[00:00:36] Camille Morhardt: Hi. I’m very fortunate to have with me today Allison Duettmann for What That Means. We’re going to talk about neurotechnology, nanotechnology, and molecular manufacturing. She is CEO of Foresight Institute, and she is a pretty smart cookie. She’s got a master’s in philosophy and public policy from the London School of Economics and graduated summa cum laude there. Welcome to the podcast.
[00:01:04] Allison Duettmann: Thanks so much for having me. I’m very excited to be here.
[00:01:07] Camille Morhardt: Can you describe what Foresight Institute does? I know it was founded quite a while ago in ’86 or ’87, but what is the goal?
[00:01:15] Allison Duettmann: We’re one of the longest standing science and tech nonprofits that is trying to advance technologies for the long-term benefit of life. We were founded on a long-term vision of molecular nanotechnology. But from the very early ages, it always had a cacophony of other technologies in the mix, including AI, implications for longevity in biotech, brain computer interfaces, and in general, newer technology. It has vast implications for space technologies if we would ever actually get to advanced molecular nanotechnology and so forth.
Because it had this sweeping focus on a variety of different technologies, over the years we’ve taken the molecular nanotechnology angle and then also spun out individual projects that focus on the other technologies as well. The way that we usually support projects in that space is through a series of virtual seminars in each of these categories, and then a series of in-person workshops where we bring top people in the field together to work on the long-term goals of these technologies for better futures that they usually don’t have another venue to collaborate on because that’s not really what normal institutions, I guess, are set out to do very much.
And then we also have prizes and fellowships in these categories. I think we’re trying to provide this extra space where people can collaborate on these more long-term visions of technologies and how they can actually help humanity flourish.
[00:02:34] Camille Morhardt: I want to just take a moment too. You’ve said benefit of life or long-term goals or helping humanity flourish. They sound really good, but I do want to ask, what do you mean by that? What is positive for you? What is beneficial for you?
[00:02:50] Allison Duettmann: Well, that is a very big question. I’ve studied philosophy before, and I can safely say that within thousands of years of philosophy, people have not figured out what the good is, or at least they have not agreed on what that is. What we are trying to hold space for is technology development that leads to worlds that can be assessed as better by many individuals. We’re not trying to really drive down really one narrative, but we’re really trying to give space where lots of different positive futures can thrive. Where really individuals have achieved the greatest possible lives for themselves, and that also includes planetary health and well-being, because without that, we don’t really have a space to be in.
Without wanting to really narrow it down on one let’s say ethical framework that we follow, we’re just trying to get to futures through technologies that can be regarded as better by a variety of different entities, maybe not only humans. In the future. We may have to expand our moral horizon a little bit and also gradually incorporate non-human intelligences and sentient beings into the mix. It has a relatively wide focus here.
[00:03:55] Camille Morhardt: Are you referencing machines, or are you referencing other biological organisms that are non-human like plants or other forms of animals?
[00:04:03 ] Allison Duettmann: Well, I’m referencing both. We are usually not great as humans to already incorporate many of the other sentient creatures that already exist around us into our moral circle. And then I think that also at least allows the worry that we may not be really great at distinguishing when it’s time to perhaps include artificial consciousnesses into our circle of formality, too.
[00:04:27] Camille Morhardt: Well, I guess really quickly, what is a conscious creature? What would you define as making it conscious?
[00:04:33] Allison Duettmann: I’m not a consciousness researcher and there have been so many different people that are just expert in that field that can define this much better, but I think we all, even as a layperson, have an understanding of what at least for us it means to be conscious. I think it’s pretty precious. I think preserving this very pedestrian notion of consciousness is I think something that I think is inspiring enough. Whether or not we can figure out all the scientific details behind it, I think we get to experience it, or at least we think we do. That’s my best bet for why it makes sense to preserve it.
[00:05:07] Camille Morhardt: I think that’s a pretty reasonable answer. Let’s dive in just a little bit to some of the technology that your organization focuses on. Let’s start with molecular manufacturing and nanotechnology, which are a little bit interchangeable in terms. But maybe you can tell us what those terms mean and then we can have a conversation about them.
[00:05:30] Allison Duettmann: People have been arguing and debating the correct use of the term nanotechnology because I think nanotechnology as it is with other terms, for example, as AI, the moment that they start pulling interest into the field, many other things get rebranded as that technology as well along the way. That’s also what we’ve seen a little bit along the path on molecular nanotechnology.
