Subscribe
Share
Share
Embed
The capability to map brain activity to language isn’t just another UI shift—it’s a paradigm shift in how humans and machines might communicate. If you’re building systems that integrate or rely on neuroscience-adjacent tech (or even simply storing neuro-derived data), you’ll want to treat this as a strategic early warning: new input modalities, new risk surfaces, and new expectations of what “internal” means.
In this episode of Pop Goes the Stack, F5's Lori MacVittie and Joel Moses unpack emerging research on decoding neural activity into language—turning brain signals into natural-language output. They explore the promise for accessibility alongside major concerns: privacy, “intrusive thoughts,” and how systems decide which signals to surface. With a massive potential “blast radius” if connected to agentic systems, the research serves a stark reminder on the importance of evaluating AI breakthroughs for practicality and risk.
Read the original research, Mind captioning: Evolving descriptive text of mental content from human brain activity: https://www.science.org/doi/10.1126/sciadv.adw1464
Read the original research, Mind captioning: Evolving descriptive text of mental content from human brain activity: https://www.science.org/doi/10.1126/sciadv.adw1464
Read the summary, "Mind-captioning" AI decodes brain activity to turn thoughts into text: https://www.nature.com/articles/d41586-025-03624-1
Creators and Guests
Host
Host
Lori MacVittie
Distinguished Engineer and Chief Evangelist at F5, Lori has more than 25 years of industry experience spanning application development, IT architecture, and network and systems' operation. She co-authored the CADD profile for ANSI NCITS 320-1998 and is a prolific author with books spanning security, cloud, and enterprise architecture.
What is Pop Goes the Stack?
Explore the evolving world of application delivery and security. Each episode will dive into technologies shaping the future of operations, analyze emerging trends, and discuss the impacts of innovations on the tech stack.
00:00:05:06 - 00:00:32:00
Lori MacVittie
Welcome back to Pop Goes the Stack. The show that keeps tabs on emerging tech so you don't have to pretend you read the white paper. I'm Lori MacVittie and no, I didn't read the footnotes either. No problem. Today's chaos comes from a place we all assumed was safely private. Your brain. Because the new new user interface isn't even a screen, a keyboard, or your voice.
00:00:32:04 - 00:01:05:24
Lori MacVittie
It's neural activity. Once we can map brain signals into language, we're not talking about a nicer input method. We're talking about a completely different relationship between humans and machines. So if you build systems anywhere near the space, here's your early warning. The definition of internal is about to change. Today, we tried to ask someone to test this technology and be a guest, but we couldn't find anyone who thought this was a good idea.
00:01:05:24 - 00:01:08:21
Lori MacVittie
So, Joel, I'm afraid it's just you and I today.
00:01:08:23 - 00:01:09:26
Joel Moses
Oh, my.
00:01:09:28 - 00:01:13:29
Lori MacVittie
That's right. So, look out, things could get wild.
00:01:14:06 - 00:01:23:21
Joel Moses
Oh, boy. Plumbing the depths of my brain is not something I expected to do this early in the morning. So, armed myself with a coffee and let's get started.
00:01:23:23 - 00:01:31:26
Lori MacVittie
Let's go. So there is a couple articles about this, and I'm sure, as usual, you've probably at least skimmed them.
00:01:31:28 - 00:02:06:09
Joel Moses
I have at least skimmed them. You know, luckily there's no AI monitoring my brain activity, so you don't know whether I skimmed them or not. So the articles in question that we're talking about, there was an article in the journal Nature and also, Science Advances science.org, that talks about research that's being done into mapping brain activity and the patterns of neural activity in the brain, and trying to apply AI models to it to try to predict the semantic context under which the neural patterns are configured.
00:02:06:11 - 00:02:38:00
Joel Moses
The idea being that if you are thinking about something to verbalize that instead of verbalizing it, a scanner can just look at the pattern of activity and can express itself for you in natural language. It's kind of the next frontier of natural language processing. You know, this is not without some background in medical science. I mean, a lot of work is done in trying to trace out the effect of different activity in different areas of the brain.
