Neurotech Pub Episode 8 – The Drinks Bring Back All The Memories

Guests: Dan Rizzuto, Nanthia Suthana, Nick Halper, and Gyorgy Buzsaki

The Drinks Bring Back All The Memories

In this episode, podcast host and Paradromics CEO, Matt Angle brings together memory-researchers Nanthia Suthana (Assist. Prof. of Neurosurgery and Bioengineering, UCLA School of Medicine) and Gyorgy Buzsaki (Biggs Professor of Neuroscience, NYU School of Medicine), and scientist-entrepreneurs Dan Rizzuto (CEO of Nia Therapeutics) and Nick Halper (Co-Founder of Braingrade) to discuss memory and neurotechnology.

Like many of you I approached, and to some extent still do approach, the concept of memory enhancement with skepticism. But the conversation today is going to be a grounded one, and I think you will see that there is some real science here that can give us reason to be cautiously optimistic about the future of memory and BCI. I hope you enjoy the episode.

– Matt Angle, CEO, Paradromics


Watch The Episode

38:05 | Predicting the Future: BCI to Decode/Reconstruct Memory
42:13 | Ray Kurzweil
42:33 | Famous Bad Predictions
42:48 | Henry’s Scale: Bluebrain Predictions
44:07 | Simonides
44:16 | The Art of Memory by Frances Yates
44:23 | William James, Primary and Secondary Memory
46:22 | Reconsolidation


Read The Transcript

Matt Angle:
Welcome to a new episode of Neurotech Pub. Today, we’ll be talking about memory. I’m joined by two scientist entrepreneurs, Nick Halper of Braingrade, and Dan Rizutto of Nia Therapeutics, both are developing BCI platforms for memory applications. I’m also joined by memory researchers Gyorgy Buzsaki and Nanthia Suthana. Like many of you I approached, and to some extent I still do approach, the concept of memory enhancement with skepticism, but the conversation today is going to be a grounded one, and I think that you’re going to see there’s some real science going on here that can give us reason to be cautiously optimistic about the future of memory and BCI. I hope you enjoy the episode.

Matt Angle:
Well, to get started it would be great if you could each introduce yourself, and then by way of introduction also tell us about how you first got interested in memory.

Dan Rizzuto:
Well, I’ll start. Dan Rizutto, and I’m the CEO of Nia Therapeutics, and my background is in human memory, brain machine interfaces, clinical trial design and operations. I’ve been doing the human memory thing, well on and off since the beginning of my career when I started out at Brandeis University as a young graduate student and I thought that I was interested in computational neuroscience. In fact, I was interested in computational neuroscience, and I worked with Larry Abbott at Brandeis during the late ’90s. Then Eve Marder recommended that I train with a young professor by the name of Michael Kahana who was pretty much unknown at that time, just starting out in his career. He was just a few years older than me, I think he might’ve been 23, 24 and I did a rotation with him and was really just incredibly excited by the field. Together we got into this field of collecting intracranial recordings from neurosurgical patients while they performed memory tasks.

Dan Rizzuto:
I did my dissertation on that topic, looking at the neurological underpinnings of human memory. It was really just a grand adventure understanding the human brain and the human memory system, which is so fundamental to who we are as people. That’s how I got started.

Nanthia Suthana:
Nanthia Suthana from UCLA and I first got interested in memory as a 19 year old. When I took a year off from school, which technically I dropped out of school but, you can say I took a year off now because I did eventually go back because I was very interested in memory. I had stumbled upon some books and was reading for fun as I’d always been interested in the different ways people learn and retain information. And I read about cases like, HM and individuals who have this amnesia and loss of memory, and just became fascinated with trying to understand what’s happening in the brain during this whole process and went back to school and here I am.

Nick Halper:
So my name is Nick Halper. I’m actually a cellular neuroscientist by training. So in some ways maybe I don’t don’t belong or something, but I found myself as an EEG lab manager coming out of school. It really got me into electrophysiology and some of the challenges but also like promise of it, both as a research tool but also as a product. While I was in school and running this EEG lab, I had always felt this like calling towards Alzheimer’s disease as something that I should be solving. Like it was a big attention and focus of mine in my cellular science research but I ended up getting a bit distracted. I joined a company called Blackrock Microsystems in a more of an engineering role because the call of neuroprosthetics is also very exciting. During my time at Blackrock I worked my way from a more engineering role into a product and business strategy role, helping to pick and choose the technologies that Blackrock commercialized.

Nick Halper:
Through that came to this kind of experience in medical devices, and really wanted to get out and work on a medical device of my own. That’s when I met Peter Schelecht who had talked about getting into memory as a therapeutic target and reaching back to my Cellular Neuroscientists days. It was an indication I was really excited about, memory is just almost the critical function of the brain. It makes us who we are and allows us to really do a lot of the daily tasks that we do and so it was a natural reaction to want to jump to it.

Gyorgy Buzsaki:
I’m Gyorgy Buzsaki, I’m a professor of neuroscience at New York University School of Medicine. I started out as a high school person who was interested in radio communication. I was a radio ham, I built tuners and receivers, and transmitters. Coding as a problem interested me very much. Also how you send signals with high fidelity from one place to another. When I ended up in medical school, I was lucky to stumble upon a person and found a person who was doing extraordinary, cool research on hippocampal theta oscillation. It was professor Grastyan who introduced me to the most important structure in the brain, which is the hippocampus and its writtens. I came to the United States to do a Postdoctral fellowship. I then went to Canada, I worked with another prominent hippocampal researcher by the name of Case Vanderwolf. One day that was back then, when most of the recordings were mostly just local field potentials, very few people recorded from units, but I was trying to record from units. One day I heard this very powerful pattern that we now know as the hippocampal sharp wave ripple.

Gyorgy Buzsaki:
That was a very unusual pattern and I was looking for some function for it. It just turned out that in the 1980s, the idea that long-term potentiation as a physiological mechanism, maybe related to memory was prominent. The big question was whether long-term potentiation as a physiological phenomenon has anything to do with memory. There were a couple of using requirements such as, it has to be a very fast powerful induction oscillation mechanism, typically 200 Hertz. There it was, my ripple that had 200 Hertz frequency. So all of a sudden I realized that this may be important. This is how I started reading very late about Brenda Milner and others. I learned about memory and things like that, mostly from the Soviet literature or Russian literature and Polish literature. But then I realized that it’s a different world. We found our ways differently, but we ended up interested in the same problems now.

Matt Angle:
Thanks, to start this conversation, there are a lot of different types of memory and some of the people who are listening now aren’t that familiar with the different types of memory. I was wondering maybe Nanthia, could you walk the audience through essentially the types of memory so that when we’re discussing different types of memory, they have a sense of what we’re talking about.

Nanthia Suthana:
Sure, and feel free to chip in, interrupt me here other panelists, because we’re all experts here. So different types of memory, we were just talking about the hippocampal sort of dependent memories or declarative memories, which include memories for events in your life, everyday events, like what you ate for breakfast this morning, what did you do for your birthday last year, so on. Also perhaps memories for facts that don’t have this episodic nature and time and place that in order to form them, it’s thought that hippocampus is important. Then there are other types of memory that are not dependent on the hippocampus, these non-declarative or implicit memories that are formed, that are able to be easily verbalized and this can include things like learning new skills, like ride a bike, how to ride a bike, how to play the piano, dance, et cetera and are more dependent on these basal ganglia systems and straddle regions.

Nanthia Suthana:
Then there’s other types of memory also that involve emotion and fear, if you’re conditioning that maybe involved in memory disorders like PTSD or Post-Traumatic Stress Disorder and other working memory, fast short-term memories that we rehearse in our minds, in the old days, when we were trying to remember seven digit phone numbers, it was relevant and now not so much but for other things, and those are a few, I’d say but there are other types as well.

Gyorgy Buzsaki:
If I may add something to what you just said, you can see from Nanthia’s description that definitions of memories started out a long time ago, and it started as everything else in neuroscience with the interest of not in the brain but in the soul originally, and then in the mind. The definition starts usually that, can be declared something verbally and if it can, then it’s declarative memory, Then we go and so this is the distinction that we make today. But if you think about it for a second, this cannot be right because if we approach it from the language point of view, then why are we doing research on animals who cannot speak. So the declarative memory is a good approach, how the mind generates remeniscense and memories, but it’s not a good way to ask what the brain actually does.

Gyorgy Buzsaki:
Over the years of course, when we realized that the hippocampus has something to do with it, then we started branching out. If you look at today’s literature, you will find that the classical definition that you read in every single book is hardly ever used by people like Michael Kahana for example, it deviates from this. Now if you go back to the ancient times, then typically we have memories with two different types. The first type was true memory and the other one is artificial memory. Everything that we talk about today belongs to the category of artificial memory, because these are acquired memories that we learn. But according to Christian philosophies, even before that, these were not true memories because true memories are there and there’s a property of the soul, and the soul is forever. Therefore it existed before we were born.