But I think maybe the best way to describe it is just imagine perhaps for a second that it was really truly possible to build things from scratch, so that rather than taking a default resource like, I don’t know, stone or wood, and then cutting it into shape and thereby wasting much of material in the process, imagine if you could actually build from the bottom up by assembling precisely only the types of materials necessary to build the object. The question is really, if resources, costs or externalities weren’t an issue, what could you build? Richard Feynman is often I guess credited with really having kicked off the field of nanotechnology.
He gave the speech in 1959 I think at Caltech where he basically said the principles of physics, as far as I can see, do not speak against the possibility of maneuvering things atom by atom. This is I think the core vision of molecular nanotechnology is really doing atomically precise manufacturing.
[00:06:43] Camille Morhardt: What kinds of things would be built? Do they span biological and non-biological? And then how do they get built? What builds them at that level?
[00:06:54] Allison Duettmann: Well, Eric Drexler, for example, who co-founded Foresight in 1986, he wrote this book, Engines of Creation, in which he envisioned a future that was defined by these engines of creation, these molecular machines that could really assemble from the bottom up very complex objects. They were also helped by engines of design, and that was AI. Basically actually being able to look and see what we want to build and where we want to place individual let’s say atoms even if you get really ambitious. But he basically detailed this path towards achieving atomic precision through molecular engineering.
The early stages really involve redesigning biology’s molecular machinery, so for example, protein molecules to really position reactive groups with atomic precision, and thereby turning them into these machines that are capable of constructing more complex materials. You build that from the bottom up, almost like 3D printing, but on a very, very tiny level. For that thought very long term, you could really build a variety of different either structures or eventually even other factories that could produce new factories. I think the promise there is almost unbound, or as people like to call it in the early day, you could really reach this abundant future.
For example, if you just imagine these devices and materials that could be designed could, for example, really improve lighting efficiency. It could promote local power generation and could really help us meet the energy demands that we have as a planet, which are a major big deal for the long term future and very sustainably. We could have, for example, new water purification methods with these very, very precise materials and could really alleviate much of the water scarcity that we have today. Going to the medical domain, develop new methods, new diagnostics, new personalized medicine and new personalized therapy tools that could combat diseases, fight aging.
I’m sure that some people have heard of the term nanobot, of this very tiny robot that could be to clean up the inside of your bodies and remove the things that aren’t working very well and repair the bits that need repairing. You have a lot of promising focus areas in there if you go all the way sci-fi. But I think even if we talk about space technologies, it’s very difficult to make progress on long-term space development exploration because it’s really expensive to send stuff into space.
But with new types of materials that could be designed to be, for example, incredibly lightweight, you could also make progress on long-term space travel in a way that is currently really beyond our imagination, where currently obviously we are not there yet, but very long term.
[00:09:26] Camille Morhardt: Right. You’re envisioning something where in conjunction with AI, for example, maybe it exists in an area where there’s water scarcity, to use the example you provided, and it looks at the surrounding materials it has to work with or the atoms that are near it and it decides how it could then maybe construct something to create purification in that area given the elements that are around it versus it may create something else to purify water in a different location?
[00:09:56] Allison Duettmann: Well, it’s unclear really how these long-term technologies will actually pan out because we know so little about it, and there’s different ambitiousness levels, I guess, of that technology. If you could just create machinery that was able to clean up the mess that we made, any kind of molecular machinery, you should be able to also clean up the mess that other molecular machineries have made.
If you just look at the climate and biodiversity and resource crisis that we have, all of this could really be helped a lot by just being able to produce better materials from the get-go that don’t even cause that much of waste and externalities, but then also by cleaning up some of the mess that we coming to the current level of civilization have left behind.
[00:10:36] Camille Morhardt: Some of the medical advancements, just to define it in more detail, would be something like chromosome replacement therapy. I mean, you were alluding to maybe repairing certain organs maybe if there’s tissue damage or removing certain toxins. I mean, I think in the longevity space, aren’t we looking at something like using proteins to recreate what the DNA blueprint is telling it to create as opposed to the damaged molecules that occur over time? We’re actually doing replacement therapy at the chromosomal level to keep our naturally deteriorating bodies repaired and free of deterioration.
[00:11:15] Allison Duettmann: I mean, if you want to go all the way out there, and we’re not there yet, but one, for example, problem in longevity is that even if, this is my assumption, even if we make all the progress that we can make, and this field has accelerated drastically, then I think it may still not be enough to get people currently alive over this threshold of LEV, which is longevity escape velocity, where people currently alive can live this long life. One plan B is cryonics.
