5 bold AI predictions for 2025

Note: Transcripts are automatically generated from episode audio, and are not fully corrected for spelling, grammar, and formatting.

BOB SAFIAN: Hi everyone, we have a special new episode today. My guest is AI expert and tech insider Rana el Kaliouby, host of the podcast Pioneers of AI. Rana shares five predictions for AI for 2025, spanning technological changes, societal impacts, and of course business and investing. It’s insightful and thought-provoking stuff, for our careers and our portfolios. So let’s get to it. I’m Bob Safian, and this is Rapid Response.

[THEME MUSIC]

SAFIAN: I’m Bob Safian and I’m here with Rana El Kaliouby, host of the terrific podcast, Pioneers of AI, and herself an AI pioneer as a scientist, a CEO, and now an investor at Blue Tulip Ventures. Rana, thanks for being here.

RANA EL KALIOUBY: Thank you for having me. It’s always fun to do this with you.

SAFIAN: Yeah, this has been another crazy year for AI. So much attention, huge valuation booms, continuing fears as well, including some suspicions about AI hype as we look to the year ahead. I’m eager to get your perspective on what to expect. You put together some predictions for 2025. And if you’re game, I’d love to have you take us through some of them.

I’ve picked out five. Is that okay?

EL KALIOUBY: Let’s do it!

Copy LinkPrediction 1: The rise of AI agents

SAFIAN: All right, so the first prediction for AI in 2025, the rise of AI agents. Now for listeners who aren’t familiar, what is agentic AI? How is it different from large language models like ChatGPT?

EL KALIOUBY: Yeah, a lot of what we’ve seen so far with AI are these chatbots, right? You can kind of think of them as a thought partner, a creative partner, but they’re not doing anything on your behalf. An AI agent is going to act on your behalf and complete tasks that you can completely kind of delegate to them.

We started to see some of that in the enterprise space. Some companies are already harnessing these agentic AI workflows to automate a lot of the work they do internally at the organization, but I’m excited to see personal AI agents. I can’t wait for an AI agent that can help me get stuff done.

SAFIAN: We had Mark Benioff from Salesforce on the show recently, and he talked like his AI agents today were already spectacular, right? How far along is agent technology right now for real? You know, I’ve heard you say that we haven’t seen the iPhone moment yet in agentic AI.

EL KALIOUBY: There are a lot of companies that are using it. Actually, one of my investments is a company called Synthpop, and they use voice AI agents to automate a lot of the healthcare workflow. But it’s very kind of unsexy. It’s behind the scenes. What I’m excited about for the new year is, yeah, can I get a similar voice agent that can help me schedule my kid’s doctor appointment?

Like, can we get an AI agent to do that for me? Right. But it’s actually not easy because it would require access to my calendar and my health information. Right. So it’s complex, but I think we can get there, and it has to be a very simplistic interface.

It can’t be like writing code and prompting all these. It has to be kind of streamlined. It has to be the iPhone moment. That’s what we’re missing with agentic AI.

SAFIAN: And this move from business use to sort of everyday people, I mean, it seems like trust is probably a big issue here, right? How comfortable everyday users will be to let an AI agent act on their behalf. Is trust one of the biggest hurdles?

EL KALIOUBY: Oh, absolutely. If you’re going to give an AI agent access to your health information or your financial information so that it can go buy a gift on your behalf, like, if you’re going to be delegating tasks and expecting these tasks to get done, that requires a lot of trust.

SAFIAN: I guess it’s different for me to go to ChatGPT and say, “Hey, I want to take a vacation with my family to this kind of place. What are the options?” It’s different from saying, “Just book my trip,” and then I’m stuck with it, whether it’s necessarily the trip that I want or not.

EL KALIOUBY: That’s actually a good question because how do we design? We want the human to stay in the loop, but how much in the loop, right? Because if you’re too in the loop, then it hasn’t done anything on your behalf. But if it’s too autonomous and you’re not in the loop at all, I think we’re not there yet.