Gyorgy Buzsaki:
This is an origin of this as the Platonic view of the mind or the soul. It took about almost 2000 years to abandon that. They said, “oh, there is no soul.” Therefore there is only in the brain. Then how do we research it, so we started to put electrodes in the brain and so on but our vocabulary, as well as the mindset remained exactly the same. Recently, well we have written about this a little bit and it will be out in the annual review of psychology next year. I’d like to distinguish between two types of memories. One what I call single utility memory and the other one is multiple utility memory. Instead of talking about the declarative and the working memory, which is working memory is also defined by the mind. You have to keep it in mind according to William James, but first you have to define what the mind is in order to understand what working memories and working memory is such a badly defined, physiologically unapproachable thing because it is the same thing as attention, the same thing as consciousness and so on.

Gyorgy Buzsaki:
It’s just a hypothetical thing but if you look at the memory from the point of view of its consequences, memories are only as useful as you can do something with it. That is you use it or look at it from the point of utility, working memory thing that we are talking about are useful because we use it once, and once we use it we want to forget it. There’s no point remembering the numbers of this Zoom site forever because we use it only once. If I remember this next time actually it interferes with my memory but there are others that are extremely useful because it’s for the entire life of the organism.

Matt Angle:
I was thinking of the pin for the bathroom at Starbucks. It’s a good example of working memory that you don’t have to remember for very long.

Gyorgy Buzsaki:
The interesting thing, of course, is that how come a lot of information, I don’t know what information is but I’m using the word as you do, comes into the brain and some of them are stuck, and others will disappear? The most interesting part of that is how we erase or select how the brain selects what should stay and what is not necessary to be there for a long time. The reason why the working memory issue is not so useful because working memory if it is used, even in the traditional way, people call working memory as something that happens for 5 seconds, 50 seconds, 5 minutes. But where I parked my car is a classical example. It’s also a single utility thing, but it can last for two days or even longer. The time is not a good way of separating different kinds of memories, but it’s always the outcome that is something we have to keep in mind. This is a long answer to your short question.

Matt Angle:
Can I ask now that we’ve elaborated on a lot of different types of memories and is there someone who could talk a little bit about the localization of memory? There’s a lot of thinking that different types of memory are localized to different types of structures. Could someone tell us the contemporary viewpoint on that?

Gyorgy Buzsaki:
Well I can react to that and say, where do you get your information from? One way of doing it is to open up a big book, let’s say Kandel’s book and say, “Oh, what kind of memories do we have.” Over the years, memory became almost synonymous with plasticity and plasticity can be defined as a change of synaptic strength, but synaptic strength change is used for many things. Everything that happens in the brain is synaptic strength change. There are as many memories if you want, as many synapses or as many circuits. There is a muscle memory or kinetic memory, or you can call it a skill memory and they say, where is it? Well, you put it in one box, it’s said a motor system. What about something else that upsets my or changes my heart rate? Oh, we can call it emotional memories.

Gyorgy Buzsaki:
Oh, where are the emotional memories? Oh, of course the box is the amygdala and where are those things that we declared verbally? We said, well, let’s think about maybe prefrontal codex, maybe hippocampus, or the interaction between the two and so on. This is the approach. We approach everything, not from the brain point of view, but from the outside world point of view and we tried to do the taxonomy without knowing how to ground these words that we are using. Perhaps it’s time, maybe this is a good forum to start. How do we reverse it and start with brain mechanisms and say, what is the memory that in 2021 we all agreed that we have to research and do we have to have 50 different types of categories or the 12 types that is in that Larry Squire’s book. Or we say that anything that is, I’ll be happy with the idea that anything that can change is memory.

Dan Rizzuto:
I like that approach Gyorgy and I wonder, I would also add and I wonder how you would react that it should be grounded not only in how the brain operates, but a deep understanding of behavior, and that our goal is to link brain and behavior. I agree that the vocabulary can be misleading and we shouldn’t get too caught up in the labels, but there are some really basic fundamental principles of memory that any brain-based explanation of memories should be able to explain. I’m thinking of some principles of basic findings in the Human Associative Memory Literature, such as the conditional response probability, where you’re more likely to recall items that are nearby in context, be that temporal context or semantic context or spatial context, but there’s a contiguity effect whereby if you let somebody recall in free association order, they’re more likely to recall events that are close in time, close in space are close in semantic space. That’s a fundamental property of human memory that the brain-based approach should be able to explain.

Nanthia Suthana:
To also add back to this question of localization, having just finished teaching undergrads and this topic precisely it’s a real challenge because traditionally when I went to college and learned about memory, multiple memory systems, and what were you alluded to in Kandel’s book, right. There are specific regions that are responsible for these functions in memory, which is a narrow sort of viewpoint and modernized. Now contemporary views would suggest that it’s in multiple areas and networks of regions can be involved in complex ways and these different memory systems. It’s a challenge to one hand try to teach students about these systems and the regions that are supported, but also provide them with a sense of the complexity of multiple regions interacting. Like you said, prefrontal, and you have the hippocampal areas and getting away from this one site, one role kind of viewpoint that’s been the tradition for decades.

Matt Angle:
In the very beginning you mentioned HM and I mean that formed a lot of people’s thoughts about localization of memory. Could someone briefly tell us the story of HM and how that informed people’s thinking rightly or wrongly about the function of the hippocampus?

Nanthia Suthana:
Okay. I can just, HM, Henry Molaison who we know now his name given he has passed away in the last few years, and was studied for many decades by Brenda Milner and Larry Squire and others. I guess the case was that he was an epilepsy patient who has intractable epilepsy. Having multiple seizures that are not reacting to medication or standard forms of treatment and so he underwent surgery to remove the medial temporal lobes bilaterally. At that time, there was no real risk to removing those areas. At least we didn’t know what would happen behaviorally or cognitively. When he had the surgery, his seizures were reduced, his epilepsy was improved, but unfortunately he was unable to form these new memories for events in his life and he also lost some recent memories from his past but old memories were still there and intact.

Nanthia Suthana:
it led to this view that the medial temporal lobe and maybe somewhat the hippocampus, although his resection was quite large, not limited to the hippocampus is really required to form these memories, these declarative memories for events in one’s life. His brain was studied and it was also realized that he had some tissues still intact, especially posterior regions but there are a lot of viewpoints now in terms of what regions in the medial temporal lobe are really important and critical for this type of memory function that we’re discussing, which is episodic like memories for events.

Gyorgy Buzsaki:
This seems like a very interesting case. And they said, “oh, memory is now grounded because we found a structure in the brain whose removal induces problems. “The interesting part to me and probably for many others is that how HM was misinterpreted. Again, open up any book and it will say basically that’s a nice situation because HM remembered everything, a lot of things before his surgery but didn’t learn anything new. The truth is that is absolutely correct because HM could never, ever recall an episode from his previous life. That already mentioned about time and space we’re going back to Endel Tulving who coined the term episodic memory and the way it has been simplified. What he said over the years is that, what happens to me, where, and when, or that was even more simplified, what happened, where, and when, and of course these are the animal researchers who use this kind of definition.

Gyorgy Buzsaki:
But when I say, what is the distinction between these two sentences, first world war started in Sarajevo, I can give you a date in 1917 or the other one is that the world trade center was born on September 11th, 2001. Now both statements have one thing in common where, when and what, but none of us actually have any personal experience from the first world war because we were not born yet, but many of us do have a personal involvement in 9/11. This brings us back to the key part of Tulving’s definition, what he calls outanotic. That is, I am part of it, this is my memory, this is something that I experienced. HM could never narrate, he could never recall a birthday party event or a fight with his friends and so on.

Gyorgy Buzsaki:
He got what we call a combination of semantic information. Of course this makes life so difficult if this is the definition that we live with for people like me, who are working with mice and rats because there is no outanotic aspect of it. But if you look at, from the evolutionary point of view, we said, what is the main distinction here? Is it the conscious part of it? Or I say, no, this is mine, I am different from somebody else. The definition of me being a different aspect from everything else, this is the part that continues throughout evolution, animals distinguish themselves from everything else. From this point of view, you can approach memory much better than from the definition of what happens to me, where, and when, and including the, to me aspect and the outanotic aspect. In other words, the connection I made without the hippocampus, you cannot put together a sequence of events, whether it was you or it’s old, you are not capable of doing that.