Cryonics is basically preserving your entire body or just your brain after you’ve died with a hope that later on we will have the types of technologies that allow us to re-thaw it without as much damage that it would prevent whatever you want to call consciousness to re-inhabit these bodies. The biggest problem in cryonics is that currently most of the techniques that are being used to preserve human bodies or just human brains are incredibly toxic. Even with vitrification methods that we currently have, we are still causing massive damage to most of the cellular structures that we have in the human body.
Molecular nanotechnology is usually very much almost a necessary condition if we want to make sure that we can, A, preserve folks better, but then also, B, be able to potentially go in there eventually later and repair some of the damage that current existing technologies are leaving. That is a very, very far out case, right? But it is something that would be uniquely enabled by a nanotechnology. Because with the current technologies as we have them so far, I see little hope that that we’ll get there.
[00:12:54] Camille Morhardt: There’s some criticism I guess in the space, but help me understand this better, that AI in conjunction with nanotechnology or molecular manufacturing can then decide what to construct and autonomously go and construct it. It can happen at such a small scale, obviously that we may not know it’s there as humans.
[00:13:15] Allison Duettmann: Well, you’re referencing the Grey Goo scenario, I’m assuming. Is that correct?
[00:13:19] Camille Morhardt: Yeah.
[00:13:20] Allison Duettmann: I mean, I think many people here probably are familiar with the paperclip maximizer scenario that has recently had quite the revival and concerns about fears of AI, but it’s basically that you’d have these self-replicating nano machines that could eat up all the resources that we have as human civilization, including humans and including our biosphere. It could happen so fast because they’re self-replicating that we may not have time to intervene or even notice before it’s delayed. This is partly a pretty out there worry based on a misunderstanding of the types of molecular nanotechnology systems that are actually being promoted by the community.
This is this more totally free ranging creation of these new entities basically. But instead, what most people in the field are arguing for and what is in fact much more pragmatic is these nanoscale factories where we could really control the input and where we could exactly know what we want to manufacture. Those are much easier to build arguably. It’s free roaming, entirely autonomous bots. This worry of a Grey Goo has really been grasping on this misunderstanding of what the field of molecular nanotechnology is trying to do.
There is even a case to be made that by the time that this fear became very prevalent, it was potentially being used to redirect funding from the more ambitious goals of molecular nanotechnology to the more near term stuff. That fear is we can never absolutely I think negate it, but there were some political motivations. Then obviously, anything that sounds like a new threat is also incredibly exciting for the media. That definitely played a role I think as well. I don’t see it becoming a near term concern anytime soon.
[00:14:57] Camille Morhardt: Especially when it gets a name like Grey Goo.
[00:15:00] Allison Duettmann: Yeah, of course. That definitely helps.
[00:15:03] Camille Morhardt: Can we talk about neurotechnology a little bit?
[00:15:08] Allison Duettmann: Well, again, it’s a very, very broad field and spans all the way from neuroscience to specific devices that are currently being created. But what we are focusing on, at least at Foresight, or the types of newer technology that we’re interested in is looking beyond what’s currently available to the types of newer technology that could actually help us cure many of the cognitive diseases that we’re really worried about, but then also allow us to really expand our capabilities and potentially even have something like brain computer interfaces that would allow us to much better interact with the real world that is already very computer mediated, but then also potentially eventually do things that are much more out there, such as whole brain emulations, which obviously is a term that many people if they do know it, probably they know it from sci-fi. But I actually do think that there’s some paths to get there that are pretty interesting right now.
[00:15:56] Camille Morhardt: Can you describe what whole brain emulation is?
[00:15:59] Allison Duettmann: Whole brain emulation is basically a potential future of state of a technology where you can really emulate an entire brain in a computer. I think in the wake of concerns around these artificial intelligences that don’t share our hardware, don’t share our evolutionary environment in which they were able to gradually learn the values that we hold, it will be very difficult to align these AI with human values. One strand that people have become a little bit more excited about is really looking at whole brain emulation as a path or as a safety hatch for artificial general intelligence.
Because the idea is that if it was possible to emulate a human brain in a computer, and that is a very, very, very big goal, then at least we have some reason to believe that it’s more easy to align those human brain emulations with our current human values and it would be for an entirely external AI. We would also have better ways to interface with AIs that we create and actually get smarter ourselves, and potentially we’d be able to run on the incredible speeds that many of the computing programs are currently running and so forth. There has definitely been a recent uptake in interest in the field.
