I think that robotics and AI paired together will be a huge threat to humanity. We may think 80% will go well. 20% or like even 1% that could go wrong is still a lot of percentage. So I think that we should as humans be prepared or thinking or trying to accelerate on a technology that will still make us relevant. We don't want to be living in a future where the current comparison between a human and a fish will be of a robot and a human. We should want to try to maintain the species because evolution and basically the Darwinian part of the evolution was always about the ability of species to evolve to try to preserve themselves to try to become stronger to adapt to the new ecosystem. If we are dropping the towel now as humans, I think our future will be already predesigned.
This is Daniel Felli. You're tuned in to the trajectory and this is the second episode in our series called 'Stewarding the Flame' where we discuss the living process. That process that creates and overcomes itself and how through innovation and regulation we might help to encourage its flourishing, potentially eventually as far beyond us as we are beyond the sea snails, and maybe even be part of that transformation. This episode is about exactly that. We've interviewed a couple folks of Italian descent here on the show and maybe a handful of billionaires, but no Italian billionaires. Paolo Ardoino is the CEO of Tether, the sort of stable coin issuer there. And his firm has invested hundreds of millions into Blackrock Neuro, which does brain-computer interface. In this particular episode, we talk about the potential necessity of brain-computer interface as a part of the next waves of evolution of intelligence and why that may be necessary for humans, but also necessary if we want the living process to persist and not be sort of trampled by potentially some kind of unconscious AGI that is too much of a separation from the living process that we do understand, which at least at present is in the biosphere. So there will be much to say at the end of this episode as always, but without further ado, this was a lot of fun. This one was recorded, by the way, live in Boston. I was not in the studio, so my background is a little bit different. Paolo's background is a little bit different. But it still came through great. I hope you'll enjoy it. This is Paolo Ardoino here on the Trajectory. So, Paolo, welcome to the show.
Thank you for having me, Dan.
Absolutely. I, you know, years ago this was called the AI and Business Podcast, but years ago some of our earliest episodes were in BCI. Braingate is based in Brown University, which I grew up in Rhode Island. And it's been such a fun space to follow, but the progress generally hasn't been that exciting compared to AI, but recently that's shifted, and obviously you've gotten into this space. There's some exciting things going on. I want to open with your vision and perspective, having now skin in the game in this domain. When you look forward a year or two at what might be capable BCI-wise that really isn't today and hasn't been in the last few years, what are the most significant things from your vantage point?
Sure. I think that with the advent of AI, I think that in the BCI field we can make important progress. So far most of the attempts to BCI have been invasive, which I believe is still the most prominent type of approach in terms of BCI, the most accurate one that provides more information and insights into the brain. There are great companies that are working on it and have been working on it in the last many years, like Blackrock Neuro, and of course you have Neuralink and many others. But I also think in the next few years we are going to see the rise of the non-invasive BCIs. There are some attempts to try to capture, in a high-level way, information coming out from the brain and use that information to govern some day-to-day tools, maybe the computer, maybe the stereo, maybe a few other things. But definitely we are going to see much more development on the non-invasive BCIs. But I would say that for the next two to three years, I expect the development still to be on the surface, high-level, and not that fast. But I truly believe that the huge unlock will come five years from now up to 10 years from now, when probably with the mass adoption of this technology in 15 years from now.
Yeah. So, and we're going to get into that whole journey because there's a lot of things to consider because, Paolo, to be frank, there's a lot of people who think we'll have AGI in, you know, 18 months from now or something like that. So if the human has to catch up, you know, in a full decade, like is that enough time to shake us out? But we'll get into that stuff. But anyway, it's good to understand your timelines writ large. Maybe we can talk a little bit about, you know, of course invasive is high bandwidth in the most literal sense of the word. But of course there's all kinds of challenges with dealing with living tissue and with dead, you know, little nanowires going into neural circuits and whatnot. Brain constantly adapting to those things and kind of resisting those things. Tons of challenges going on there. Non-invasive obviously is going to be our near-term trajectory. Let me ask, when it comes to the technical things that'll start to make the non-invasive more powerful, obviously you folks at Tether are now sort of working on some of this. So, when you think about breakthroughs, technology applications that are starting to open this up and let people use their computer or do other non-invasive things that are kind of useful, even if it's early stage useful, still very practical and useful. What's leading to those breakthroughs? What are some of the technical things that the audience should know about that we're really making headway on in the field?
Well, I think that with the information that we can extrapolate from the non-invasive BCI, we can still do very interesting things. So we have the ability now, and that is I think one of the most important breakthroughs of Tether. We have now the ability to run very lightweight models directly on the consumer smartphone or consumer device. So recently we were able to showcase our Med-PSI, that is our medical health model, the most recent one that we built. We built two versions: the 1.7 billion parameters and the 4 billion parameters one. Both of them, especially the 4 billion parameters one, is exceeding the performance and accuracy of the 27 billion parameters Med-PaLM medical health model from Google. So with one-seventh the size, we are able to be even more accurate than the 27 billion parameters from Google. So there are so many breakthroughs on the local edge compute and LLMs that allow us to be confident that when we extrapolate even imperfect information from non-invasive BCI, we can still use that to run models and to transform that information into real actions. You know, of course, again, the level of precision is probably 1/100th of what you get compared to invasive BCIs, but still it's enough for certain day-to-day actions like moving things, or turning up lights or down lights, or changing the music on your stereo with your brain, right? So you can start feeling that. We can start perfecting that information. But think about it: the more we get to experiment these things in day-to-day life, the more people we try it out, the more people will also provide their brain waves to us in a very anonymous way – not us as a company, but us as a research group. Yes, even universities, right? We want to see an open effort to use this information that comes out from the brain to be used to train models, and open models. We really believe in open source models that can be helpful to humanity, that can be used then also in the medical field more and more.