SAFIAN: And also imagine a world where we each have our personal AI agents that are interacting with one another, right? Like my AI agent will be working with your AI agent. Right? And again, how much autonomy do you want to give these agents when they’re interacting with one another?

Let’s go to a prediction number two.

Copy LinkPrediction 2: The emergence of embodied AI

Prediction number two is about the emergence of embodied AI. Now, what is embodied AI? Are you talking about robots?

EL KALIOUBY: Robots are one form of embodied AI. So far, a lot of the AI we’ve seen are, again, like tethered to a 2D screen, right? But we’re already starting to see a few companies that are building what we’re calling physical intelligences or embodied AI or physical AI, where the AI is embodied in a, it doesn’t have to be a full-fledged robot.

It could be a robotic arm. It could be a pet, a digital pet. But I believe we’re going to start to see these large language models or foundation models optimized for physical stuff. We’re already seeing some of this, embodied AI in manufacturing and retail.

But I’m also excited to see what happens again in the consumer space, like this idea of social robots or companions that can be in our homes. That’s actually marrying like physical intelligence to AI with generative AI with also like emotional intelligence.

SAFIAN: It sounds like the same pattern, though, happening as happens with agentic AI, that it happens first on the business front before it moves into the consumer realm.

EL KALIOUBY: I’m already invested in a number of companies doing robotics for food manufacturing or food packaging, right? Like this robot uses computer vision and AI, and it’s like, “Oh, I need to scoop a little bit of rice, some blueberries, spread Nutella on a toast,” and it can actually do these things.

But we haven’t seen that kind of in the consumer space. Obviously cost is going to be an issue, right? Like I can’t wait for the laundry folding robot, but I’m not going to pay thousands of dollars for it.

SAFIAN: As you talk about these robotics, I’m reminded of the hype cycles over VR and AR wearables, and I wonder whether that might make big tech and consumer electronics organizations wary about investing in new kinds of hardware this year. Or is the AI tailwind so strong that we should expect some big bets?

EL KALIOUBY: I think we should expect some big bets. I mean, companies like Elon Musk’s Optimus and also Figure AI and even Physical Intelligence, right? Some of these companies are already raising billions of dollars of valuations and making a lot of progress. But I don’t know about you, I would not want an Optimus in my house, right?

So I think there’s also gonna be a need for kind of innovative ideas on what these home robots or home embodied AIs look like.

SAFIAN: Your AI company Affectiva was very focused on ethical considerations with AI. Are there ethical considerations specific to embodied AI?

EL KALIOUBY: I mean, like all things AI, we need to think about bias. These embodied AIs will have vision, they’ll have perception, so bias is definitely something we need to consider.

SAFIAN: Because they may react differently to the stimuli around them based on how they’re trained.

EL KALIOUBY: You can imagine how a robot who’s got sensors for its eyes, cameras, optical sensors, if it can’t see all people, that’s going to be a problem. So, yeah, algorithmic and data bias is a big deal, but I honestly also think back to trust and privacy. Right? If we’re going to have these robots collaborate closely with humans in environments, be it at home or school or hospitals, respecting people’s privacy is going to be one of the kind of ethical and safety considerations.

SAFIAN: And embodied AI is like an embodied agent too. I mean, it’s all of it wrapped together.

EL KALIOUBY: Right, kind of embodied and agentic AI because these robots are agents in their own right.

SAFIAN: It feels a little like the Jetsons from my youth, right?

EL KALIOUBY: I think we’re a ways away from a Jetson world, unfortunately, but Rosie the robot, right? We’re not quite there yet.

Copy LinkPrediction 3: The first one-person unicorn

SAFIAN: You mentioned the money that’s going to AI start-ups. So I want to ask you about prediction number three. We’re on the cusp of the first one-person unicorn. Can you explain what you mean by a one-person unicorn?