Nevertheless, I should also say that human brain emulations, I would say we won’t get there anytime very soon. I think without significant funding and significant interdisciplinary collaboration, it is still very much like a high hanging fruit, let’s say, within the newer technology space.
[00:17:22] Camille Morhardt: But it would be running on circuits, not biological material. Is that correct?
[00:17:28] Allison Duettmann: Here you’re getting into the long-term possibilities of having biological computing. There are a few companies and labs that are working on biological computing. Cortical Labs, for example, is an interesting one. One could imagine that potentially the hardware could even be biological in the future. Yeah, it’s possible.
[00:17:49] Camille Morhardt: Very interesting. We’re talking about with things like Neuralink, right, is another common reference in neurotech space and looking at stimulating the brain, stimulating brain function, reading the brain, and then allowing for communication externally that wasn’t possible otherwise. What kinds of things are being looked at in this space?
[00:18:15] Allison Duettmann: There is a number of growing companies or research labs in the space that are trying to do non-invasive BCIs because they wouldn’t have to be replaced. Caltech is working on one and there’s a few others as well. At the same time though, I think it would be difficult to get the entire readout that you can get from the invasive ones because they just have a better reach locally. There are, I think, a few really interesting projects current out there, and I’m assuming that in the future may even be not necessarily a patchwork, but I’m just hoping that we can just develop also better technologies that can be more non-invasive while still being invasive.
I think that there is a lot of progress in this space, but I think we still need to work on durability. We need to work on reach. We need to really work on readouts. I think it would be great if there was more collaboration between individual companies and what they can do currently in terms of readouts and really trying to have a bit more of an open access to pushing some of the progress along.
[00:19:10] Camille Morhardt: What about this concept of reading versus writing when you’re talking about brain computer interface? What is writing?
[00:19:18] Allison Duettmann: Well, it would really involve actually being able to communicate your thoughts. It’s really early stages in that it’s extremely difficult to be doing something that can be reliably upheld in that way.
[00:19:32] Camille Morhardt: Where I was trying to head with that is reading whether somebody intends, I want strawberry juice right now or something, as opposed to grape juice or whatever, and you’re able to actually read that when a person can’t speak or communicate any other way. That’s sort of the inside out. You’re trying to understand from the outside what’s happening on the inside. What about the reverse of that? What about the outside in where you’re implanting information or potentially implanting thoughts? It is written about. It’s not like that so far out there concept right now. Help me understand where are we with that, and what kinds of things are we able to implant?
[00:20:08] Allison Duettmann: Well, I mean, there was a recent paper that has made quite the wave from I think Yu Takagi and Shinji Nishimoto, and they basically used generative AI models to recreate images from human brain activity that relatively well matched with what the human subjects were actually looking at in that scenario. There’s a really fantastic paper out there that’s rather terrifying if you look at the individual depictions. And then they were able to use generative AI to actually really give a pretty good understanding of what the person was actually looking at. It was this reverse engineering. A fear for that at least is that this technology in the hands of the wrong people could really lead to massive infringement on our privacy.
I think the place where we really most have privacy currently still is our mind. On the long run, people worry about something like mind crimes. There’s been some research at FHI that was looking at these digital minds and the types of crimes that you can intrude on minds if you actually had an ability to change thoughts, to change the environments in which they occur and to really engineer experiences for people is I think pretty massive. Because at the end, the mind is the filter to which we perceive everything else in the world. If one tinkers with that filter, I think that is probably one of the worst crimes that one could possibly commit.
[00:21:27] Camille Morhardt: Very, very interesting. One of the things is this cross-section, I guess, of neurotechnology and nanotechnology. We used to say data is the new oil, but at this point, we’ve got the data, we have a lot of data, and AI is working hard on the ability to be able to process that data and then generate predictions and insights from that data. One of the things we have a lot of data on is biological data that people, myself included, provide willingly for the benefit of a dashboard back to myself on how well I’m doing.
I know that there’s data anonymization and other techniques to separate my personal identity from the biological data that I’m submitting, but a lot of that data ends up in the cloud. There’s really only I can count on my two hands the number of hyperscaler clouds that probably contain vast quantities of information on human biometric data. I don’t know that I have a great question to formulate. I’m just wondering what your thoughts are on that, when at the same time we’re approaching things like molecular manufacturing and we’re approaching things like neurotechnology.
