is going to depend on the device, correct? And if you've got Gemini, and then Google's going to have a hand there-

Jacob Bourne (08:54):

What kind of phone you've got-

Marcus Johnson (08:56):

Exactly.

Jacob Bourne (08:56):

... for example.

Marcus Johnson (08:57):

Yep.

Jacob Bourne (08:57):

Mm-hmm.

Marcus Johnson (08:58):

Okay. So that's some of the models that are out there. In terms of the who, there are some numbers on the demographics of who's using them. So Experian Data Quality had some numbers from earlier this year, US consumers who use AI-driven chatbots, and basically, unsurprisingly, Jacob, it's folks who are younger, folks who are higher income, they're more likely to be in the jobs that you would typically use one of these chatbots for work potentially. And then also interestingly, folks that live out west where the AI companies are based. So 24% of people out West and 19% in the Northwest regions using them versus 14% Southeast, Southwest and 12% Midwest. So quite big discrepancies there across the country.

Jacob Bourne (09:49):

I mean, it makes a lot of sense that the younger generations would adopt this technology faster. I mean, they grew up [inaudible 00:09:56] native, and so it's kind of a natural thing to try it out, for example. But I think that older age groups will catch on. And the reason, again, it goes back to this convergence between chatbots and voice assistants. Studies are finding that elderly people who live alone, for example, are really benefiting from voice assistants and chatbots, or can benefit. And so I think we'll see some more adoption on that front. And I think just the more people get exposed to chatbots, and what they can get out of generative AI tools in general, understanding the use cases and how it might be applicable to their daily life, then we'll see more adoption, and of course, things like Apple Intelligence, whoever has the latest iPhone is going to probably be using it by default. Yeah, so I think we're going to see an expansion in terms of the demographics that use generative AI.

Marcus Johnson (10:45):

Yeah, currently, if you're under 40, close to 30% of people are using those. 40 to 55, you're looking about 15%. And then folks 55 and older, it's about 5% now, but to your point, that could change with these emerging use cases, particularly ones where you can talk to these devices and get some form of companionship or conversation.

(11:06):

So how do folks feel about them? According to a CivicScience survey from this summer, 45% of Americans have an unfavorable view of customer service chatbots, that's up from 2022 a little bit. And that unfavorable view is two and a half times as many people saying unfavorable as folks who have a favorable view of them. So my question here, Jacob, is what are the two main obstacles to widespread chatbot adoption given how people feel about them in the moment?

Jacob Bourne (11:36):

Well, I mean, there's the customer service chatbot, and then their chatbots in general, so I think there is a distinction between the two. I think one of the reasons why people don't like customer service chatbots is because if they have a real problem and they need an immediate challenge they want solved, they might feel like it's whatever company is providing the product or service is trying to get out of solving that problem by using a chatbot.

Marcus Johnson (12:01):

On that point really quickly, that would be my biggest frustration with them as well, personally. But there was some data, this is from a survey from real estate company [inaudible 00:12:13], "6 in 10 folks said they were concerned about the lack of empathy from AI in customer service," and that was higher than folks saying there was a lack of human interaction, AI wasn't understanding or resolving the problem, or things took longer to get resolved. So they're frustrated with those things and those things had high shares, but number one was it's just not empathetic like a human can be.

Jacob Bourne (12:34):

Yeah. And I think that's a fair assumption that a chatbot wouldn't be as empathetic. Now, interestingly, ChatGPT has scored higher than some human doctors in terms of bedside manner.

Marcus Johnson (12:47):

Oh, interesting.

Jacob Bourne (12:49):

Things could change. It depends on how advanced the chatbot is that is being deployed for the customer service function. In some cases, you might get better service from a chatbot than a human, but not necessarily. Of course, I think it depends, it also depends on how well the human customer service agent is trained. This comes down to effective deployment too. It's not whether or not people are going to be willing to use chatbots or not, it's about how you deploy it. Is it an advanced model to serve the needs of your product to service? Can your customer, if they want to speak to a human, can they access that human if the chatbot can't meet their needs? So I think those are all important pieces of the puzzle.

Marcus Johnson (13:30):

Yeah, the deployment piece matters, but it's also surprising that it doesn't seem like companies are really listening to the customer on this because, "6 in 10 people said they would prefer companies not use AI for customer service, mostly because they worry or make it even more difficult to reach a person," that was from a Gartner survey. And our colleague Gadjo noting, "Half of people said they would consider switching to a competitor if they discovered a business was preparing to use AI for customer service." Companies are using them anyway, most likely because the cost savings are just too great.

Jacob Bourne (13:59):

The cost savings, right.

Marcus Johnson (13:59):

Yeah. Last year, 7 in 10 employees across industry surveyed by Forrester Research said their companies were either experimenting with or implementing conversational AI and chatbots on their websites.

Jacob Bourne (14:10):

Yeah, and if everybody's doing it, then everybody's doing it. Of course, there are other considerations consumers have when choosing what businesses they do business with. So this is one area of generative AI and kind of more historical AI chatbots, I think we can probably certainly say they're not as capable as the ones currently powered by generative AI models and so-

Marcus Johnson (14:33):

Or the ones in a few months, or the ones in a year.

Jacob Bourne (14:33):

Exactly.

Marcus Johnson (14:35):

So people saying, "I don't want them." They're saying, "I don't want the current version," and companies are probably saying, "Bear with us, it's about to get much better."

Jacob Bourne (14:42):

So consumers could come around, I think. The other thing to note is that people are adopting generative AI in general. I mean, according to Harvard University economist, David Deming, "GenAI has had a faster adoption rate than the internet or personal computers." Now, that could include image generators too, but I think mostly we're talking about chatbots, like ChatGPT there. So there are obstacles, but they're not major obstacles really. I mean, people are really picking up on this technology.