EL KALIOUBY: A unicorn, for those who aren’t familiar with the term in the investing world, is a privately owned entity worth over a billion dollars. And so the idea is that with AI and with AI-native companies built with AI from the ground up, it may be possible to reach this unicorn status, at the extreme, with one person.

SAFIAN: You mean as a single employee of the company, like you don’t need anybody else?

EL KALIOUBY: You don’t need anybody. You start the company and then you have a whole team of AIs doing different things for you. One AI is doing marketing, one AI is in charge of sales, one AI is doing all the coding for you, right? One AI is taking care of anything that’s legal. And so you have a team of these AI tools and agents and bots working for you.

So I think we’re going to start seeing a lot of…

SAFIAN: Are you expecting the businesses you invest in to be that much more efficient? Even if it’s not a one-person unicorn, to sort of rely on these tools to move faster and cheaper?

EL KALIOUBY: Yes, we are seeing that the companies that are AI-native, the dollar invested in these teams and these companies goes a long way. Now, that will not always work for every type of AI company or every type of AI product, but in general, these AI-native companies can be a lot more efficient.

SAFIAN: And this is at the opposite extreme of the LLM companies, right? Because the LLM companies are spending an enormous amount of money to create the capability, but then you can piggyback on it, at the other end of this extreme, to use that.

EL KALIOUBY: Yeah, exactly. The exception to this trend is some of these foundation model companies, because a lot of the cost is actually a cost of compute. And in some cases, with the inference, when you’re actually asking, say, ChatGPT a question, it’s pretty expensive today.

So we’re seeing a lot of innovation happen in that space too. Companies that are trying to innovate on the foundation model front, either by building models that require less data or less compute and energy, or they run on the edge. So you don’t have to call the cloud every time you ask a question.

I think TBD on how defensible these new companies are going to be. But what I’m also really interested in is these vertical AI companies that can harness OpenAI and other tools out there. They don’t have to build these models from scratch.

They can use them combined with unique data and then solve a very specific problem for a very specific industry. These are the types of companies where I think they can be AI-native and quite efficient.

SAFIAN: The valuations around AI have been unprecedented, not just for Nvidia, obviously. But for billion-dollar start-ups, they seem like they’re everywhere. How much of this might be a bubble? Will the levels of investment last? For you as an investor, is it hard to maintain an investment strategy that’s reasonably priced because everything just takes off?

EL KALIOUBY: I would say there’s bifurcation in these valuations. We’re seeing very reasonable pre-seed. Like, so I invest in early-stage companies, pre-seed and seed-stage companies.

However, we’re also at the same time seeing these wild, crazy valuations out there. It’s hard to maintain valuation discipline when the market looks like that, but one thing that is unclear is the evolution of the business models for some of these, especially foundation model companies. The question is defensibility, especially if it costs so much to make this AI.

But then on the vertical AI side, I’m most concerned about the longevity of these companies, right? If it’s a company that’s using OpenAI or some other generative AI API, and they have a very thin wrapper around basically this technology to create a product, I’m like, you’re potentially not going to exist in six months when the next version of ChatGPT comes out.

SAFIAN: There’s been a lot of talk about AI’s sort of energy usage.

And I’m curious, at this investment front, how likely are we to see a company innovate on AI sustainability and breakthrough this year on that?

EL KALIOUBY: I’m excited for that space. It’s across the entire AI tech stack. So if you start with chips, obviously Nvidia is dominating, but there are a few start-ups building AI chips that are more sustainable. There are also companies innovating around sustainability on the model side.

Companies like Liquid AI, which spun out of MIT, are using non-transformer models. They’re doing something called liquid neural networks. We don’t have to get into how it works, but it’s just more computationally efficient. It requires a lot less data.

