Episode 1 | Transcript

Episode 1 | Transcript

Episode 1: "This Is Mind Uploading and Why Humanity Needs It"

Published Thur, 18 Aug 2018 | Transcript

Allen Sulzen: Welcome to the first episode of the Carboncopies Podcast. My name is Allen Sulzen and I'm pleased to host our first series of presentations, featured originally as part of the Carboncopies Workshop. You'll hear an array of different scientists, scholars, and professionals throughout this series that aim to reach a greater understanding of the human mind and how we will eventually reach whole brain emulation, commonly referred to as mind uploading. Carboncopies.org is dedicated to advancing research in order to preserve, restore, and even improve your mental experience beyond the limits of biology. If you're thinking: "that seems like a lot", you'd be right. The road ahead is long, but the path, as you'll hear in this series is becoming increasingly clear. The presentation you're about to hear was given by Dr Randal Koene, Chairman of Carboncopies.org. He explains mind uploading why humanity needs it. I give you Randal Koene.

Randal Koene: Most people who are joining today have probably already heard of either mind uploading or whole brain emulation in one way or another. I'm just going to say a few words about why we at Carboncopies have made it our mission to push for its development and accomplishment and why we think it's important to humanity or human society as a whole. So in a world where physics, biology, chemistry, currency, and even intelligence are becoming code, one of the statements that I'm making is that our species can flourish only if we also come to grips with ourselves as information processes of sorts. The purpose of this introductory talk is to introduce concepts and to provide the context for the remaining sessions and speakers in the workshop. If we're going to talk about mind uploading and whole brain emulation, it's important that we understand what we all mean by that, that we have some common terminology.

We begin with what I would call the experiential perspective or human experience to that which a person or a human being experiences. We could say that it's an abstraction level of sorts. What it does include, for example, are our cognitive faculties such as sensation and perception, learning and memory goals and desires and awareness and several others. Of course, without those, it's really hard to imagine the human mental life, but what the experiential perspective doesn't include, for example, are interactions of atoms and ions and the chemistry of cells, metabolic processes, molecular signaling processes, and I know that last one is potentially going to sound a bit controversial to the neuroscientists listening to this, but what I'm trying to say is at this moment I really have no idea which synapse in my brain was involved with what I was thinking or which receptor channels were being activated or which ones tried to activate but failed.

All of these little details about the underlying biology are not part of my experiential cognitive existence. Of course, that doesn't mean that those things don't matter at all and the cutoff is definitely far from absolute. What it does mean is that we need to recognize that depending on our goals, different abstraction levels may require particular attention. So for example, if you're trying to cure Alzheimer's disease, then it's probably really important to pay close attention to molecular processes. But if you're trying to build an intelligent machine or a device that replaces the information processing carried out in a small part of the brain, then it's really the effect on the cognitive experience that matters most directly, and this is where I put up the analogy. Imagine for example, that you want to run an internet browser on a computer. You don't want to have to care about the precise layout of every transistor involved in that process. Otherwise, for example, it would be impossible to design an upgrade to new CPUs.

All you want to do is you want to make sure that every CPU is able to run the same program code and produce a satisfactory user experience. So there is an abstraction level that's important. In this way we could say that the path to a successful mind upload begins with identifying your success criteria. For many of us, that means at a minimum, retaining what our conscious day to day experience is presently like, and then aiming to upgrade to an ever better performance. Now, the term emulation or emulator has been used to describe the separation of scales. Being able to abstract away the details below the level of interest and still ending up with a desired result. In the example of the Internet browser, as long as the program code runs properly, everything below that can be accomplished in a variety of ways. In fact, in that example, it might be sufficient just to ensure that the browser interprets html coded websites correctly, allowing even the underlying software to be rewritten from version to version.

Similarly, a computer system can run an emulator to permit using programs that were designed to be run on a different or older computer system. Now our cognitive processes are normally run by the human brain. Our default hypothesis is that everything about our mind that is relevant to the human experience is an emergent result of the processes is taking place in the brain. And to date, that's a hypothesis that has passed every test with flying colors. We see that lesions in the brain directly correspond to changes in cognitive behavior and we haven't yet found anything that, uh, that would indicate something lying beyond this brain processing or human cognition.