[00:22:36] Allison Duettmann: I mean, I think that part of the reason why Foresight has so many different focus areas, including biotechnology, molecular nanotechnology, decentralized computing space, newer tech, is because we think that progress will occur at the intersection. I think we see that a lot right now, at least in terms of potential in this space that you just mentioned. There has been an incredible progress recently in the field of decentralized privacy-preserving machine learning, and there’s wonderful work at OpenMined. We had a fellow, Georgios Kaissis, who was looking at privacy-preserving machine learning for medical and health data.
I think that there is such a big opportunity for us to learn from the type of data that is the most sensitive to us, and there’s usually financial data or healthcare data. But at the same time, it also has crazy privacy restrictions for good reasons, because it is extremely sensitive and could also expose our family if not dealt with correctly. But we do see in the computing realm now a lot of really wonderful tools like these privacy-preserving machine learning tools that could really be used, I think, to leverage these data things.
You could imagine, for example, encrypting the data at the source and only letting metadata through that is still significant enough for specific machine learning algorithms to make sense of and to give you back useful information without you necessarily having to share that private data out. People are making progress on this. It’s still relatively cost ineffective as compared to large normal algorithmic approaches, but the problem is that usually these approaches don’t really have a chance to go against the existing big data trained AI models. But in a space like healthcare where we literally cannot share data, they have a comparative advantage because they’re the only ones who can actually properly make use of this data. I think it’s a really exciting space to be looking at.
[00:24:21] Camille Morhardt: I’ll take it for granted that we’ll get there on personal privacy-preserving data, and we do have policies and laws in place. They’re not universal, but everybody recognizes that. But what about the problem where we have data from a vast swath of humanity on exactly how our biology functions. That’s now sitting in the hands of essentially the few, the giant clouds, that can actually hold the literal sheer quantity of data that exists on this front. Who has access to that data? Because I know you work on having the benefits not only rest in the hands of the few. How do we tackle that kind of thing?
[00:25:00] Allison Duettmann: Well, for many of the type of healthcare data, actually you are not allowed to share it broadly. We have pretty strict regulations on this, and that’s also the data that would be very useful for us to actually make progress on longevity because many of the things that people currently are doing, if we were able to cross compare across what different people are already doing, we could really make big progress. I think that yes, in general, we have lots of our data in the cloud, but in specific healthcare fields that are really sensitive, we don’t yet. Hospitals are not allowed to share.
[00:25:31] Camille Morhardt: Right, right, right, right. But what about watch data, Garmin, Apple, all that stuff? I mean, that’s all going to a cloud without my name attached, but it exists online.
[00:25:42] Allison Duettmann: That’s true. I mean, we can talk about the risks associated with that, but I think if we want to talk about new types of opportunities that could be unlocked if we actually were able to share relatively sensitive data or not share it, but actually have it either homomorphically encrypted, user-friendship privacy, federated learning, all of these approaches that are currently being explored, I think we could be doing much more than we currently do with our data. We are already arguably sharing way too much.
I think if we made progress by funding a few of these more privacy-preserving solutions because they’re not far out from being actually applicable, then I think we could really take everything home again and we wouldn’t really have a need anymore to share everything to the cloud. Cryptography, unfortunately, is still a really underfunded field. It’s still mostly being practiced in academia.
There’s a few individual labs and so forth. But I think with more investment in these spaces, we would have really amazing tools at hand to actually be making progress on this. I should say that this not only holds for longevity, but the type of promise that we see in privacy-preserving machine learning technologies, cryptography technologies, security technologies, and auxiliary approaches, they would also hold for AI.
If we’re worried about centralized AIs becoming too powerful by having access to this big data, then I think we should be focusing much more on these decentralizing solutions that come out at the security and cryptography space, because they have an ability to allow us to make sense of the data and allow individual organizations to collaborate and actually reap the benefits of the data without actually sharing it around. We could perhaps compensate a bit for the normally centralizing dynamics that exist in AI.
[00:27:16] Camille Morhardt: Oh, that’s an interesting perspective. Disseminate it more so that we have broader access to it and therefore diminish that centralized potential power that could exist. That’s an interesting concept.
[00:27:26] Allison Duettmann: It all relies on cryptography technologies. They need a lot of love of these things.
[00:27:29] Camille Morhardt: Just so people know, we did do a podcast on homomorphic encryption and also on data anonymization. We talked with some technical experts who are working on those things if people want to get a little bit more. I think the data anonymization one talks about differential privacy as well. Allison Duettmann, CEO of Foresight, thank you very much for joining us today. I’ve really enjoyed the conversation.
[00:27:58] Allison Duettmann: Thank you so much for having me.