Dan Rizzuto:
I would agree and I’d build on that, one of the fundamental misunderstandings of the HM work was that mainly in the popular literature and in the layman’s understanding of memories is that hippocampus is somehow the, be all, end all of memory. It’s true if you lesion the hippocampus, you have profound memory deficits but these are specific to being able to recollect into an episodic form, this auto Noetic narrative, but there’s still other forms of memory. I would say that in my view, the hippocampus is the glue that binds different aspects of memory together into a personal narrative, but it is not just where all memories are stored. There a particular memory may include sites that activate visual cortex, smells that activate and sounds that activate the auditory cortex, and the hippocampus is that glue that binds it all together into a single episode of your life that allows you to re-experience that episode. With large lesions, you can disrupt ability to re-assemble that original memory even though the particular aspects of memory may be very much alive in visual cortex, auditory cortex, et cetera.

Matt Angle:
This would be an interesting time to bring in a question we had from one of our previous guests, Mathew Kaufman, who’s at the University of Chicago. He’s a motor researcher. We asked him if he had any questions about memory and he had this question:

Matt Kaufman:
The thing that has always really baffled me about making new memories is how do you get a network to recapitulate its activation. And so people have worried about how do you build something that looks like the hippocampus that lets you do this mapped plasticity. Let’s say for sake of argument, that you activate it once and it captures that activation and it will reactivate itself that way in the future.

Matt Kaufman:
But we know that this activation spreads out to cortex and to other parts of the brain. Where when you have a memory, you get reactivation all over the brain that is recapitulating some of the sensory stimulus, but the plasticity isn’t occurring in cortex, at least not immediately. But you can retrieve a memory immediately. So how do you build a system such that you don’t just have somewhere that can recapitulate its activity? How do you make it so that it can do that? And then it can recapitulate the activity that drove it to have that activity, that’s distributed across the brain. This is something that I’ve yet to see a theory at least that I understood that explained it.

Gyorgy Buzsaki:
Let me give it a try. So if you go by the definition that episodic memory is what happens, where and when, and forget the me for a moment, then that was a extraordinary moment in neuroscience. Because having accepted the definition, neuroscience had a roadmap, how to figure out the neurophysiology or neuroscience of memory. All you have to figure out is where is the what stored in the brain, where is the when stored in the brain and where is the where stored in the brain? And this is exactly what happened. We said, “We found the where, because the hippocampus is the spatial device.” And many laboratories looked at it and a Nobel prize was given for that.

Gyorgy Buzsaki:
Recently time also conversion to the hippocampus. They said, “Okay, well there are time cells in the hippocampus.” And then we can figure out from various patterns, not only where the animal is, but where the animal is heading, whether the right or left hand, left part of the tibia is, or radial is. So, hooray we have got the internal hippocampus system where the, where, what, and when converges. And this is an extraordinary simplification in terms of the computation that was just asked about.

Gyorgy Buzsaki:
Because instead of every single episode that we had in our lifetime, now we have to adjust separately store in one box. The what, another box though, the where, and then a third box the when. And then when we have to recapitulate the memory, we multiply the marginals and there we go we’ve got the three aspects of the memory and we have it. Well, I have extensively written about this problems that there is no such thing as where and there’s no such thing as when in the brain. Time is not made by the brain. Time is not sensed by the brain. There are sequences that are present. So if you throw out the time and space as such, then how do you make the simplification? One possible approach is that you can conceive the hippocampus as a librarian, or you can call it a search engine who’s like a search engine or like a library. And it doesn’t have a lot of knowledge, but it has the necessary sufficient knowledge to point in the neocortex where the information is residing, such as in books in the library.

Gyorgy Buzsaki:
So the fundamental mechanism of the hippocampus is generators sequences. And the reason why it can do it so effectively, or more effectively than any other structure is because the hippocampus is a single giant cortical module. Unlike the neocortex, which is a modular system that can replicate many functions, pretty similarly, hippocampus just grows and then it keeps connections with every part of it. And if you are looking for a big random graph in the brain, then the hippocampus is your best bet. So we can go from any neuron to any other neurons in hippocampus in just two steps. And that takes about a hundred milliseconds, which is the time of a hippocampal theater cycle.

Gyorgy Buzsaki:
So the hippocampus generate the sequences and it is capable of generating those sequences in a way that can be tied to learning and it’s capable of pointing to the relevant sources in the brain, in the neocortex. So we have a enormous savings because you don’t have to code and decode as well as store all the memories, all the events that happened to you, all you have to remember is the semantic patterns and how the semantic patterns can be tied together by something. And the something is the hippocampus.

Nick Halper:
In that way could you think about the hippocampus as some endogenous amplifier effectively, that sits there and replay circuits until you get that plasticity, in a more general sense spread throughout the brain so that you create these associations between these kinds of semantic events?

Gyorgy Buzsaki:
Yeah. That’s a nice metaphor. So it’s a sequential amplifier.

Nick Halper:
Sure.

Matt Angle:
What do we know right now about the function of the hippocampal circuits and what I mean by that is, suppose that you had the ability to read and write from 1 000, 10 000, a 100 000 neurons. What could you decode? What do we think we could decode based on our understanding today? What do we think we could encode? Is there a mechanistic understanding of those circuits right now? Or where are we?

Nick Halper:
This is through direct hippocampal access?

Matt Angle:
Yeah.

Nick Halper:
So you can’t cheat and stimulate sensory neurons going into hippocampus or anything like that?

Gyorgy Buzsaki:
I don’t want to talk too much, but I have ideas here. Mainly that if you… To answer the question the way you ask, it’s very difficult, because it already presupposes that they shovel information into the brain and the brain is there to absorb knowledge. And again, this is a funny way of thinking about it because this is what’s called the tabula rasa idea, that the brain is there as a blank slate and we fill it up with information. And from this perspective, it’s very difficult to answer your question.

Matt Angle:
Because you think of the hippocampus more as this kind of lookup table that points to information and other areas. And so you feel without the understanding of the cortical areas and what’s being encoded in the cortical areas, the hippocampus would be a meaningless graph. Is that your position here?

Gyorgy Buzsaki:
Perfect. Yeah. So what you can say is that the hippocampus is part of the brain, is already there and it can generate enormous amounts of sequences. Even a rat hippocampus can generate many sequences, but the richness of the hippocampus is not in the hippocampus, but it’s the reader. The same sequence can be dealt with by five different ways. And this is why the neocortex is so enormous, much bigger in the human than in the rodent, because the job of the neocortex is to observe and suck out that information that exists there.

Gyorgy Buzsaki:
So instead of thinking that how we generate sequences, you can say that the fundamental goal of the brain, or the fundamental task as the brain, is maintain its own dynamic. And whenever we lose something or not, it doesn’t change the dynamic, the dynamic should stay and does stay pretty much the same. So learning is not making the brain or hippocampus much more complicated, but it’s a lookup table, as you say, and then we can choose any of these lookup sequences and link it to something interesting that happened out there. So learning to me is more like a magic process rather than a building from brick by brick, a new house.

Dan Rizzuto:
Something interesting. So I think what this makes me think of is the difficulty of writing a new memory into the brain, which might’ve been… I think it’s certainly in the popular psyche right now that people who are building these memory prosthetics are able to write memories into the brain. And I think Gyorgy is pointing out that the difficulties of doing that, at least from a hippocampal perspective, you might need electrodes in sensory cortex in order to do this. But there is the fundamental process of the hippocampus is the fate of rhythm. And this is thought to underlie associated information processing not only in the brain, but also… Not only in the hippocampus, but also in the larger brain networks and the interactions between hippocampus and neocortex.

Dan Rizzuto:
And it could provide a physiological mechanism that supports distributed cellular activity, synchronized activity, and allows a communication channel between neocortex and hippocampus. And so one possibility. And it doesn’t require tens of thousands of micro electrodes, maybe perhaps just a few macro electrodes. But one possibility, very intriguing is to magnify learning by modulating these fundamental mechanisms, such as the fatal rhythm in the hippocampus and elsewhere, rather than a more specific detail of a memory, rather than writing in details of a memory, but rather amplifying the learning process itself.

Matt Angle:
I’d like to get from each of you a prediction into the future, and I’m going to make your Gyorgy last because he’s too influential. And I think he tilts the group. And let’s look forward in the future. What is the date where there’s a 50% likelihood that the following is possible? So I’m asking you to look forward to a time where you think… June 2029. I think there’s a 50% likelihood this would be possible. When do you think there’s a 50% likelihood that it would be possible to decode a volitionally recalled episodic memory based on a BCI readout? Where someone recalls a birthday party and you’re recording from their brain and you reconstruct the birthday party and you can say what kind of cake they had. How far away do we think we are? When’s the date when there’s 50% likelihood that’s possible.

Nick Halper:
So without prior knowledge, not being able to say, “I’ve looked at this person’s brain when they’ve thought about this birthday party before and I’m going to tell you whether they’re thinking about that same birthday party again.” If I’m saying-

Matt Angle:
You can have decoding model. You can have some training.

Nick Halper:
Okay. In that case, I think actually it would be not be so far away.

Dan Rizzuto:
If there’s only two training items then we can do it right now.

