Arm Viewpoints: AI, healthcare & agentic systems: How 2025 defied expectations

Brian: [00:00:00] Hello, and welcome to another episode of the Arm Viewpoints podcast. I’m Brian Fuller, editor-in-chief at Arm, and we’re excited to welcome back our old friend from the 311 Institute, futurist Matt Griffin. Joining in the fun today is the rest of our amazing Arm content team, Jack Melling and Omkar Patwardhan from Cambridge and Kurt Wilson from San Jose.

In the first of the two-part conversation, we discussed how Matt’s bold predictions for 2025 actually played out What landed, what accelerated Beyond expectations and what still sits just over the horizon. In this episode, you’ll hear highlights including which predictions stuck the landing, where the predictions fell short.

Why AI development accelerated far faster than forecast? How advances like AlphaFold and synthetic drug discovery may be the most underestimated [00:01:00] breakthroughs of the year, the complex future of ag agentic ai. That’s ahead as we look back at a whirlwind 2025 and set the stage for where the future is headed.

Let’s dive in. So here we are with few we happy few we band of brothers. We’re back together after a year and what an eventful year it’s been with our good friend Matt Griffin. Welcome Matt. Welcome and thank you everyone. So let’s dive right into it. We did this last year in exercise in editorial calendar building, I guess you wanna call it ’cause you’re our favorite forecaster for prognosticator.

You gave us some bold predictions. Sleeker AR glasses, invisible computing, all kinds of interesting stuff. Before we get into your crystal ball for 2026, give us an honest scorecard which predictions stuck the landing and which.

Matt: So if we have a look at [00:02:00] AR glasses, basically then those got smaller, they got sleeker, they got better and more capable.

In fact, actually Mark Zuckerberg, I think it was yesterday, actually decided to disinvest in his metaverse franchises by about 30% so that he could actually ramp up spending in wearables, AR glasses and other sort of forms of compute. So that one basically seemed to actually land. I say, although.

I think Mark Zuckerberg is really the main proponent driving, the sort of the new wearable and the new UI revolution outside of sort of Sam Altman. We talked about artificial intelligence, basically hitting new IQ levels basically. So we saw chat GPT, for example, hitting 146 in actual IQ, whereas last year it was 155 just in verbal iq.

And we saw Google Gemini hit an IQ of 136. Even though we can’t necessarily directly compare human intelligence to AI intelligence, broadly, when we look forward two to three years, [00:03:00] we can see IIS with an IQ of about 300 plus. When we talked about things like post PhD level artificial intelligences, we saw Grok 4 and we saw so Grok 4 from XAI and we saw Google Gemini.

2.5 starting to hit that benchmark. And then GPT five smashing that benchmark, albeit that GPT five cost a billion dollars worth of compute to train, which you guys would like instead. Yep, absolutely. Oh yeah so generally we were in the ballpark, that’s it, on that and quite a lot of other different

Brian: things.

Anything any, anything didn’t materialize that you thought might. Not really. It’s

Matt: generally a matter of timing. So actually, I suppose on the one hand I thought that some of the healthcare technologies might accelerate a little bit more than they have. But I think overall, when you have a look at the macro investment environment, we saw 5.7 trillion being dropped into AI development, energy data centers, compute GPUs and everything else.[00:04:00]

And really. From a futurist perspective, I think the current AI hype cycles have been quite exceptional. ’cause they’ve been relatively robust now for about the past two years. Normally you see a hype cycle lasting around 18 months. For example, we saw this with Web3, blockchain, metaverse, those kinds of bits and bobs.

But when we actually have a look at healthcare. It seems like quite a lot of investors deprioritize their investments in healthcare and in other tech, other areas such as ESG but obviously American political reasons to actually almost go all in. On AI and AI infrastructure. So it was a bit of a shame to see some of the healthcare breakthroughs not accelerate as much as they could have other breakthroughs and things like 3D printing.

So when we talk about infrastructure those increased a little bit, but again, didn’t really accelerate too much. We saw pretty much everything being pumped into computing and [00:05:00] ai.

Jack: It’s interesting that you mention healthcare there as. An area where the technology’s perhaps going to materialize, but do you see there being like a potential healthcare as a use case that will be really accelerated by AI now and potentially in the future?

Matt: Yes. And somewhat now, ironically, so when you actually have a look at how Americans especially are using chat, GPT and artificial intelligence, Sam Alman has gone on record and said that up to 50% of the inquiries being put into chat, GPT are healthcare related. Now, this is perhaps one of the reasons why when we have a look at the product that he and Johnny Ive basically are actually producing, which we should at least see a prototype by this time next year.