(15:09):

Now in general, I actually think the biggest obstacles with generative AI in general is actually the high computational and power resources needed to deploy it. And I think that does limit the chatbots' capabilities and the speed of output. And if those two challenges were solved, I think we would see probably more capable generative AI tools and potentially lower costs to use them for consumers as well.

Marcus Johnson (15:36):

Yeah. Okay, circling back quickly to the customers aren't thrilled with the current experience point, according to a CivicScience survey, "The share of people saying that customer service chatbots were helpful is going up." It went from 34 to 44% from 2022 to 2024, so that is going in the right direction but-

Jacob Bourne (15:57):

It could be because of Generative AI as well.

Marcus Johnson (15:58):

Yes. Yep. But the cost savings here I think are interesting because The Economist was noting a median salary of a customer service job in America, $40,000 and almost three million people working in customer service jobs according to the Bureau of Labor Statistics. And there was an Economist article noting last year, Gartner predicting that GenAI would lead to 20 to 30% reduction in customer service jobs by 2026, so there's money to be saved there for companies [inaudible 00:16:27]-

Jacob Bourne (16:26):

Yeah, and it depends too, because a lot of these roles can be outsourced as well. So is that 20, 30%, is that domestic or is that global companies to look at too.

Marcus Johnson (16:35):

Yeah, exactly. But yeah, it seems like customers and companies just not on the same page when it comes to using chatbots for customer service. "33% of customers said AI will never improve how they interact with businesses versus just 2% of business leaders who agreed with them said the same. And 49% of business leaders said, AI will improve customer service now versus 16% of customers thinking the same," as from a 2023 LivePerson survey. So currently not on the same page, but that's true for a lot of areas of business, not just GenAI.

Jacob Bourne (17:11):

Yeah.

Marcus Johnson (17:11):

Let's talk quickly about the next year or so. What do you see being some of the coolest chatbot applications coming out over the next 12 months?

Jacob Bourne (17:19):

Well, we're going to continue to see a lot of different tools hit the market, and it's really hard to keep up with them. One area that doesn't get talked about as much, but I think will be interesting is not necessarily Neuralink, the Elon Musk's company because that's an invasive procedure, but alongside that, we're seeing some non-invasive computer interface technologies advance. And what that allows for is for AI companies to use that data from brainwaves to then feed an AI model. So we could see some interesting applications come out of that, like for example, potentially dream analysis or something like that, healthcare monitoring. Obviously there's privacy issues that would come up with these types of applications.

[NEW_PARAGRAPH]Beyond that, I mean, some of the stuff we're already seeing signs. I mean, video generators are here. They're not necessarily being deployed yet commercially because of the high costs that they incur, as well as just concerns about deep fakes and the fact that we have a big election coming up and things like that. But I bet by next year we're going to be seeing some more video generators hit the market commercially. We're going to see a lot of content come out of that. It's going to have impacts on social media, impacts on the entertainment industry, including the movie business, even the video game industry as well. So I think that's another area to watch closely.

Marcus Johnson (18:46):

Let's end the show with this, Jacob. So Gadjo, our colleague, recently wrote about, "Salesforce moving from chatbots to intelligent agents as the next stage for AI business. These AI agents basically can automate various business processes," so they can do things for you completely or they can do most of the job and then check in with you to see if you want them to keep going or basically complete the final step. Help us out, what are the main differences between AI agents and chatbots?

Jacob Bourne (19:21):

Yeah. Well, I think that chatbots are mainly designed to be interactive, so they're creative, it's a back and forth that's really dependent on the human giving a constant prompt to give output. Whereas the AI agent is, yeah, you might prompt it initially, but it's really designed to take action on its own in the background. So it's doing work for you that you don't necessarily have to prompt it to do step by step. And there's a little overlap between the two because you can get some sort of agentic behavior from a chatbot, but it's not really designed that way. It's designed for this back and forth. It's more of an interaction between the human versus the AI agents that really is supposed to be an autonomous AI worker essentially. And we actually just saw Microsoft announced that it's going to be rolling out some agentic capabilities to Copilot.

(20:10):

Kind of as a segue to your last question about what we're going to see, some interesting applications, I think this is another area we're going to see a lot of growth next year is in AI agents. I think the AI companies are really focused on this. I think it's an area of big risk though, because again, you're putting your faith that whatever you're allowing the AI agent to do on your behalf it's actually carrying it out in a way that you want, it's not making mistakes like sending out a mass email that you don't want sent, or something like that. Obviously, there's some pitfalls there. But I think in terms of, especially for workplace use cases, I think this is an area where we could definitely see some more ROI from generative AI in terms of really improving productivity if you have AI workers that are just getting these kind of more menial tasks done in the background and freeing up time for human workers to do other things. So yeah, it's definitely an area of a lot of growth that we should expect to see more of next year.

Marcus Johnson (21:17):

Very good. Well, that is all we have time for for today's episode. Thank you so much to my guest. Thank you to Jacob.

Jacob Bourne (21:24):

Thank you so much for having me.

Marcus Johnson (21:24):

Yes, sir. Thank you to Victoria who edits the show. Stuart runs the team. Sophie does our social media. Thanks to everyone for listening in to the Behind the Numbers Daily, an eMarketer podcast made possible by TikTok. You can tune in tomorrow to hang out with Sara Lebow on the Reimagining Retail Show, where she'll be speaking with Susie David Canyon and Blake Droesch, all about the best ancillary retailer services that help get customers in the door.