So if we can run cognitive processes by emulation on another platform, then that ultimately results in what we're calling whole brain emulation. We sometimes then call that substrate independence because a brain emulator could be implemented in many different ways, no longer necessarily depending on biology. Now in neural engineering today, the effort to build neural prosthesis is the earliest attempt to achieve something that is closely related to brain emulation. A neural prosthesis is meant to replace or support the cognitive function that's carried out by some set of neural circuits in specific regions of the brain. The prosthesis is built by creating computational models of the processing carried out by those neural circuits and those models are then run on a computer or an embedded in some device that interfaces and communicates helping neural tissue a patients. The more accurate the model that was built using brain data, the better the implementation can be.

The principal operators in the brain are called neurons. Tiny simple processors that only experienced an increase or decrease in their membrane potential. And at some threshold they respond with an electric discharge. And ultimately, it's the orchestration of billions of neurons that is the information processor that plays the symphony that's our experience of being. You can imagine that progressive neural prosthesis for every piece of the brain is ultimately identical to whole brain emulation. Now, it's pretty easy to see why individual neural prosthesis are useful and desirable at a minimum they help with a brain dysfunction, for example, and beyond that, they hold the promise of augmented capability. It's also pretty easy to see why mapping and modeling brain functions is important to science, medicine and the development of analogous functions in AI. But why does Carboncopies specifically promote attention to and advances towards whole brain emulation?

The biological brain isn't optimized for access or monitoring and it's not optimized for modification or augmentation. Once neural prosthesis reaches the point where you can emulate the whole brain, you reach a very special milestone as long as part or all of cognition rely on the biological brain, there are hard limits to the sort of cognitive function that's possible. I'm trying to show that here in this a simple graph where say you know the range of things that people can do and experience is a fixed amount of stuff. There's a limited range of what we can do and what we can experience. Then if you have neural prostheses, then you can do a bit more because you could tie in augmentation that is carried out on a device and tie it into the rest of the brain, but of course you know as long as there is still a part of that cognition that relies on the biological brain, well, you're still going to have to deal with those limits.

So for example, the limitations of neural responses mean that it's impossible to cognitively be aware of or respond to events at a microsecond scale. Something that machines are capable of doing. So this is the human thriving arguments for whole brand emulation and understanding that comes with it to expand the range of possible human experiences and capabilities instead of submitting to the narrowly constrained way in which we can experience the universe today and ceding the bigger picture to machines. There's also the important survival argument though, without being able to actually modify our mental abilities, we're constrained to what I might call an evolutionary niche of sorts, and the history of evolutionary change demonstrates that it's very likely that this niche may disappear and that humans may soon play an increasingly minor role in the future society of intelligence. Adapting to such change may very well be a survival requirement, so the main reasons we have are both thriving and survival.

This ability to move beyond the limits of what the biology gives us now since Carboncopies was founded back in 2010 and even a couple of years before that. We've been keeping track of the biggest challenges and possible roots to solutions. It began with the 2008 roadmap that came out of the whole brain emulation workshop at Oxford. The purpose of this roadmap is to determine how to map the structure of the brain, how to record the response characteristics of its neurons and synapses, how to build a working model with parameters defined by those data maps, and then how to bring to life all of that in a real implementation of the whole brain emulation. So what are the biggest challenges today that need to be overcome for neural prosthesis and ultimately whole brain emulation to be possible. The greatest challenge is undoubtedly still the problem of acquiring sufficient data from an animal or human brain.

To build a prosthesis, you need to be able to adequately approximate how specific set of neural circuitry transforms incoming signals into outgoing signals. That process of approximation is called system identification. It's a systematic process of developing a computational model of the transformation from brain data. The more we know about the structure of the circuits in the neural tissues, the better we can constrain the possible models, reducing the search space and easing the computational burden for system identification today. Developments in the area of connectomics seek to give us access to both the general and patient specific neurocircuits structure. Knowing structure alone, though is not enough to build a working relationship. After all the 3D connectome image is certainly not a working emulation. More inference is needed to convert a structural diagram into a working model in which the specific response functions of the components were correctly selected from a range of possibilities.