Because for a $6.5 billion purchase, you’re going to want to see something that they’ve actually produced as opposed to just, vaporware. When we actually have a look at the new, say, new chat, GPT and open AI device, one of the. Points that has come [00:06:00] across is, and whether this is true or not, it just seems rather bonkers, is that it’s lickable.

Okay? Now either the reporters in the US have actually got their wires crossed and they actually said it’s likable. But they’ve been quite a few news outlets that have actually said that the new device basically is lickable. Now, on the one hand, that might sound actually really outrageous and odd, however.

If it is, this is a device that ties into healthcare, maybe not prioritizes healthcare, but if it’s a device that ties into healthcare, then artificial intelligence is actually making huge strides in both predictive healthcare and quantitative healthcare. For example, we can collect a whole variety of different biomarkers and biometrics, and we can very quickly determine, your blood pressure, your heart rate.

Your anxiety levels in your emotional state, whether you are depressed or not, whether you’re getting the onset [00:07:00] of dementia, whether you have pancreatic cancer we can tell whether or not you are going to be getting ill basically from whatever virus is going around in the next two weeks and so on and so forth.

On the one hand, when you have a look at just vanilla, artificial intelligence is absolutely accelerating healthcare and healthcare discovery. So we’ve seen companies like InSilico that have used artificial intelligence to develop 30,000 new drugs in 21 days. We saw alpha fold basically finally coming out at the start of this year, and AlphaFold has now simulated and modeled over 300 million proteins, which have been open sourced.

So congratulations to the team on that, which fundamentally changes healthcare. When we have a look at some of the latest. Thoughts from Sir Demis Hassabis, of Deep Mind fame. He now believes that within the next 10 years, artificial intelligence will cure all disease. Now, on the one hand, we can already see [00:08:00]how artificial intelligence is rapidly accelerating new drug discovery, the discovery of new novel antibiotics.

So we’ve seen AI producing 90 new types of antibiotics, which is just unheard of. So on the one hand, basically we can see how artificial intelligence can accelerate drug discovery and help with human longevity, but whether or not people actually have access to those, that’s gonna be another sort of question.

But when we have a look at the use of artificial intelligence within wearable devices, within cameras and all these kinds of different systems Yeah. As, as well as, for example, basically chat bot being used in healthcare to diagnose. Or they then go and see an American GP. It’s absolutely accelerating healthcare.

But the thing is, I think the main shame is that it would be nice to see more investment flowing, not just into ai, but [00:09:00] into AI for healthcare. And that’s where I think when you have a look at.

The AI foundational models, some of the sort of more vertical stack and niche models, and then the infrastructure layers beneath that. And healthcare seems to have been a little bit forgotten about in terms of one of the most transformative uses of ai.

Omkar: You mentioned a point on accessibility, and I wanted to add onto it.

I recently wrote a story about UK. Hospitals in the UK started implementing AI for cancer diagnosis, for early diagnosis especially. How do you see this use case pan out even more to more hospitals around the world, especially in the Americas, where they can potentially now even the patients will start having access to that kind of [00:10:00] technology.

Matt: So when we actually have a look, for example, at the use of artificial intelligence with traditional blood tests, we can now use AI to identify a hundred percent of the pathogens basically within a blood sample. Now, when we start applying this to cancer care, one of the biggest problems that we actually have with cancer diagnosis is generally we are diagnosing you with cancer when the symptoms are already starting to manifest, and in some cases we identify it too late.

Now when we actually have a look at the use of artificial intelligence within gene sequencing especially, increasingly what we can do is we can take a biopsy or we can take a blood test, and from the blood test we can use artificial intelligence to identify very faint genetic cancer markers, which then tell us that there is a cancer in your body.

Now the sensitivity of these tests is now such a, is now so good that even if you have a [00:11:00] few cancerous cells floating around your blood screen, we can generally pick them up. Whereas before you’d have actually needed quite a large cancerous mass basically before you could actually identify. When we think about general accessibility, on the one hand, artificial intelligence, along with other sort of tools and technologies like rapid gene sequencing and so on and so forth, is actually lowering the bar to accessible healthcare.

But as long as healthcare is still reliant on large machines, for different things, like surgeries, for example, MRIs, PET scans, et cetera, et cetera, even x-rays and generally ultrasounds. Then access to some of the, should we say diagnostic tools is actually still going to be limited.