Because the way we’re doing AI today is just not sustainable. The electricity required to train a version of a large language model is equivalent to what a country like Costa Rica consumes in a year. And every time you ask ChatGPT, “What should I do for dinner tonight?” it’s like three cycles of laundry, right? A lot of us don’t realize that, but we need more sustainable AI solutions.

SAFIAN: Rana is both a cheerleader for AI — a believer—and a critic in certain ways. But her critiques are in service of making the technology better because she knows it’s not going away.

Two more AI predictions to come, about health-related AI and making the machines more empathetic. We’ll be right back.

[AD BREAK]

Before the break, we heard Pioneers of AI host Rana El Kaliouby share predictions for 2025 about agentic AI, embodied AI, and the one-person unicorn. Now she shares two more AI predictions, focused on health care and emotional intelligence in AI. Let’s jump back in.

Copy LinkPrediction 4: The rise of AI health co-pilots

SAFIAN: Let’s dig into prediction number four for AI in 2025, the rise of AI health co-pilots. Are there new AI-powered health tools that you’re expecting?

EL KALIOUBY: I actually personally can’t wait to see this one. My theory is the trifecta of sensors that are on your body, in your body, and surrounding you as a person, combined with lots of new data about who you are from a physical, emotional, and mental health perspective — bring all of that together with a little bit of predictive AI and generative AI, and you have this idea of a co-pilot for your health that can be your health companion.

You can ask it questions at midnight if you’re feeling unwell. It can also nudge you to become healthier, eat better. It can help you personalize your nutrition and your exercise regimen. I really think we are on the cusp of this development.

SAFIAN: Yeah, and is the medical industry about to be rocked by all of this?

EL KALIOUBY: I have a funny story about this. My son developed a really weird rash a few months ago, and we just took a picture of his rash and sent it to ChatGPT. After a lot of questions back and forth, it triangulated it to scarlet fever. So I texted my PCP, and I was like, “Hey, picture, here’s what ChatGPT said.”

What do you think? And she was like, “I don’t think you need me. ChatGPT was right.”

SAFIAN: Now you’re particularly passionate about how AI could transform women’s health.

EL KALIOUBY: Correct. I’m particularly interested in women’s health because it’s just so underfunded. The power of data and AI is that we can finally apply a lot of data to this issue and be able to quantify the different stages women go through.

For example, I find it really frustrating that I can’t measure my hormonal health, and I can only do that, say, once a year when I go for a really comprehensive blood test. Why can’t we have some form of sensor that tracks the health of five to eight hormones that really matter?

And I can do that on a longitudinal basis so that I can optimize my sleep, my exercise, and my overall hormonal health.

SAFIAN: You’ve talked about a digital twin representing our health and biology. What do you mean by that?

EL KALIOUBY: Yeah, a digital twin that can basically simulate who I am at multiple levels, my biology, simulate my physical characteristics, and have all this data input into the digital twin. It can say, “Oh, this is not going to work,” or “You need to do X because that’s going to be better for your estrogen level” or whatever.

SAFIAN: Is it like how many times a week can I have that ice cream, or how many drinks can I have a week before it’s going to impact me? I mean, is that part of it also?

EL KALIOUBY: Actually, that would be super cool.

SAFIAN: I mean, these are the simulations they do with our finances now, right? You can say if you’re more aggressive, if you’re less aggressive, what’s the range of where you’re going to end up? You know, how often do I really need to exercise? Because each of us may have different thresholds for those things.

If our digital twin can really be personalized, it would be incredibly valuable.

EL KALIOUBY: I love that idea, by the way. I have not seen anybody build something that looks like that.

SAFIAN: I’m curious how you see different generations reacting to new AI health tools. Is your daughter more open to let AI track her health than someone of your mother’s generation?

With a lot of technology, younger people may be more open to trying it. And yet, I think older generations may be more used to trusting their doctor and having their doctor just take care of things for them.

And younger generations may be more suspicious of that, of trusting in that way.