Yeah. So, you know, it strikes me, Paolo, that however many tens of thousands of years ago, language became more robust and popular, and people experimented with making noises and sounds, and now all of our brains are completely rewired to be vocally and symbolically communicating at all times. We swim in this parallel symbolic world. Obviously, I live most of my life on a computer screen, just like you do. And my fingertips are really good at a bunch of stuff that my great-grandparents never thought their fingertips would be good at. In terms of the vast majority of my work, it strikes me that even non-invasively, there will become new habits and things we take for granted that we just do mentally, we no longer do with our hands, we no longer do with our mouths. And that if we have enough people hooked into these sorts of things, it could be the banal stuff like turning on and off the lights. It could be before I go to sleep, I want to remember something that I was going to do, and so I kick off a Slack message in my mind while I'm in bed, sort of before I'm going to bed, I kick off a couple notes to myself or whatever, in some cases maybe even without having to type or do anything like that. Little habits like that might compile. And then when you look at a day in the life of an average human, even with non-invasive, you might notice many telepathic advantages in terms of speed, productivity, convenience, etc. And it seems like getting mass adoption even of that low – you said 1/100th, which is probably, I don't know about accurate, very very low bandwidth – but even at that level, if we do it in bulk, it feels like that could be pretty transformative as a first step in the first place.
I agree, especially if you can add the concept of silent speech. So you in part use the information coming directly from the brain, plus the small movements of your mouth – any additional information will make the systems much more precise. So I think, as you said, there are things like remembering information, taking notes for yourself, doing little things, turning on and off stuff, commanding things with voice mode, commanding the television, even moving the cursor on your screen. These are things that could be done also with non-invasive BCI. And so what I do care about though is the privacy of this, because we start entering the field of the most intimate thoughts.
Oh yeah, incredible security concerns here. Yeah.
Yes. So that is why we believe that having this, especially because it's low bandwidth, the information contained in those brain waves and sound speech in a certain way is limited, so we believe that local devices are perfect to analyze that because they do not require super important power. But yeah, so that's where we see the biggest breakthroughs. And we still believe that we can use existing LLMs or adaptation of existing LLMs. Our team was able, through well this in case partly comes from very complex stuff, but we were able to train models, adaptations of the Whisper model in our brain-to-text challenge, and a model that allows basically public information taken from the human brain to be used to train new models that would use brain waves to capture what people were thinking in that specific moment in time. So these things are possible. We believe that it's important to create anonymized public data sets so that researchers of all the world will be able to build the most amazing user experiences.
Yeah. Um, yeah, obviously. I mean, so there's a thinker, a fellow Italian, Luciano Floridi, who says the mark of a mature civilization is that it can be brought to its knees by a cyber attack. And I think he's basically right. And one wonders what are the new possibilities and risks of everybody looped into things that are reading their thoughts all day long and feeding those thoughts into personal AIs that help them with their habits or their work or whatever. Clearly new security concerns. Clearly some elements of decentralization and encryption and whatever else are going to have to be evolved in parallel. Any thoughts from your vantage point? Because for that ecosystem to thrive, I suspect people will need to know that they won't be bombarded with brain-customized advertisements or have politicians lightly bend their perspectives on almost everything they think about just enough to vote for the right kinds of things. People probably won't dive in without some degree of that security. What do you think is going to evolve ecosystem-wise to make people feel safe jumping into this space?
Well, there is a lot to unpack there. First of all, I think you described one of the episodes of Black Mirror. And it's a real concern. So first of all, connecting to what you were saying before, you have the huge breakthrough. You use the example of your fingers and your voice. Our voice has been an analog system, so we have almost like a digital brain, we have analog communication system that is our voice and fingers. And now we have the chance to start using something like a mathematical processor from our brain. But the more we start relying on it, it will create a huge dependency and that could be exposing the entire civilization to big threats. Completely agree with that, and that is very scary. That's why again we are pushing for at least allowing people to have full control over their own thoughts and their information. Sure, encryption, all that. But you know, I would not trust my information, my brain thoughts, to be stored on cloud. Also because if we were doing this 10 years ago, imagine that even then we had this technology that was able to allow us to extract thoughts from our brain and store them in cloud with strong cryptography. Then we already talked about quantum compute. So someone with a quantum computer could in the future break the stored information in cloud that was containing our thoughts even if encrypted. So we need to also think about: encryption is great, but also the locality of data is fundamental, because if someone had access to that information even years later with a more powerful technology, could decrypt it. So not storing that information in cloud is very, very, very important.