And again, when we think about ultrasound, for example, we can now put ultrasound into a smartphone and use artificial intelligence through the smartphone to identify, should we say, problems that are going on [00:12:00] in your body. AI is definitely helping move the bar when it comes to increasing access to healthcare for people that traditionally haven’t had it.

It’s also starting to level up the world as well. If we actually have a look, for example, at robotic surgeons, we’ve seen more and more robotic, remote, robotic surgeons taking place around the world. And if we get to the point where we can have a surgeon, for example, in Bangalore.

Operating on a patient in Nigeria across a G Teleoperated network with the use of artificial intelligence that actually helps around 158,000 people in the world who currently have no access to surgery, which could be used to help relieve different pains, symptoms, and all sorts of other bits and bobs.

But we also need to get connectivity sorted. So if you have a look at this, Sahel, for example, Sub-Saharan Africa, Sub-Saharan can. Sub in desert, 66% of people there don’t have access to the internet. [00:13:00] So there are some challenges still. When we look at accessibility, the first thing is you need an internet connection for some of this stuff.

Kurt: I guess looking back to last year, were there any predictions that you either didn’t make or became bigger, something that became much bigger than you thought it would this last year?

Matt: Ai. I think again, it’s really gotta be AI development accelerated faster than I thought it would. If we went back a year ago, basically when we said, in the next 12 months we will see almost $6 trillion investment being invested into ai, should we say, or into the technology field.

I think the vast majority of people would’ve said maybe we see, a trillion, maybe 1.5 trillion, something like that. And even though outta that $5.7 trillion, basically that sort of been put up for grabs, most of it isn’t actually committed. Yeah. There’s been a massive acceleration in AI development, and I think part of that is, is due to, on the one [00:14:00] hand, a couple of egos, when you have a look at Sam Altman for example, I don’t think it’s too much of a stretch to say that he would very much basically like to be the person.

Or the individual that helped develop AGI. Same with El Musk, basically with the Colossus computers. When we have a look at Sir Demis Hassabis, his ambition at the moment is to solve intelligence. ’cause once we solve intelligence, he believes we can then use that to solve everything else. And then when we have a look at the American government, basically with Eric Schmidt.

Eric Schmidt went on record in Congress and in front of the Senate a little while ago, about six months ago, and he said. Whoever wins the race to AGI wins everything. And when you dig into that statement, artificial general intelligence is believed to be such a powerful and transformative, but also self-developing technology that whoever reaches it first, these AGIs will develop themselves.

They will develop new products, new services, almost [00:15:00] instantaneously. And even if the country. That developed AGI in second place was like 10 minutes behind. They wouldn’t be able to catch up. When we think about new semiconductor designs, for example, we have actually seen a huge acceleration in the use of artificial intelligence to design some really quite interesting new computer chips.

So I think that acceleration is something that. Would’ve been hard to account for. If we put it into context, the vast majority of individuals thought that artificial general intelligence would be emerging in around 2035. And now basically we’ve got some of the magnificent seven talking about AGI or even sub AGI, but certainly something that looks AGI like by 2028-2030.

That acceleration caught me a little bit off guard, but that’s where the wild cards always come in.

Jack: Was that like the most interesting example [00:16:00] of AI acceleration from the past year or was that like another, insane example that you’ve seen this past year? We thought, wow, that is incredible. I cannot believe that happens.

Matt: So I think probably one of the most, probably the most incredible development was realistically. And its ability to simulate 300 million proteins because that’s, if you step back about a year and a half ago, artificial intelligence was nowhere near being able to do that. You step back a year ago and it was gonna struggle with that a lot and then suddenly bang, it’s done it.

So I think when you actually have a look at the use of ai, basically to develop synthetic proteins and to model proteins. That’s got the biggest societal, but also industrial impact as well, that a lot of people I think are actually underestimating. Yeah. When you actually have a look at, when you have a look at AGI, so for example, chat, GPT has a thousand times more general knowledge than any human.

It’s learning at per rate 300 million [00:17:00] times faster than us. When you get to AGI, where AGI is defined by open AI’s board. As an artificial intelligence that is capable of outperforming all humans at all economically valuable work, that starts taking it to an interesting place. Then when you bring in OpenAI and Microsoft’s commercial definition of AGI, where they define AGI as the point in time where artificial intelligence is able to generate 100.