00:02:38:03 - 00:03:04:09
Joel Moses
If someone is is having like a brain tumor removed, oftentimes the surgery will occur while the patient is awake, so they can sample reactions to verbalizations or ask you to do a physical activity and that way they know, with probing activity, what areas of the brain belong to that activity. So there is a little bit of a background in medical science for this.
00:03:04:11 - 00:03:22:26
Joel Moses
But I got to tell you, reading the article, this is a different world altogether. This makes a lot of assumptions about brain science that I have tons of questions about. It's kind of scary. What did you think when you read the article?
00:03:22:28 - 00:03:46:27
Lori MacVittie
I thought this is crazy. I mean I had flashes of all, right look, we've all read science fiction. We've seen the movies, right? We know, right, this kind of thing is terrifying, right? It's one like no, I do not, sometimes I don't want to know what I'm thinking. Like I don't, you know, I don't want any
00:03:46:28 - 00:04:19:04
Lori MacVittie
I don't need machines, like going, "Yeah, she's thinking about that." No, please do not share. It just it seems invasive in a way. It seems very risky. But on the other hand, you know, I've watched things like Neuralink, which exist to help people who, right, they're disabled in some way--they can't move, they can't type, they can't talk--that help them verbalize and communicate with other people
00:04:19:04 - 00:04:32:12
Lori MacVittie
and that's an amazing thing to do. So obviously there are some uses for this that are very good. I'm just not sure that this is ever something we want to put into, I don't know, production. You know, like.
00:04:32:14 - 00:04:56:06
Joel Moses
So, yeah. You know, Neuralink is kind of what I started thinking about and for those who are listening to the podcast who don't know what that is, that's one of the Elon Musk properties that's experimenting with implants. Neural implants that monitor portions of the brain and they've had some, I don't know if it's necessarily success, but they've had some interesting research outcomes
00:04:56:08 - 00:05:19:17
Joel Moses
that suggests that people can learn how to activate neural interfaces in such a way that they can express certain thoughts through the neural interface. Which is a different mechanism than just monitoring general brain activity and figuring out from the pattern of activity what someone intends to say or needs to communicate. That's a, this is a completely different thing.
00:05:19:18 - 00:05:47:03
Joel Moses
It's the same basic science, but one is intrusive and one is designed to be kind of on the side. And, again, I have some skepticism here because, like with Neuralink, it's learning over time how to activate the neurons that affect the interface rather than just thinking. And, you know, when I think about how I think, my mind is a little bit like an eight track player, right?
00:05:47:03 - 00:06:23:25
Joel Moses
I have a lot of different thoughts going on simultaneously. I feel like I have a fairly active subconscious and, you know, medical science and psychological science kind of rationalizes that the subconscious does exist. It does drive some of our activity about what, you know, forming opinions and things like that. So which track in that eight track, if you're viewing all your brain activity from remote, which track is this neural model actually modeling? And that's what I find fascinating about this study.
00:06:23:28 - 00:06:42:19
Joel Moses
It presumes that they can pull out a verbal semantic context and that that's really the only thing that this is interested in doing. I have a feeling there's other semantic contexts and brain activity and maybe some of it's private, maybe you don't want people to hear it.
00:06:42:22 - 00:07:02:24
Lori MacVittie
I think that's true for most people. We have things we think that we do not share with anyone and do not want to share with anyone. Right? Sometimes intentionally. Sometimes things just pop in your head and you're like, whoa. Right. I mean, the notion of intrusive thoughts, right, exists for a reason because thoughts can be random.
00:07:02:26 - 00:07:24:05
Lori MacVittie
But what I like about what you're saying, right, so there's a lot of signals going on in your brain and the question is: how does a system like this decide which of those signals are actually the ones that matter to surface them? Which is a lot like security research, right? You have tons of thousands and millions of signals.
00:07:24:08 - 00:07:37:04
Lori MacVittie
How does it know what to go after? Right. So, you know, there's some similarities here, at least in how is it deciding what to pay attention to. There's that attention
Joel Moses
That's right.
Lori MacVittie
idea again, right.
00:07:37:05 - 00:08:04:14
Joel Moses
That's right. And, you know, maybe it only works on people with one track minds. In which case it would probably work on hundreds of thousands of teenage boys out there. But I mean, quite frankly, I don't want to be on stage wearing one of these neural caps and ask it to summarize for my audience what I just said and have it tell the audience: I like pizza, I hate conferences, and that guy in the front row has a big nose.