Nick Halper:
Thinking about specific birthday party, a memory. And I can record from anywhere in the brain. And I’m looking at these sensory replays and hippocampal patterns.

Matt Angle:
No. I’m not asking you to… I’m saying like open set speech decoding, where you train on phonemes and then the person can say anything they want. In the same way you could build on, let’s say a set of memories, but you ask them to recall a memory that’s never been trained before. Being a little more specific. No cheating in other words.

Dan Rizzuto:
In humans?

Nick Halper:
Yeah in humans. Let’s go with 2045.

Nanthia Suthana:
Okay.

Dan Rizzuto:
25 years.

Nanthia Suthana:
Goodness. I’m more of an optimistic. I mean, this is a hard, impossible question to answer because I think it’s so much relies on the technology. We just don’t have the technology to access the neurons or the signals that would probably be needed to do this, but you may disagree other panelists. But I think there’s a lot of effort right now and to either improving the technology or just getting it to humans where it can be safe and usable in humans. And so if that happens all of a sudden out of nowhere, then we could get there sooner I think maybe.

Matt Angle:
What’s your 50% date?

Nanthia Suthana:
What are we? 2021. So maybe cut that in half, 25 years in half.

Matt Angle:
Okay.

Nick Halper:
2032.

Matt Angle:
Okay. Dan, what do you think?

Dan Rizzuto:
Yeah, I’m probably closer to Nick here. If we assume we have a incredibly powerful technology today, that would certainly accelerate things, but we don’t, and we know the regulatory pathways to developing such novel technologies are long, they’re arduous. They require a significant investment. And we really do proceed at a slower pace than we would like. So yeah, I think 20, 25 years sounds realistic here.

Nanthia Suthana:
It’s definitely more realistic. I’m going for the all of a sudden something miraculous happens here.

Nick Halper:
Our minds might change as we see an acceleration of this technology. I mean, companies like Dan’s and ours are going to be getting products into the brains of so many more people now and looking at memory systems chronically that I think, ask this question again in five years.

Matt Angle:
Okay. Gyorgy what do you think?

Gyorgy Buzsaki:
I think we should invite Ray Kurzweil. He would give you a date and it would be January 17 of a particular year. Kurzweil like all other visionaries and many fortune tellers, always make a mistake, which is they give a date, and those dates are never ever met… Just a second. I brought something from my door.

Matt Angle:
Send us a copy of it we’ll put it on the podcast.

Gyorgy Buzsaki:
I call it Henry’s Scale, this is Henry Markram’s The Human Brain Project. And it says, “We will understand the single neuron in 2005, the neocortical column in 2008, complete rat brain 2014, which we have past, complete human brain reconstruction is 2023. Now…

Dan Rizzuto:
Two years.

Gyorgy Buzsaki:
Yeah. The usual thing in humans is that things are exponential. And you would like to say, “We had a hard time, but now we have the tool set.” “We have this one, from now on it’s a free ride then we will get there soon.” In reality, when you think how much we progressed since HM. Or I would put it in a more historical concept. What do you think? Who was the person who described episodic memory in contexts or in connection with space and time? When was that? We know Endel Tulving of course.

Nanthia Suthana:
Yeah.

Gyorgy Buzsaki:
But somebody before and very well probably already talked about space and time and memory.

Matt Angle:
I don’t know. Are you going to tell us?

Gyorgy Buzsaki:
Yes. So his name, that’s the first recorded name is Simoni Dez from an island in Greece, 2000 years ago. Here is a wonderful book that you would like to read, The art of memory. And when you will read it, you will be surprised how much thinking has been done before William James, before the British empiricists and how much have been repeated and how much we are repeating today. So I don’t want to have a timeline, but I guarantee you that within 100 years, we will not be there to say, “We have a device that can work almost like episodic memory in a human.” I know it sounds pessimistic, but I bet you 25 cents that I’m 50% right.

Nanthia Suthana:
25 cents?!

Gyorgy Buzsaki:
25 cents in 100 years.

Dan Rizzuto:
Another interesting question is, when might we actually see a commercial memory prosthetic, and I think that is much more close and being able to read out. True. You’re absolutely right. And let’s say an implantable embedded memory prosthetic and-

Gyorgy Buzsaki:
Simple problem. Because if I can have this in my class and I can cheat any time, when you say, “Who was Simoni Dez?” In two seconds with the help of my externalize memory, I can outbeat everybody who doesn’t have the device.

Dan Rizzuto:
And yet an Alzheimer’s patient-

Gyorgy Buzsaki:
That’s all semantic. That’s semantic.

Dan Rizzuto:
Right. Exactly. That’s right.

Gyorgy Buzsaki:
That is what we have to strive for, you guys, as technologists. I think the best way to proceed is always the path of least resistance and perhaps the biggest bank. And the biggest bank is in semantics rather than trying to do some spooky stuff such as my own autonoetic recall. That’s a difficult thing. Because every single time I recall my birthday, I add something to it from the present time, this is the reconsolidation problem. Nothing is so unreliable as memories are.

Matt Angle:
I have a question from Mark Schnitzer. I asserted to Mark that there was already some clinical evidence that neuromodulation can affect recall and this is what he asked me.

Mark Schnitzer:
Well, you mentioned the clinical evidence that a memory prosthetic could work and be successful. I would just be keen to hear more about that evidence.

Dan Rizzuto:
Yeah. I think the evidence is really clear. This is the really exciting thing about this field, is that we already have demonstrated proof of concept in human subjects that we can improve their memory. And it’s important to clarify the details we’ve improved. For instance our laboratory at the University of Pennsylvania, which is the technology that my company Nia therapeutics is based upon. We’ve demonstrated that we can improve verbal recall performance. So neurosurgical patients who participate in our studies are able to recall better with stimulation of lateral temporal cortex. Then they can without stimulation of lateral temporal cortex. And these are very rigorously controlled clinical studies, sham controlled, blinded, randomized studies.

Dan Rizzuto:
I think at this point we’ve replicated this and published it in top tier journals pretty extensively at this point. So at this point we’re just refining and extending it. Now we’re going beyond neurosurgical patients. Now we’re in the process. We just published this to medRxiv, not yet in a journal it’s being submitted. But we’ve now extended these results to neurosurgical patients with a history of traumatic brain injury. We’re able to improve verbal memory, verbal recall performance in patients with traumatic brain injury and compared to their own sham controlled randomized performance. So, that’s-

Matt Angle:
Where are you stimulating?

Dan Rizzuto:
Yeahl, this is where there’s a lot of variety in approaches in the field right now, both in terms of the types of signals. So many people are going for what’s called closed loop neurostimulation, where they’re identifying biomarkers in the brain that are indicative of particular memory states and stimulating based on those control signals. And so that’s one type of difference amongst approaches in the field. And the other is in targets of stimulation, where are you stimulating to improve memory?

Dan Rizzuto:
Our team, we came in, we were funded by DARPA in 2014 to the tune of about $24 million to undertake a pretty massive study of neurosurgical patients and brain stimulation and memory. And we didn’t know where the ideal location or target for stimulating to improve memory was going to be. Hippocampus seemed like it would be a good bet given all the wealth of literature that connects hippocampus to human memory. And we did stimulate hippocampus and there were some modest results there, some not so great, but then by chance we happened upon lateral temporal cortex and we found a very reliable effect. So we’re talking lateral, just over your left ear typically. And we found that we were able to improve recall performance between 15 to 20%. And in some patients up to 40%, that is patients were recalling on average, 20% more words on a given list. We use lists learning to gauge their memory.

Matt Angle:
If I were to pick up the Kandel book, what would it tell me that lateral temporal cortex is doing?

Dan Rizzuto:
I’m not sure what Kandel would say. Ojemann would say that it’s involved in verbal memory. He’s done extensive studies of recordings from lateral temporal cortex while subjects, back in the eighties and nineties, maybe nineties, two thousands from again, neurosurgical patients, single unit studies, local field studies in lateral temporal cortex, showing that this area modulates during verbal memory tasks. MRI studies also indicate that there’s a function of verbal language associated with this area of the brain.

Matt Angle:
Is anyone else surprised that… I mean, it works. I mean that the study is there, but is anyone else surprised that it works? Do we have a mental model for what happens mechanistically when you stimulate that cortical area?

Gyorgy Buzsaki:
So I would put my money on hippocampal sharp wave ripples, because we understand it. And last year we had a paper, I advertise also our stuff in Science, where we prolonged optigenetically the duration of the ripples. And we improved memory by the same ballpark figure, 20, 30%. In a simple task of course, the choice is between left and right. And that’s a very simple situation. And I think most of the problems are with these memory enhancement problems, that they are not real world situations and the choices are limited. But if you say, “Where should I stimulate in the brain in order that I will remember every part of our conversation today?” Or even more demanding I would say, that I would recall the key aspects, the most important aspects of this conversation, rather than all the details that are not necessary.