Billion with a ‘B’ billion dollars worth of profit by itself. That’s just staggering. I also came across a video of Sam Altman talking to JP Morgan which you did a little while ago. So since 2019 I’ve been talking about the use of AI to build companies, and finally we’ve got a soundbite basically of Sam Altman actually in this JP Morgan investor conference talking about the point in time where AI.[00:18:00]

A single entrepreneur build a multi-billion-dollar business before AI then builds its own multi-billion-dollar businesses. So when you actually have a look at some of the developments that we’ve actually seen, some of the things that I’ve been talking about really for about the past five to 10 years, and now starting to actually become public knowledge and.

I think when you scratch the surface on what AI is really able to do today, so for example. We had Hyperganic in Germany and Leuk 71 in Dubai that used artificial intelligence to produce a new rocket engine in anywhere between six hours to a day. When you have a look at some of the transformative things that artificial intelligence is actually doing, I think the technology’s being underestimated.

Hallucinations are still a problem, for example, when you look at open AI GPT-4, about 5% of responses were. Mistaken, should we put it like that? Or [00:19:00] hallucinations with GPT five, that’s now down to N 0.6%. So you have a look at the use of artificial intelligence, for example, to develop new semiconductors.

We’ve seen an artificial intelligence that developed the world’s fastest computer chip, but it did it in a fundamentally new way where even the IEE E experts just went, this computer chip works. We don’t know how. And it’s when you have a look at Nvidia, Nvidia go on public record now, they, no way they could develop the latest generations of GPUs without ai.

And then we saw an artificial intelligence engineer recently develop a hundred new semiconductors, pretty much in the blink of an eye as well. So I think, when we look forward to 20 26, 20 27. I think some of these shadow cases that really haven’t actually made the press or haven’t really made the headlines in the way that they probably should have, I think they, they start becoming more visible to the general [00:20:00] public.

Brian: Last year we talked a fair amount. You talked a fair amount about agent ai. How did that play out in 2025 and where do we go from here?

Matt: About 10 years ago. There were lots of people 10 years ago talking about, in the future we will have a personal digital assistant, that we can talk to that will do things for us and on our behalf, shall we say, 20 24, 20 25.

We give those things a new name and we call them artificial intelligence agents. So on the one hand. I think that the, I think overall basically the industry, the technology industry is racing far ahead of where it’s actually, of where it actually is. So I’ll give you a couple of examples of it. When you have a look at things from an enterprise adoption perspective, the vast majority of organizations are only really just getting their head around how to implement and where to use generative artificial intelligence.

If we have a look at retail, [00:21:00] 40% of retailers are using generative ai. 15% of chemical companies are using generative ai. 30% of automotive companies are using generative ai. These sorts of things. So when we have a look at ai, a lot of the vendors that I talk to, the adoption rates are still actually low.

But some of the things that the AI agents can do are actually quite interesting. Walmart has been using AI agents to do all sorts of fancy things, when we have a look at Agen ai some of the big problems that we have with Agen AI is everyone is talking about it. I went to Davos in February, and every single organization, like KPMG, like Deloitte, like Infosys, everybody was talking about Agen ai.

Increasingly we need to give AI agents their own id, which is something that Okta has been talking about because if you are using an Agentic workflow [00:22:00] to do work for you, what happens if one of those agents actually turn out to be on an on a North Korean server? What happens if those agents are compromised?

So when we actually have a look at the security of agents, that’s quite lax. When we have a look at how we can compromise them and their behaviors, that’s fairly straightforward. When we have a look at organizations that are adopting agent ai, it’s still really low. When we have a look at the amount of noise around AI agents, there’s a lot of noise around AI agents, let’s face it.

But when you have a look at the press, for example, and go and have a conversation with people, or you watch YouTube videos or listen to podcasts. The main use cases, the main use cases of AI agents that everyone still talks about is booking your flights for you. Those sorts of things still show, still tell me that We’re in the, we’re in the very early innings of agenda AI still, and it’s going to take probably a couple of years before it really start when we think [00:23:00] about commerce.

So we touched a little bit on last year as well as payment. There are more people now starting to use AI agents as part of their shopping journey than ever before. And actually the ramp up of AgTech Comm has gone from like zero to 40% of Americans in a year. So when we look at AgTech commerce especially, that’s where agents really seem to be playing quite a pivotal role in disrupting an industry.

But when we have a look at the rest of the use cases. Organizations are still general trying to wrap their head around, generative ai. And then we have companies like Microsoft that are developing you cyber marketplaces. So there’s still a huge amount of build out around agents and we’ve got a long way to go I think before we trust them and our enterprise.