00:08:04:15 - 00:08:24:28
Joel Moses
I don't want that to happen. So I think that these systems, if they do exist and if pure research validates that this is something that can be done, I have a feeling these systems are going to need just as many guardrails as other AI systems.
00:08:25:00 - 00:08:48:16
Lori MacVittie
And that's an important point, right. We have, I think, some of these systems most of the conversational AI today has adopted that idea of guardrails. Right? Most of us probably heard from our parents, right, when we were young, "hey, if you don't have anything nice to say, don't say anything at all." Right, that's a guardrail that is put in place. Right?
00:08:48:16 - 00:09:16:13
Lori MacVittie
Is it kind? Is it helpful? Is it true? You know, we run these kind of guardrails almost instinctively. We don't actively think about it. It just happens because we've learned that. So guardrails are an important part of the human experience and so it makes sense to have them on systems. But and maybe that's okay, right, the guardrails didn't exist for conversational AI until it was shown:
00:09:16:17 - 00:09:36:18
Lori MacVittie
oh yeah, we need those. Almost like an afterthought. So maybe it's important for this research to actually figure out can it do it before it starts putting in guardrails to go: okay, now we know how it works, now we know how to apply the right guardrails to it.
Joel Moses
Right.
Lori MacVittie
You kind of have to know, right, both sides.
00:09:36:20 - 00:09:56:21
Joel Moses
You know, the other thing I can't stop thinking about is what is this technology really intended to be? Like I can see only one practical application of it. And that is for someone who is nonverbal, giving them the ability to communicate in a verbal sense. So, you know, dealing with someone who has lost the ability to talk, maybe.
00:09:56:23 - 00:10:24:24
Joel Moses
But, I mean, I read some discussion about this technology, you know, being obviously pointed at the idea that we can instantly and in a handsfree and nonverbal form, generate text messages to other people. First of all, I don't know whether that text message would be coming from me or my subconscious. But second of all, is that something that I really prefer over just literally just verbalizing it?
00:10:24:27 - 00:10:48:09
Joel Moses
I mean, it doesn't seem like it's necessarily going to be any faster. I'm going to have to think it and then I'm going to have to say it. But there's no major increase in speed with which the message is sent. There's a lot of uncertainty around it. And, I just don't necessarily know what the practical applications are or the advantages above just pure verbal communication.
00:10:48:11 - 00:10:52:09
Joel Moses
So I'm confused.
00:10:52:11 - 00:11:26:08
Lori MacVittie
You're confused. I think, I mean, it's cool and it's interesting and I think there are definitely applications in certain spaces. I think this is a good example of AI and evolution and what it can do that the demonstrate. It's not always applicable to every task. Right. AI is not something you're going to use for everything. This is not something we want SREs to start using.
00:11:26:11 - 00:12:03:07
Lori MacVittie
I, you know, people express their frustration, "delete everything." Well
Joel Moses
Right.
Lori MacVittie
now, you thought it, you know, you didn't type, it's too late. Right? The it just basically took everything from root. It's gone. That's just not a good way to do this. And we're still struggling with moving from, you know, typed scripts that are executed by humans to hey agents that--and we don't know how it thinks or chooses or makes decisions, not really--is going to be executing things. Like we still have to get there.
00:12:03:12 - 00:12:29:15
Joel Moses
Yeah, I definitely don't want this technology connected to agentic structures for a good long while at the very least. You know, the other thing I was thinking about as I was reading this research is it describes a general mechanism for creating semantic context out of patterns of brain activity that are observed. And it struck me that it's all centered around describing what happens with the language centers of the human brain.
00:12:29:17 - 00:12:54:04
Joel Moses
But then I it occurred to me that there are other possessors of brains, and they also have language centers. They may not be the same. Our pets. Is this the mechanism that I finally get to use to understand what's going on in my cat's head when he knocks my water bottle off the shelf? You know, does he just hate something that sits upright and has to knock it over?
00:12:54:07 - 00:13:07:16
Joel Moses
I'm curious about that. And perhaps something to apply this technology to would be greater understanding of the animal kingdom. That would be an interesting area of research.