Gyorgy Buzsaki:
And then it’s a difficulty. And what they didn’t mention is that it’ll… I’ve seen 20 posters from Michael, from the DARPA. And all the other processes, negative. And that red was a hotspot that happened to work. And that’s what you’re asking, is that what is so special about that hotspot? And the explanation would be maybe they can trigger something that others cannot. Particularly in hippocampal sharp wave, is a difficult thing, but you can do it with up and down stairs.

Gyorgy Buzsaki:
Other people… Again in rodents have shown that if you couple, this is my ex postdoc, that if you stimulate or couple the hippocampal sharp waves, particularly up and downstairs you can also improve memory. And there are many of these things, but again, all of these are very simple situations and neither case there are hotspots, but almost doesn’t matter where you stimulate. Maybe you happen to stimulate in an intersection of roads, which is most effective, such as an intersection of roads, which is most effective, such as this… In case of Parkisonian stimulation… The subthalamic nucleus that has multiple projections from multiple motor areas.

Dan Rizzuto:
I think that’s right. I mean, we don’t think that lateral temporal cortex is the spot for verbal memory in the brain. And in fact, what we see, is that by stimulating, again at this spot in the brain, lateral temporal, and at the right time, which is during these poor predicted memory states, you can change the larger brain dynamic. So you actually see a global, or somewhat global change in the state of the brain network. From a poor memory state, as defined by our multi-variate classifiers, into a good memory state. We published this in Nature Communications in 2018. So there’s a localized stimulation, that translates into more of a global state change in the brain, through functional conductivity. We think the effect is partly reliant upon connection with hippocampus. And we have additional data suggesting that the degree to which we’re near white matter in lateral temporal cortex. And that white matter is connected to hippocampus… Predicts the effectiveness of our therapy.

Nick Halper:
To add on to that, in thinking about the memory network. Because it is a network, indeed. On the subject of phase amplitude coupling, if you think about the hippocampus as a coordinator of long range activity, in synchronization between these memory areas. It’s impossible to say, “I’m going to affect memory or improve memory in a clinical way,” without really looking at that whole network. And if you can find a hotspot, like lateral temporal cortex, that affects that whole network, then great. But I think this is going to be basically disease-dependent.

Nick Halper:
And then from this, we can take a lesson from pharmaceuticals. We can look and say, “Okay, if I want to make a clinically relevant memory prosthetic, or memory implant, I need to first look at what’s going wrong. I need to understand how the brain is functioning when it’s going right. And I need to look at how that deviates when it’s going wrong. And if I can restore that state, then I can probably restore function.” And that’s what we see, right? When we take these… There’s a whole bunch of evidence, whether its invasive, intracranial stuff that came from Penn, or whether it’s even noninvasive stuff done by Hansen Myer or others. What you see, is that networked coordination of activity, and restoring that network activity at those multiple sites, is what is critical to restoring function. I think there’s a lot of evidence there. Really strong evidence in humans to support that.

Nanthia Suthana:
I’ll add… Just from what I’m hearing, and what the state of the field is, and how much has changed the last 10 years. It seems that there are multiple ways to modulate memory. Multiple regions, multiple methods, in terms of changing the physiology. And a lot of is going in an accelerated way, before the science maybe has a chance to fully catch up, in terms of what’s really going on. And so I think we have an opportunity here to go in parallel, and take advantage of these findings where we can modulate memory. And pursue them, perhaps in clinical settings. But also in parallel, perhaps in the same opportunities, try to understand what’s really going on. Is it sharp wave ripples? Is it theta gamma coupling? Is it consolidation, retrieval and coding? What is going on, that’s allowing this electrical signal to modulate the circuit and the network?

Nick Halper:
I think you bring up a good point there, Nanthia. Which is that… It’s actually relatively common that clinical products proceed mechanistic understanding of the exact target. Or basically, disease that they’re targeting. And so, it’s often the case that we end up developing something that is functionally useful as a clinical device, for example, without having the really nitty gritty evidence of… This is the individual neuro firings that are coordinating this, and this is how this device works. I mean, we’re only now just figuring out how deep brain stimulation for Parkinson’s works/ and it’s been around for 50 years. So I support that point of pursuing these activities in parallel with society.

Dan Rizzuto:
And I think this is also… What’s exciting about the new crop of neural devices that are being built, is that they have… It’s all about closed loop. It’s all about neural sensing, and this is going to open up whole new avenues for human research clinical studies. And it will also accelerate the development of technologies. If you can imagine how fast DBS for Parkinson’s would have evolved, if they had built in sensing from the very beginning. As opposed to waiting until 2021 to actually release, say commercial device, for sensing and DBS for Parkinson’s. I think we’re going to start to see a shift and an acceleration in the technology development.

Nanthia Suthana:
I guess that’s why I was so optimistic with my very unrealistic timeline. It’s because a shift is happening in the clinical treatment, opportunistic, science investigation world. Where we’re going to have maybe a hundred thousand patients walking around with sensing devices in their brain. And then several thousand epilepsy patients with hippocampal devices that can record sharp wave ripples. We’re close to that being a reality here. And so the scientists, I think, could do a lot with that kind of information. Just imagine that during sleep, during walking around, navigating, during recalling episodic details of an event that happened earlier in the day. That’s why it’s so exciting. And I’m very optimistic about the next 10 years. I’ll be… A very unrealistic timeline for what we were discussing earlier.

Gyorgy Buzsaki:
Certainly the new thing, is to take into consideration that the brain state matters. The brain state matters at a longer scale, such as sleep waking. For example, a drug taken at night, or taken in the morning, can have an opposite effect. If you remember all the young board studies, that you take scopolamine during the day, that’s bad. But if it takes a couple of men during the night, actually it increases the probability of a sharp wave ripples, and the memory improves. On the shorter scale, I think what we should remember, or think about, that what is being recalled is some kind of memory. And the memory, you can say, is information. Information requires a code or a cipher, which has an agreement between the sender and the receiver. And in every single communication system, the information is sending packages. It has to be packed into smaller chunks, just like in the Morse Code.

Gyorgy Buzsaki:
If this conversation that we are having today would be just one long word, that would be no way to understand anything. And the reason why, despite my accent, you understand me, is because your brain has the same written as my brain. And the segmentation is pretty much the same. So in communication system, you package the information in shorter chunks. And you can call it part of the neuro syntax. And in the brain, if you are looking for any rule that the guides the neuro syntax, that’s the system of the rhythms. And the reason why that is such a potentially useful method, is because every known brain written is based on inhibition.

Gyorgy Buzsaki:
So inhibition is a natural full stop in a sentence. It’s a natural coma. It’s a natural separator. And then you have showed the messages, and longer messages. And where we have a hierarchical system that is made for all of this. Now restoring all those complicated things that we are dreaming about is difficult, but assisting and making the gadgets good enough to improve memory is indeed…. I agree with everybody, that’s not so far away. Because if you can just make this packaging artificially better, then not inevitably, but hopefully, it will help. Especially in with brains, they don’t have this ability, because they lost it. Such as in epileptic patients.

Matt Angle:
What do we think is some of the most promising work going on right now, that would suggest new approaches for memory prosthetics? Or enhancing memory, or helping people with damaged memory systems? Are there particular threads for research that you’re looking at right now, and you feel very promising?

Gyorgy Buzsaki:
Animals or humans?

Matt Angle:
Either.

Gyorgy Buzsaki:
In animals, I do see a program. But that requires that you go inside the brain, and deep. In humans, that’s a little bit more complicated, if you would like to stick to the existing non-invasive methods. But I think in humans, what I hear back from you, is probably very similar. What we’ve been thinking about, is that the brain doesn’t let information in all the time. It is not like a Shainin System where the information is sent, then they receive it… Automatically just absorbs everything. It’s the opposite. It’s always the receiver who calls up the center and says, “Send it to me now.” So the brain is in charge. And monitoring that state in the brain, and delivering the messages the right time, is something that devices can do reasonably well. So I know that different phases of our selections, I will get to that. I know that certain levels of vigilance in certain areas are resistant to letting information. So why don’t we wait, or why don’t we just wait, that’s one way of doing it. And the second one is more complicated, of course, is then to modify it.

Gyorgy Buzsaki:
Is there any hope, for example, that we can make the brain a continuous device? That, irrespective of its vigilance fluctuation, it will be just like a computer that absorbs information independent of its own state? That’s not probably something that is too hopeful. But to exploit the knowledge that we know, the times and when the brain is more sensitive, that’s a different story.

Dan Rizzuto:
Perhaps not controlling the state of the brain… I would agree with you. It’s going to be very difficult to treat the brain like a computer, and control its state. But I think nudges are within our current technical abilities right now. Where you can nudge the brain. And that can have meaningful results, I hope.