Omkar: I have a question and it pans out into my [00:24:00] next theme that I wanted to focus on as well. We’ll start off with agentic AI first, but I remember when publicity first introduced their new AI browser, I think it was called Comet, right? Yeah. And they showed this a video where you put a voice command in that browser.

It goes through Amazon compares the products that you want and makes the payment as well. Or if you have WhatsApp open on that browser, you can put a voice command to it and it can send a message to help whoever you want, right? And it’s incredible what it can do and the capabilities of how far it can reach.

My question over here is, what about security? When we look at security, and because I, I bring this up most often because I remember Chad GT at one point had made ta GT conversations available for Google search. So when someone is looking for something like this and agenda AI starts putting out information about what you’re doing in your private life online and the security factors [00:25:00] is minimal,

Matt: so from a privacy perspective, it now looks like artificial intelligence is able to gather 165,000 points on every individual. Now, these include things like, online behaviors, offline behaviors. The speed at which you type, for example, indicates your emotional state. The words that you use indicates your emotional state.

The language you use indicates your emotional state, what you’re generally talking about. Basically, we can pull a lot of psych, psychometric database from that, the psych that you are using. We can pull a lot of behavioral data from that. When you actually think about things from a privacy perspective, the vast majority of people are putting more personal information into a chat bot than they ever put into a Google search bar.

For example, if you were, if you had a health query, then traditionally with Google, you would say, I have a headache. What could the, what could be [00:26:00] causing this? And you’d get something like, WebMD up or whatever it happened to be. And then you’d go and search something else and Theis would have to try to piece together what were you really trying to get at?

Whereas now basically you’re increasingly putting financial data. Into these chat bots, you are putting your personal aspirations and dreams. So the number one use case for AI this year, according to Harvard and different sort of research groups, was actually the use of artificial intelligence for companionship.

The second I see was for life advice. Organization, life, organization. The third was individuals trying to use artificial intelligence to help them understand their life’s purpose. So when we actually have a look at these ais, the data that we’re actually putting into these things. Goes way beyond what we’d have done with a Google search bar.

If you are trying to use an artificial intelligence basically for financial advice and in the. 48% of adults are now using artificial [00:27:00] intelligence to help them with their taxes, finance, and investment decisions. Increasingly you’re saying, I’m this kind of person, I’m slightly risk averse.

I’m interested in these kinds of stocks and shares. This is my income. This is what I, this is how I believe the world is working. What should I invest in? How should I manage my finances? How can I develop the most tax efficient? Inheritance scheme for my children, whatever it happened to be.

These systems, increasingly when you talk to the big seven, they are increasingly creating a digital twin of you from a mental perspective, a physical perspective, but also basically from an emotional and behavioral perspective as well. And the power of that. You know it out about seven to eight months ago, he said he envisages a point in time where artificial intelligence will remember your entire life.

So when we talk about the use of personal data, these ais aren’t [00:28:00] just listening to you and saying, today you’ve got a query about a headache. Increasingly, thanks to large context windows and gigantic data centers full of hard drives, memory and flash, these AI are going three weeks ago you were talking about, you were talking about a funny pulsing in your right arm.

I’m connecting these two now. Maybe basically you are gonna have a stroke in three weeks time. When we actually have a look at artificial intelligence as a data collection machine, it’s phenomenal. It goes way beyond the big data that, companies like IBM would’ve talked about in, 2010 AI is in many cases increasingly useful and helpful.

Even when we think about legal advice or, in the United States, there was a company that, that tapped into the GPT for at the time, sort of API. [00:29:00] This company was a healthcare company. They went from zero to $3 billion valuations in about a two week period. And all that company did was develop an interface that people could ask you at United States based healthcare questions of, to help people understand better how they could navigate the US healthcare system.

There some seriously interesting use cases of ai. This is where, when you actually have a look at some of the actual real really interesting use cases, we’ve got this $5.7 trillion worth of investment. $1.4 trillion is committed just by open ai. We’ve got AI revenues of roughly about a hundred to $150 billion.

Which is why everyone talks about the AI bubble. You are we in when actually have a look at how. How consumers are actually using ai. A reminder outta 800 million monthly [00:30:00] users for chat, GPT and 680 million users of Gemini, I think about five to 10% of those by volume are actually businesses.

There are a lot of individuals that are finding benefits in ai, even if it’s just as a love companion.