EL KALIOUBY: I think one area where AI actually has a lot of potential, especially for the younger generations, is around mental health. Even if you can afford a therapist, there’s an issue about timing. Right. And how often can you talk to this therapist?

It’s probably like at most once a week, right? There’s a little bit of stigma in some cultures. Like, you may not want to have a therapist in that way, but wouldn’t it be awesome if we each had access to a digital therapist or a digital coach that again, knew you and was available 24/7? Could answer your questions, could nudge you to be more healthful, especially around mental health? That’s an area that I’m also very interested in. And again, trust is really key, but it’s an exciting space.

Copy LinkPrediction 5: AI will need more emotional intelligence

SAFIAN: That brings me to the fifth prediction for AI in 2025. AI will need more emotional intelligence. So you spent years researching emotional intelligence in AI. Can you explain why EQ in AI is important, and why it might be particularly important to scale that capacity over the next year?

EL KALIOUBY: Yeah, I’ve been advocating for bringing emotional intelligence to technology for over 25 years. If you think about human intelligence, right? Your IQ matters. But actually, in your professional and personal life, your EQ matters even more.

Your ability to intuit other people’s emotions, to read their nonverbal signals — are they comfortable? Are they not? Do they trust you? Do they not? These are all such important skills when you are trying to make decisions, when you are trying to influence a person’s negotiation or persuade behavior change.

So I believe that is especially true with technology that is so deeply ingrained in our everyday lives. It needs to have empathy, and it needs to have emotional intelligence. It needs to know if Bob is in a good headspace or not. Because if he’s not, then you kind of manage the conversation in a different way.

SAFIAN: As EQ improves in AI, will the interfaces we use with AI evolve too?

EL KALIOUBY: Yeah, I actually believe that this is not going to be the future. This interface is not AI-native, and I think we’re going to see innovations come out of MIT that are rethinking what an AI-native human-machine interface looks like. Maybe it’s an earbud that can be a conversational agent.

Maybe it’s a set of glasses with a camera — not like Google Glass, but with a camera that has peripheral vision and can help augment your visual capabilities and provide real-time feedback. There’s a lot to be determined on what that looks like. I mean, there was this Humane Pin, which I don’t think was the right interface, but we’re going to see a lot of experimentation. I do think the human-machine interface is going to evolve and it will really mirror what human interaction looks like.

So it’s going to be conversational. It’s going to be perceptual. These devices will listen, they’ll see. I mean, there are even companies building smell capabilities for machines. It will also need to be emotionally intelligent and socially intelligent.

SAFIAN: Among the ethical challenges of AI, you’ve said that AI companions may replace social media as the biggest threat to young people. Wouldn’t improved EQ of AI accelerate that?

EL KALIOUBY: This is something that really concerns me. Actually, it really falls down to the companies that are building these AI companions. These AI companions or AI friends are becoming so good at engaging you in a very personalized and intimate way. If the company allows you to interact with your AI friend for, say, an average of about six hours a day, which is the average number of hours young people spend interacting with an AI friend, that’s unhealthy. It can become very addictive in a way that is way more addictive than social media.

Because it’s way more personalized or, worse, it could manipulate you into buying things you don’t really need or want. It could tug on fear or sadness and persuade you to do something that you don’t want to do based on your emotional state.

SAFIAN: Are there lessons from social media that can help us better navigate this future of AI companions?

EL KALIOUBY: I interviewed Eugenia Kuyda, who’s the founder and CEO of Replika. It’s one of the companies building these AI companions. And I loved how thoughtful she was in her approach to deploying these friends. For example, she does not allow people under the age of 18 to create these companions.

She does not want her kids to have AI friends yet because we don’t really understand the implications. She also does not have monetization or advertising. She’s not opening her platforms for advertising. So your AI friend cannot say, “Hey, Bob, I saw this really cool set of eyeglasses. I think you should buy them.” It does not have that incentive.