Nanthia Suthana:
Yeah. Nudge, but also predict when is… Like you mentioned earlier, when’s a good time. Is it the phase of an ongoing oscillation? Is it during this particular phase of sleep? Or sharp wave, or full, et cetera. So that’s one approach. And then the second piece of the puzzle you need, is what’s the language? How are you going to nudge? Is it just going to be blasting it with this crazy amount of global stimulation? Or is it going to be focused? Is it going to be hitting a network of white matter? Which our results also suggest that that’s important. Is it going to have a pattern to it? Is it going to be a high frequency modulated, low frequency pattern, that could induce more plasticity? Et cetera, et cetera.

Nanthia Suthana:
I think as we learn more about this, we can be smarter in terms of how we’re communicating and nudging the brain, as you say. And also, I’m very excited about deep learning and machine learning methods that are being used on these big data sets, that hopefully we can gather in the memory field. Such that we can learn new information, in terms of these signals that are communicating back to us, and what they mean. And so in that way, to find out better how to nudge, or how to communicate back.

Dan Rizzuto:
Completely agree. And that’s the big trend that I see the whole field is riding on right now. Which is data science, and these very large data sets. And using machine learning, and artificial neural networks, to mine these massive data sets. And identify signals and biomarkers that can be useful for guiding our nudges.

Gyorgy Buzsaki:
So who decides when to nudge?

Dan Rizzuto:
The AI.

Gyorgy Buzsaki:
Who decides?

Dan Rizzuto:
But who trains the AI?

Gyorgy Buzsaki:
I’m serious.

Dan Rizzuto:
So are we. So in our approach, we know… I can tell you when a patient… In fact, I don’t need to tell you. The patient reports when they’re in a good memory state, when they’re in a bad memory state. So if you give a patient 12 items, if they remember six of those items, by definition, they were in a good memory state. Went for the items that were ultimately recalled. And they were in a poor memory state, for the items that were ultimately forgotten, or not recalled. So there’s a patient self-report, in identifying their own brain states. And that’s the behavioral signal that we use to train our algorithms, that then give the nudges.

Gyorgy Buzsaki:
It’s exactly as if I’m interrogating. So we discussed a few minutes ago, that it is a mystery. How my brain absorbs some things, and reject other things. And it’s a selection process. When I see in a cocktail party, through the photons impinging upon my retinol, that somebody is kissing somebody else, that’s not a big deal. But if it happens to be my wife, all of a sudden, my entire brain lights up, and then it goes on and on forever. Because this is me.

Gyorgy Buzsaki:
Now, when you are reading out my brain state, and you are delivering the stuff that you want to be delivered into my brain, is very different from the way how my brain selects by my outside world. And this is the key thing to think about is that, is there a way you just switch on my brain, or your brain? Say, “Okay, in the next half an hour, I will be on a hundred percent absorption mode, and I will learn everything,” which is not possible. But then, you can take the windows, but then you read those windows. Somebody should deliver messages. And who decides about those things?

Nick Halper:
I think this is where responsive neurostimulation comes in handy, right? You can look at indogenous signals in the brain and say, “I’m going to nudge these further into the direction they’re already going,” for example. Or do a manual correction to these. And so you’re relying… I mean, by the time your hippocampus… I’m treating this like it’s sequentially arriving here. By the time you’re at hippocampus, you’re in a really high order cognitive state. There’s high order information. And if you can simply take the brain’s natural processes, and up-regulate them, for example, I think that answers that problem itself, in some ways.

Dan Rizzuto:
I would echo that, or amplify it. We’re not saying that this hypothetical device is going to control your attentional process, or control our emotional process. What we’re saying is, it may amplify the learning process. And you still have the ability to completely focus your attention on whatever is most salient to you.

Dan Rizzuto:
We’re not going to be able to increase the learning of a completely random event that you have no interest in. And this comes back to this idea of control of a brain state, versus a nudge. But there’s another fact, which is the variability that you’re pointing out. The variability puzzle in human memory. Human memory function is very variable from moment to moment, and also from person to person. But impressively, within a given person, it’s incredibly variable from moment to moment. And it’s always been a mystery, as to what causes this variability.

Dan Rizzuto:
And my scientific co-founder at Penn did a massive study of 500 list-learning experiments, with individual healthy subjects. And found that it wasn’t exogenous effects, like the memorability of a given item, that was the largest source of variance. But it was list to list variability. That is, there’s something indogenous in the person, that varies on a pretty slow timescale from list to list, that drives the performance variability. And if we can nudge that process, I think we have a chance of significantly improving people’s lives with deficits.

Gyorgy Buzsaki:
Can you give an example then? How would you envision that? I’d like to be better prepared for this podcast, and I have your instrument in my head. On my head. What would you do? And what should I do?

Dan Rizzuto:
Yeah. So in our model, there’s an initial training process, which is actually incredibly intensive. It takes several weeks. But after that training process, once and again, this is where machine learning comes in. What that training process involves, is lots of list-learning. Patients are remembering lots of lists. Through those list-learning studies, we’re gathering brain-sensing data from distributed areas of the brain. Not only hippocampus, but neocortex. And using machine learning to identify those neural signatures that predict… Or really, it’s predicting the brain states that allow patients to remember information. And what we find, is those indogenous rhythms of the brain allow us to predict their performance. It actually allows us to predict with very high accuracy, if a patient is going to remember the information that’s being presented to them right now. It allows us to predict a minute later, two minutes later, whether they’re going to remember the information in front of them. And that is-

Matt Angle:
Is that generalized across tasks?

Dan Rizzuto:
It does. It generalizes across verbal memory tasks. We have not yet generalized it to spatial memory. But it generalizes from free recall of word lists, to free recall of categorized word list, to paired associate, learning to cued recall. So free recall is ABCDEFG. Recall them in any order. Cued recall would be AB, CD, EF. What went with A? What one went with C? What went with F? For those types of verbal learning, list-learning studies, it does. These classifiers generalize across… And we’ve published on this. Happy to provide references. So once you’ve done that, then you can… You know what the good brain states look like, what the bad brain states look like. Bad, good. And you can nudge the brain when it’s in these poor memory states. And like I said, if you nudge it in the right place at the right time, it makes the brain state look more like a good memory state. And you actually improve recall, without any effort on behalf of the patient. And again, that’s Ezzyat et al 2018.

Gyorgy Buzsaki:
Yes. But in that case, I think what has to happen, is I have to interact with you. Because you know, and I accept what I have to learn. And you are looking at my, or the patient’s, brain state. And said, “Oh, let’s nudge it. Because I’m going to present blue.” But that’s not exactly what I am interested in. I’d like budge my brain to those things that I want to attend to.

Dan Rizzuto:
That’s fair. In our publication, we were synchronized to only the encoding periods. We’ve since replicated this to… Again, I do believe our results only translate to verbal memory. But again, this is the type of memory that is most impaired in patients with memory loss, such as Alzheimer’s and traumatic brain injury. It’s verbal memory loss that is the issue. That causes them to have to leave work, leave school after their injury. And so, we believe we can improve verbal memory through this training and nudging process. And without any need for… There’s a training process, which is intensive and synchronized. But then when you release the patient at home, there’s no further interactions that’s required. It’s continuously operating in the background, giving nudges.

Nanthia Suthana:
So I guess the challenge, though, will be whether this will translate to real-world experiences that are complex, and a mixture of verbal and nonverbal information. And second, whether there could be a place for… as Gyorgy mentioned, individual’s preference. Maybe having the DBS system on or off in Parkinson’s, or something. But I don’t know. That could be a future discussion. But it would be ideal if there is no interaction, or need for the patient to do anything. Because then it’s more likely to work continuously.

Nanthia Suthana:
But I hear what you’re saying, Gyorgy, in terms of… Patients may wanting to… Have an important event, and really remember this graduation ceremony, or birthday party, et cetera. But it sounds like, Dan, from your saying, that would work in this kind of responsive way, for this event. If the verbal memory results can translate to a complex real-world scenario. But I mean, this is touching upon a bunch of ethical issues, that will need to be discussed if this really comes to fruition. In terms of who is in control, and what can be changed, in terms of the individual patient versus the clinician. And how do you make these decisions?

Matt Angle:
I have a question from Jacob Robinson at Rice.

Jacob Robinson:
My naive question would be… When I think about the science related to learning and memory, it often relies on technologies that are far more advanced than what we have available for human patients. And I’m curious. How do you resolve that gap? We have very limited ability to interact with the human brain in a clinical setting. We have the increasingly expanding ability to do so in animals. Is that a bridge too far? I would be concerned that the hype is going to be like, “Look at what we did in a mouse.” And then people are going to expect that in a person. And we’re like, “No, we are decades…” The technology lags by 30 years. What we can do with clinically, versus what we can do in research labs.