We do not have these libraries of possible response functions and ways to select them and their parameter values. Today, functional recording with implanted electrodes is the most feasible way to collect data about the actual response to the system and statistical regression or machine learning can then be used to build models and fit parameters constrained by what we know about the circuit structure. It's this method of functional recording followed by models building upon the known surface structure of the CA3CA1 transition and hippocampus that the Berger lab at USC and its associates and collaborators use in their neural prosthetic efforts. In addition to identifying, tracking, mapping, and seeking solutions for scientific and technical challenges, Carboncopies also tries to draw attention to opportunity challenges. What I mean is there are problems around projects, collaborations or organizations in which researchers can take on the challenges. Most of these projects and collaborations still depend entirely on academic groups, but there is no limit to the variety of organized approaches that we consider explore and support.

In the past few years there was opportunity to explore how private startup enterprise could enter the mix. Incentives and obligations in the for profit structure are different than those in academic settings leading to a different set of advantages and disadvantages. It makes sense that there are types of projects on the roadmap to whole brain emulation that are well suited to development in a startup and other types are better suited to development in academia. A couple of high profile efforts along these lines of had a little over a year to gain traction at this point. Kernel and Neuralink are the ones we mentioned in our schedule. Their founders declared that their companies exist to create bridges between human and machine, even to achieve neuroplasticity, to lift up humanity. What do these companies contributes that is difficult or impossible in academia and what may still be beyond their reach? This year, our team of Carboncopies researchers is launching a technological review project.

This is because a living and useful roadmap needs ongoing renewal and the tech review is going to be a major input to that. When applications of neurotechnology are discussed in general, when we talk about neural interfaces and prosthesis and even brain emulation, access to brain data is such an urgent problem at the front of the queue that questions about how to interpret and use that data are often set aside for later and yet getting from a pile of data to a working prosthesis or brain emulation is nontrivial. Indeed, it's an entire research domain in its own right. So how do you approximate brain function from brain data? How do you do it so well that the result satisfies the criteria for a successful emulation? This is why we picked the special topic: Discovering neural circuits in brain data. Whole brain emulation depends on a lot of things.

Even neural prosthesis depends on many, many things and so these dependencies and the opportunities, they lead to a kind of temporal order that makes sense. And what I'm trying to show here is that we might, for instance, imagine that we're going to get some kind of application that is either for patients or even for normal users that involves noninvasive recording as a neural interface and that can give you a brain diagnostics or something else and stimulation interfaces. Of course they're going to be used in some form or another. Perhaps we'll start wearing sensors and various interfaces a lot of the time. Then at some point, neuroprosthetic medical devices are going to start popping up in patients and you could say that patients who received these implants, these devices may end up being the first people with some kind of super power.

Let's say for example, that you have a retinal implant and the retinal implant is tuned to allow you to see x-rays. Well, this is a kind of superpower. Or let's say that you have an implant that replaces a piece of the hippocampus. Then perhaps you can in some way categorize what is happening over time in that implant and you can choose which memories you want to strengthen and which ones you don't want to strengthen our which ones you want to help recall more easily and that way you have a way of selecting among your new memories that you're forming in a way that we currently don't have and that itself would be a kind of super power. Even if perhaps the quality of those first prostheses and patients won't be perfect. You know, even if perhaps they don't work quite as well as a biological hippocampus does. They may still have the ability to do certain things that we currently can't. Then eventually, hopefully you'll see a neural prostheses in many regions of the brain and at some point you could say that leads to whole brain emulation either directly through this multiple neural prostheses approach or a parallel approach where we learn how to replicate brains by knowing more about how the structure can be, how the function can be inferred from the structure. So this is sort of an outline of a possible future.

Allen Sulzen: So that wraps up our first episode. If you enjoyed it, please visit us at Carboncopies.org to learn more and get involved.