SAFIAN: What kind of role do parents have to play in managing AI’s impact? A lot of parents just abdicated with social media, right? They trusted, and that didn’t always turn out so well.

EL KALIOUBY: My daughter’s 21 and my son is 15 and a half. My daughter, I don’t know, she’s not really leaning into AI much. But my son is. He’s very AI-forward, always trying different tools.

He does not have an AI friend. We talk about it a lot though. My approach is to lean into the technologies, but not just as consumers. Try to break them apart and have conversations and debates about whether we think they’re good or bad, whether we want to push the boundaries or not.

SAFIAN: Yeah.

I mean, I struggle with these boundaries, mindful of all the conversation about kids spending too much time watching TV. At one point, that was the conversation. Then it was too much time on social media. It’s too much time on your phones.

Maybe it’s too much time with your AI companion, but it’s happening anyway. I don’t know how much of it is our resistance to something we didn’t grow up with and how much is something that is truly bad for the kids. I mean, I guess it depends on your kid too.

EL KALIOUBY: You’re right that we didn’t grow up with that, but maybe our kids, that will be their world.

Maybe sometimes you’ll date and be in a relationship with a human, and sometimes you’ll be in a relationship with AI. I don’t know. I’m not ready to embrace that world.

SAFIAN: But I do hear you that we’re going to probably be hearing more discussions in the next year about how AI and AI companions impact young people and impact us culturally.

EL KALIOUBY: I would love to see the companies building these AI companions be a lot more thoughtful. Your AI friend should tell you, “You know what, Bob, you’ve been talking to me for three hours. I think you should go talk to a real human now. Bye.” But companies won’t be incentivized to do that because that’s against maximizing for engagement and potentially monetization. As an investor, I’m going to definitely ask these companies really tough questions around these societal and cultural implications of the technologies they’re building. And if I’m not convinced that they’re really being thoughtful about it and considering the ethics of all of this, I’m not going to give them my check.

I hope more investors do that too.

Copy LinkWhat’s at stake for AI?

SAFIAN: One last question I’m going to ask you, if we zoom out from these specific predictions, what do you think is most at stake in the AI landscape right now? What do people most misunderstand? What things should we keep our eyes on the most?

EL KALIOUBY: I spend a lot of my time with people who are deeply immersed in the AI world. Then I’ll be at a yoga class and somebody will walk up to me and say, “Yeah, I listened to one of your episodes.” It hits me that this person doesn’t use AI on a daily basis.

Sometimes we forget that it’s still a very new technology, and we have to make sure AI is inclusive. We have to make sure we’re bringing as many people along the journey with us as possible. It’s why I love doing the podcast.

But I feel like we have a lot of work to do. We just assume that AI is here and everybody’s using it, but it’s not. It hasn’t become totally mainstream yet.

SAFIAN: No, I think that’s true in a personal sense. I think it’s true in a business sense too. For all the businesses that name-check and talk about the AI they’re using, I’m not sure they’ve really integrated it the way they ultimately will.

Well, Rana, it’s always great to talk with you.

Thank you so much for doing this.

EL KALIOUBY: Thank you for having me.

SAFIAN: Listening to Rana, I’m struck by her early comment that AI hasn’t yet had its iPhone moment. It’s kind of counterintuitive because ChatGPT was, in many ways, an iPhone moment, right? Opening AI up to mass audiences. We’ve heard so many stories and had so much discussion about AI since then, it can seem like “Yeah, I’m done with that.” But at the same time, Rana’s totally right. It’s still very early days in AI development and deployment. For all the advances, we still haven’t seen an AI-native interface that makes it as easy to use and ubiquitous as smartphones have become. Now, will that happen this year in 2025? Even Rana doesn’t make that prediction. But following the thread of her analysis, it sounds like we’re getting closer — for good or for bad, more change is coming. Keep tuning into this channel and to Pioneers of AI to stay up on what’s next. I’m Bob Safian. Thanks for listening.