Nanthia Suthana:
Ooh, interesting question. I would disagree with that, as of 2021. I think that these last couple of years… I mean, we just published a paper earlier this year. That I, in grad school, would never believe if you told me, 20 years later, that I could do this in a human. And so, I think that we are closer to bridging that gap. There’s a lot to be done. We just published a paper recording intracranial oscillatory signals as a person is walking around. Then yes, okay. People were doing this 30, 40 years ago, in animal studies. But this is just the beginning.

Nanthia Suthana:
Because of all the animal studies, we have so much knowledge. That I think when we get into a human, in direct brain access, it’s going to be much more accelerated. We can hopefully catch up quick, because we have so much knowledge from animal studies. So, I would direct to our recent nature paper in 2021. There’s also a paper in Nature Biotech by the UCSF group, recording signals in naturally behaving Parkinson’s patients. And like I said, a hundred thousand people could have these devices in the next five years. So that gap is going to be bridging. Hopefully very soon.

Nick Halper:
I think that gap is being bridged for multiple reasons. It’s not just between the science and tech gap. I think overall, there’s that lag time between animal studies and what can be done in humans is just decreasing. Especially in neuroscience world, partially just due to the increasing acceptance of… I’ll call it bioelectronic medicine and implantables. Stuff like this moves through the pipeline faster. There’s more standardization of technology producers. Whether that be in really mundane things, like hermetic sealing, that being a service you can just pick up now, versus a really big in-house challenge, like it was 20 years ago. To manufacturing processes and miniaturization of computation that gives… The clinicians are basically people developing these clinical products, a lot more freedom in being able to kind of meet those demands. Or

Nick Halper:
A lot more freedom in being able to kind of meet those demands or follow on what was happening in animal research. And so I think there are other factors at play that closes that gap besides just the scientific gap. But I think the second kind of point I would add to that is that a lot of people, when they look at the scientific gap and memory, I think they look at a lot of the single unit work that’s been done, but so much of memory is about rhythms in the brain, right. And that technology has been around a lot longer in humans already. And so I think we’re at a place where we can start just leveraging those more intelligently instead of having to maybe translate or overcome some of these hurdles related to a single unit tech or science.

Matt Angle:
And can you tell us a little bit about what you’re building now and how you’re working to bridge this gap?

Nick Halper:
Yeah, exactly. So I think earlier you asked a question about what area of research in memory excites me, it’s network effects. It’s kind of like almost surprising. If you look at the titles of memory papers that come out, they all focused on these like single target effects. And this is partially because of how science caters towards narrowing down that single variable, right? Stem the single target. See what happens to him, the other single targets, see what happens. But I think those are just really promising area of memory research that is about these kind of network interactions multi-region interactions. And so that’s the problem where basically the approach that Braingrade taking is network effect interaction. So looking at the hippocampus and hippocampal assembly, as you know, we’ll call it a data wave generator or earlier use the term indogenous amplifier and trying to connect that device to other regions in the memory circuit and synchronizing those that is our kind of primary approach. And so that is the area of research that I find most exciting.

Gyorgy Buzsaki:
I like that because it really exploits the mini path. What I call the good enough, it feels a bit perfect, but then call it a nudge. But this would be a different kind of nudge. It would be a little elevator that rigs things a little bit to another level. So you complained about the single unit studies and they said, oh, when I have 15 euros, then I can tell where the animal is very, very high precision in the hippocampus. But if you have 50 electrodes in the hippocampus, which is recording nothing else, just a filtered LFP between five and 15 Hertz. And all you have is the amplitude and phase readout of those locations. You can tell better than with 15 euros where the animal is.

Nick Halper:
Exactly.

Gyorgy Buzsaki:
So for the practical reason the brain doesn’t use brain LFP or brain waves oscillations as such, because that’s a derive signal, but for the experimenter, this is a much easier signal. For the first, if for nothing else, it’s much easier to record it forever then unit activity. And you can derive the same information. Now, there are a couple of studies in humans with human MEG or even EEG. You can tell which direction the animal, not the animal, sorry the human is looking at. So they had their action system, which is deep down somewhere. But with from the phase distributions, you can read out and use it practically. Even, if you don’t understand the thing about it.

Matt Angle:
Nick you mentioned studies looking at inter areal communication and in memory. And are there particular studies that you’re looking at as inspirational in this regard?

Nick Halper:
Yeah. So Nitin Tandon and Kamin Kim did network effect studies using intracranial electrodes in humans looking at phase synchronization. So when you’re studying epilepsy patients, you don’t always get to choose the targets that you’re in, which is one of the challenges of this type of research, but looking at data phase synchronization between hippocampus and signals that they’re seeing in fornix. For example, I think I find that paper particularly interesting. It was a small study, but I think it marries a lot of the theories that have existed in memory research for a long time, and also explain some of the shortcomings or inconsistencies that we see in some single target approaches.

Nick Halper:
And there’s a lot for every paper that stimulates in hippocampus. For example, there’s another paper that says the stimulating, the hippocampus actually disrupts memory. And you have these kind of a tug of wars between scientists who seem to be doing basically the same experiment, but aren’t really able to resolve it. And I think this is where closed loop neurostimulation comes into. And I think that alone would solve some of these issues, but especially on a network of time scale. Right. So I can see what two regions down the chain is doing and stimulate into hippocampus system million, 10 frontal cortex at the appropriate time.

Matt Angle:
Nanthia you mentioned, some work that you did recently, would you like to talk a little bit more about that?

Nanthia Suthana:
Yeah. I mean, it’s all, I think very similar in terms of the big overarching questions that we’re trying to answer using electrical stimulation, we are have been focusing on a different region. So another region in this big network, which is the entorhinal perforont pathway, which is what we think we’re targeting, it’s the major input pathway to the hippocampus. So where all the sensory multimodal information is con you know, coming in and focusing into get sent to the hippocampus. We have electrodes that are in this white matter area where we, where it’s known that human perforont pathway fibers extend to the hippocampus. And we’ve shown in a couple of studies now that that stimulation when provided during learning can result in improved memory. And it’s been multiple papers in different post-doctorate postdocs hands. So I’m, I’m rather convinced that this is a finding that is consistent across these different studies.

Nanthia Suthana:
And we’ve done different types of lab-based tasks, like verbal memory that was mentioned, and also learning people’s names face based name, associated tasks, as well as some spatial navigation and some other object recognition task. And it seems to generalize in terms of improving memory there. And now my ultimate question though, is how much does that translate to real world experiences? So my lab has really moved beyond those, those lab based tests to try to do this in real behaving humans, although it’s much more challenging and in parallel trying to record the signals and use deep learning to see if we can glean something from those signals that can tell us when was an optimal time to respond.

Nanthia Suthana:
And we’re also working in areas of memory impairments that are unwanted. So in the case of post-traumatic stress disorders, we have a clinical trial to use responsive neurostimulation for PTSD, we’re targeting amygdala regions and using a similar approach where we’re trying to, to find the signal that can predict when the individual is being triggered. You know, let’s say they hear some fireworks during 4th of July, which is very triggering for a veteran population. Can the device stimulate the network to suspend the behavioral outcome of that trigger. So we’re very excited. We have our first two patients this summer, and those are, those are some of the things that we’re working on now.

Matt Angle:
What do we think a natural sequence of BCI products might look like over the next 10 years with regard to memory? Do we see that there are some kind of low-hanging fruit, some more ambitious projects. If we wanted to look a little bit more near term, what do you think people have to look forward to? What’s the first species product that hits the market?

Nick Halper:
So I think utilizing existing technologies, we’re talking, LFPs sensing a memory prosthetic with the ability to sense local fuel potentials and stimulate using macro stimulation. I mean, these are proven technologies that don’t need, with longevity proven out decades, you don’t need to develop anything new to get a device that can work for decades using sensing LFPs and stimulating macro electrodes it’s there. And we know how to, and now we have proof of concept about how to use these two tools to nudge the brain and improve memory.

Nick Halper:
I think that’s the first step looking further down the line we’re talking about expanding that to additional brain regions. I mean, right now the most complex brain stimulator has two leads in the brain. You know, these Medtronic DBS devices or the neuro paced devices, they have two leads that we ultimately want to get out to larger numbers of leads, targeting larger numbers of brain regions, and potentially getting more precise in terms of the local fields. I don’t think we need to go down to single units. Maybe I could be wrong. I doubt it. I think that level of resolution at while incredibly useful for neuroscientists is not necessary for a memory prosthetic, but you know, what is the optimal size of the LFP? What is the optimal size of a stimulating electrode in order to facilitate memory performance? We don’t yet know.