00:13:07:18 - 00:13:17:10
Lori MacVittie
That would be interesting. I'm not, you know, I'm not sure I want to know what my dog is thinking.
00:13:17:12 - 00:13:20:05
Joel Moses
Your dog is thinking what all dogs think: feed me please.
00:13:20:07 - 00:13:43:24
Lori MacVittie
Yeah.
Joel Moses
Pet me please.
Lori MacVittie
Yeah, exactly. Mine, mine, mine. But are they thinking, right. Now you're getting into like the metaphysical philosophical realm of are they even thinking? Do they? Right, is instinct thinking? Is there a difference? I mean,
Joel Moses
Well,
Lori MacVittie
these are good things we could learn.
00:13:43:24 - 00:14:13:28
Joel Moses
strikingly, this article doesn't make any comparison to neural patterns associated with semantic meaning or thinking. And I think that's a critical distinction. And in fact, it doesn't even assume the brain is thinking. It simply seems, it tries to recover information out of the patterns that it observes. It doesn't, you know, all it's trying to do is predict what you are going to say based on your patterns, which is pretty interesting.
00:14:13:28 - 00:14:34:20
Joel Moses
It doesn't try to really get into the area of thought. So again, I have a lot of questions about the, first of all, the veracity of the research. Obviously there are a lot of people who can publish a lot of things these days. There's a lot of research going on in medical science related to brain activity in the brain, the human brain especially.
00:14:34:23 - 00:15:02:13
Joel Moses
And so this is really interesting research, but, practical applications still kind of escape me. It is fascinating, though, to see AI, generative AI and, you know, large scale processing and creating semantic structures in graph form being applied to all sorts of things. This is a really radical one. But, I think this is just a harbinger of research to come, frankly.
00:15:02:15 - 00:15:34:10
Lori MacVittie
I agree. I think the bigger lesson from things like this is that everything is changing. AI is changing everything. It is changing how we interface with our systems. It's changing how we operate them, manage them, secure them, how we interpret data, how we do things on a daily basis. It absolutely is changing that. But there are
00:15:34:13 - 00:16:00:13
Lori MacVittie
more forward-looking, maybe even science fiction-y looking, uses of AI that are not going to change how you do things today. They're not going to impact your data center in the next ten years. Just not. And that sometimes we have to recognize that these things, while cool or interesting, and they are fascinating are not really practical.
00:16:00:14 - 00:16:18:04
Lori MacVittie
They can't be applied. And not all AI is really useful for what you're doing. So we should be able to clearly call that out, to be like, that's not, no, we're not doing that here, kind of thing.
Joel Moses
Yeah. Absolutely.
00:16:18:10 - 00:16:46:20
Joel Moses
Yeah. I think it's probably decades before we see anything necessarily meaningful coming out of research on this level about brain activity. But you know, in the meantime, no one should be prepared for their CEO, you know, creating a company strategy by mere thought. I think that that's safe to say. It's more likely to tell you that he's thinking about tacos for dinner.
00:16:46:23 - 00:17:26:01
Lori MacVittie
We should just cut it there, because what else, I mean, do we say about this? I think this is interesting, but yeah, I mean, we're just finally getting past GUIs to APIs so that we can get to agents. And this is like 20 steps ahead of that. So, not all the news that you hear about, you know, the great things AI is doing are real or applicable, and you should probably very carefully evaluate them for practicalness and also blast radius.
Joel Moses
Oh, absolutely.
00:17:26:03 - 00:17:51:07
Lori MacVittie
I think this has got, yeah, huge, huge blast radius. So take some of this stuff with a grain of salt and, you know, go back to how can I apply this to accelerate what I already do? To automate, to be more productive to, you know, find problems faster, and not create problems maybe. Right?
Joel Moses
Right. Yeah.
00:17:51:13 - 00:18:08:29
Lori MacVittie
Don't go looking for trouble. Oh, yeah. I think that is a wrap for Pop Goes the Stack this week. So hit subscribe so you never have to fake a white paper opinion again. And we'll keep reading the citations. You keep looking brilliant.
00:18:09:01 - 00:18:09:13
Joel Moses
Take care.