Gyorgy Buzsaki:
Well, I don’t have a company, so I can say whatever I want, but I would ask the question differently, whether you are talking about invasive or non-invasive, you’re talking about the white public, or you are talking about patients. And I think we have to look at first the patients. So because these are the ones who really need to come back to where most of us are, rather than a creative super guys or NGOs who can outperform the average population. So in that case, we know that in epileptic patients and in also Alzheimer patients, the biggest, well, one of the biggest problem is the interictal spikes, the seizures come every month or every two weeks, rarely more often, but interictal spikes can be many, many, many, and they hijack the pathways that are needed for making memories. I don’t know what we can do with closed loop stimulation or closed loop with spikes because they are so short, but the consequences are devastating because every single hippocampus interictal spike produces a long lasting downstate, at least in the prefrontal cortex.

Gyorgy Buzsaki:
But that is a long time in the millisecond range where we can think how those unwanted downstairs can be brought back. So any device that can kick and change that hopefully will do nothing else, just bring back the physiological level. And then it leaves it to the rest of the brain to do something. Amazingly, we have done several experiments in the past and recently you can switch up the hippocampus completely for a couple of data cycles, and it comes back in one single data cycle to the exact state where it was before. So indeed, any short interference that can nudge, I forgot what the other was word was. You can manipulate the state of the target area and let it go back to how it would do under normal physiological conditions would be a low-hanging fruit and would lead to some memory improvement.

Matt Angle:
Nanthia and Nick, do you have any thoughts about what kind of products we might see in the next 10 years?

Nick Halper:
Is it cheating to just talk about my own product?

Matt Angle:
No that seems reasonable.

Nick Halper:
Yeah, I mean, I agree with Dan that the first products and I think it’s a little bit odd to call them the first products onto the market. So I think what these are is next gen devices of existing deep brain stimulators. And what we’re looking at is improvements to those, whether that be in adding sensing capabilities or adding multi-region capabilities or increasing the number of LFP sensing electrodes that you have in hippocampus, all of these things are improvements on existing designs.

Nick Halper:
And so I think that’s what the product’s going to look like in the next period. And we’re going to see other improvements alongside those, right? There’s going to be additional miniaturization. We’re already seeing incredibly longer battery life. 25 years was the most recent mark that I saw for a DBS device, which is very impressive. But I think ultimately when we talked about BCI for memory, what we’re going to see is devices that are capable of interacting with more of the circuit. Cause I think that’s been the limitation on existing DBS devices. If somebody could pull off a Boston scientific device or a Medtronic device and just stick it into hippocampus or internal cortex and get the exact improvements they wanted, then Dan and I wouldn’t be building what we’re building. And we already see that in companies like functional neuromodulation, which haven’t done as well for these targets. So I guess, yeah, what I foresee is the next generation of things that are here.

Nanthia Suthana:
Yeah. I’d say similar response at UCLA. You know, we have a collaborator here, engineer Dejan Markovic, who was funded by the DARPA program to develop a higher channel, better sensing, neurostimulation, close loop device. And we have a prototype now that we actually are testing the version bedside with epilepsy patients and recorded LFP single neuron activity with the tiny little miniaturized device that these other systems we use like BlackRock can do. And so I think as these technologies get miniaturized and integrated, we may see better devices. But I agree with Dan that in the short term likely be existing devices that are already FDA approved that are good enough to go forward and provide some improvement and quality of life. And then as that hopefully progresses and shows some positive effects. We can start putting these new devices to the regulatory system so that we can upgrade at a certain point and do better and optimize it further.

Matt Angle:
My last question, when this podcast goes out to the millions and millions of listeners that listen to neurotech pub, what would you want to ask the world for? What do you think couldn’t, whether it’s a technological improvement or some change in the way that regulators were, or improvement data science, or someone to test a specific biological hypothesis, what do you think would move this field forward? In other words, there’s limited science limited. What is it?

Dan Rizzuto:
Connectors. The ability to connect. Large numbers of electrodes to a can or an implantable pulse generator that’s in the brain or in the chest that’s restricted by the running fine wires between those electrodes and the can, the pulse generator, the implant and connectors are holding us back. Well, there’s only one connector company right now, which is called Balseal. And you can’t stack more than eight Balseals in a linear array, which means you can’t have more than eight electrodes on a given lead. And I think ultimately, Matt, I think you’re already going to multiplexing, which is probably the ultimate solution, but for companies that don’t have a… there isn’t a commercial multiplexing solution out. And so that could be fair enough. Multiplexing could be part of the solution, but it’s how you connect to large numbers of electrodes to the electronics for processing the data.

Nick Halper:
And those connectors could be wireless, right? Like a drastic improvement in wireless technology.

Dan Rizzuto:
Could be. That’s right. Wireless connectors. Yeah.

Matt Angle:
We had Vanessa Tolosa on the podcast a couple of episodes ago, and I think she sent the word connectors about 20 times. I think it’s on a lot of people’s wish wishlist.

Nick Halper:
Oh, definitely. I mean, connectors would be a definite pick of mine. I mean, if we’re talking about for this field in general, too, there’re certain problems that are hard to solve without funding. And by this, I mean, funding for I’ll call them risk-taking products. I think so many companies die at the starting line just because they can’t make that jump. We hear about the valley of death in fundraising a lot. And I guess the kind of specific situation I’m referring to is kind of what we’re looking at now, right? There’s only so much research you can do in a human kind of I’ll call it preclinical human works in humans, but before you actually have a product that’s meant to be under clinical trial investigation, for example, there’s only so much work you can do there before you eventually have to build something for your target population, that’s catered to them. Right. And to bridge that gap, to build that thing that that funding is missing. And so I think, yeah, I don’t know. I mean I’m another entrepreneur talking about more funding needed, but you’re right.

Nanthia Suthana:
I will also extend that funding requests to funds to develop training programs, to train the next generation of individuals we need in this field, which is actually a big challenge. This neuro engineering or a technology field where you’re, we have somebody who knows the language of these very different fields of engineering, neuroscience, medicine, clinical industry, right? It’s so many different things, so many different hats, and it takes a very unique person to be able to really merge all of those.

Nanthia Suthana:
I’m not sure as a neuroscience community, we’re doing as well as we could in training these future generation of neuro technologist that can span basic science to industry. And so at UCLA, we ha I have a T 32 for this translational neuro technology was just trying to do this, but it’s like one little piece of what really is needed, which is really just a better way to train these students who are coming from all these different fields and really string them specifically in this application of developing and applying and translating neuro technology for neuroscience and bedside to clinic, bench to bedside is what we say right from basic science to clinic. That’s, that’s what we need.

Gyorgy Buzsaki:
The history of humankind is the history of externalization of brain function. And that’s typically memory, whenever you make an artifact that lasts forever, it outlives you. So, that is extremely helpful. And of course the technology since with America and then eventually worse, and even verbal communication changed a lot. So we’ve got tons and tons and tons and tons of improvement when it comes to access information. And the reason why it is so exciting and interesting is that there is no other species like us, where they share of human cries knowledge or the spacious knowledge is so little in the individuals. My share of humankind’s knowledge is ridiculously small. Now, how? So, what I would ask from the large audience out there, the millions and millions of people who are listening to this program is that, what is the need? Is it good enough?

Gyorgy Buzsaki:
Just make it faster so I can have access to everybody else’s knowledge faster or more effectively, or should I concentrate, or should we assign this constant rate on gadgets that allow outperform everybody else? And I will get an advantage over others who cannot pay, and I can make my memory better by implanting things. And what other costs you accept for that? Because we know that there are people with big memory. There are people out there since ancient times who can memorize all speeches of Cicero. There are people who can memorize the cannot forget anything, but none of those survive very well in society. So we have to be prepared. And this is what you might be asking from them. That what happens if you learn a lot of things, but including all those bad emotional memories, that there is no choice for you to select or another level of communication, could they, the public, is that who should we help? First? What is the low hanging fruit?

Gyorgy Buzsaki:
We are scientists. So good. Neuroscientists are so good making the brain worse and breaking things. And can we utilize that? In fact, many of the simulation studies, both from UCLA, as far as from University of Pennsylvania, showed that it’s so much easier to make things disappear or not to remember than to recall something, maybe PTSD would be a good target. We said, okay, we have many ways to show how to forget it, how to access those memories online and do a cold closed loop loop system. That allows me to remember something, or when I spontaneously remember something, I just push a button and said, erase. This would be my vote as a first 10 year program rather than enhancing my memory, which I know it is detailed writing, but you know, I have enough help from here. And I tried to go.

Matt Angle:
Thank you all for taking the time with me. I think the audience is really going to like this discussion. I certainly did.

Dan Rizzuto:
Thanks Matt.

Nick Halper:
Thanks for inviting us.

Gyorgy Buzsaki:
I enjoyed very much talking to you guys, and I wish you good luck with your companies.

Nanthia Suthana:
Awesome. Thanks everyone.

Read more

Listen To The Conversation


// Director & Host: Matt Angle
// Producer: Lili Byrne
// Editor: Jasper Sams
// Graphic Design: Midnight Measure Design