From Brain Preservation to Reconstruction
Summary of the April 2018 Carboncopies workshop on whole brain emulation
Keith Wiley, 6/4/2018

On April 29th, 2018, Carboncopies hosted its second workshop of 2018. This hybrid-format workshop, viewable both online via a live interactive video stream and in person onsite in the San Francisco area, had the theme of considering how a cryopreserved brain might be emulated via whole brain emulation (WBE). This topic was chosen in response to the recent announcement by the Brain Preservation Foundation (BPF) that they had awarded their Large Mammal Prize in March for the successful preservation of a pig’s brain (http://www.brainpreservation.org/large-mammal-announcement/). The winning preservation protocol, Aldehyde-Stabilized Cryopreservation (ASC) lends itself directly to WBE and so the BPF announcement was chosen as the inspiration for Carboncopies’ second workshop of the year. The workshop addressed topics such as what sort of model parameters should be required of a whole brain emulation (WBE) so that the result may be judged a preservation of a person’s mind and identity, what features of a preserved brain should be required so as to provide the necessary information for an eventual scanned and emulated model, and what real-world contemporary examples can we draw from for inspiration or as an illustration of the potential progression of such technologies as our technological capability evolves.

The opening talk, From Brain Preservation To Reconstruction, was by Dr. Randal Koene (founder and CEO of Carboncopies). This talk introduced WBE and investigated the technical aspects of how WBE would be performed on a preserved brain, with attention paid to the open questions, such as determining which information from the brain needs to be captured and utilized in an emulation. What information can be gathered from the brain to understand and recreate memory engrams? At a lower level, Koene asked what neural parameters are the determinant factors of how neurons accumulate and perform memory functions for the retention of information in the brain from past experiences, and with the effect of altering current behavior and subsequent state changes in the brain. What are the circuit architectures and neural or synaptic properties of memory and cognition? What library or reference guide might we build of neuron and synapse types and their varying properties? Koene emphasized the need to better understand large-scale organization, such as brain regions and algorithmic models of regional behavior.

Another question Koene broached was that of determining the success criteria of WBE. How do we know that a WBE has successfully reproduced an individual’s mind? Koene proposed the concept of a neural fingerprint, some objective measurement of brain structure or function that distinctly identifies an individual, such that if a WBE exhibited the same fingerprint we could judge it to represent the same person who preceded the preservation. What sort of validation test or data would inform us on such matters? Koene proposed that in addition to neural modeling and neural circuit validation, we might also desire (or require) psychological behavioral validation by comparing the WBE’s behavior to similar behavioral tests preceding the preservation process.

Koene also emphasized the importance of acknowledging model imprecision. No model is perfect and while this realization may shut the doors to philosophical acceptance of WBE as identity preservation for some readers, it certainly will not be considered prohibitive by others. For those readers willing to accommodate realistic variations between a model and its source data, what are the tolerable model variances in terms of error or generalized noise and randomness? How precise must one neural fingerprint measurement be to another for them to fall under the same identifying label? What parameters can we utilize in our model variance tolerance function?

Koene finished his introductory presentation with a reminder that WBE can serve multiple purposes. While longevity is an oft-touted goal, another huge source of motivational objectives is the almost limitless possibilities to improve and expand human cognition as the future unfolds.

Following Koene’s talk, Dr. Keith Wiley, a board member with Carboncopies, presented Why a Whole Brain Emulation from your Preserved Brain is Probably You. This presentation was purely philosophical. It did not touch on recent or ongoing neurological experiments. The focus of this talk was what Wiley calls the copy problem, the question of whether the best interpretation of WBE of a preserved brain isn’t identity preservation, but rather that some sort of metaphysical identity copy emerges in the WBE and that the original identity is left behind in the brain. Wiley’s talk was primarily a series of counterarguments to the copy judgment. Wiley tackled multiple concerns that are commonly used to support the copy claim. Such concerns include continuity streams of consciousness or of neurological activity, the question of how identity purportedly spatially relocates from the brain to the WBE computer, and what implications should be taken from the possibility of a nondestructive process in which the person and/or brain revive despite a WBE having also been created. Wiley urged a philosophical position that personal identity is primarily psychological, instantiated by information patterns of neural configuration. As such, he specifically argued against the alternative of body identity, in which identity is physical instead of abstract and informational. Finally, Wiley proposed a variation of psychological identity called branching identity, which handles most of the paradoxes and conundrums that popularly confound body and simplistic psychological identity.

Dr. Kenneth Hayworth, from Howard Hughes Medical Institute and Janelia Research, then presented the Aldehyde Stabilized Cryopreservation (ASC) technique that won the BPF mammal preservation prizes, and then asked whether ASC is sufficient to preserve the information critical to a later WBE. Hayworth’s talk focused on the neural underpinnings of memory and how such issues inform the requirements of successful brain preservation. What features of the brain must be preserved to enable whole brain emulation? Hayworth emphasized that the starting position of such considerations is an assumption of physicalism, i.e., that the mind is solely the product of processes, ostensibly computations, occurring in the physical brain, a conclusion supported by a century of experiments.

ASC has been confirmed to enable indefinite storage of the neural features considered by modern neuroscience to be the critical components of neural function, such as connectomic topology (which neurons connects to which), as well as individual connection properties (chemical and morphological traits of synapses and neurons). The assumption is that ASC should enable virtually unlimited technological capability for revival. The only question remaining is whether sufficient information is actually preserved. Hayworth argues that whether an emulation is of sufficient quality is in the eye of the beholder and that if the mind is computational, then there is no further fact, as the saying goes, regarding imperfect copies of neural function or personal identity.

Regarding the issue of continuity of consciousness, Hayworth described how we have a self-model and associated narrative that is continually written into our long-term memory and that our identity (our sense of self) persists across breaks in consciousness, such as anesthesia, only because of our memories. We feel that our identity has survived sleep, fainting or anesthesia exclusively because our memories upon reawakening create the experience of our long-term self continuity; there is no other property at any level of abstraction, physical, psychological, or philosophical, that is relevant to this basic fact. Therefore, the crucial property in question is preservation and continuity of long-term memory, be it declarative, episodic, learned, or innate. These memories are stored in patterns and strengths of synaptic connections.

Hayworth seeks a computational description of these synaptic properties. In other words, not only does he want to understand the topology and transmissive functions of synapses, but he also wants to abstract those brute physical facts to their implied computational properties and implications for memory representation. Hayworth concluded his talk by emphasizing that ASC appears by all counts to preserve the necessary structural and molecular features of memory, and by restating his desire that the professional neuroscience community take up these questions of further validation and investigation in earnest.

Alicia Smallwood presented a concise and inspiring overview of analogous tasks and projects in a talk titled Reverse Engineering Demo: What Integrated Circuit Reverse Engineering Teaches Us About Reverse Engineering the Brain. In this informative presentation, Smallwood walked the audience through a set of videos by Ken Shirriff that demonstrated reverse engineering of integrated circuits (ICs), a problem that is quite similar to reverse engineering the brain. In both cases, a physical substrate comprises thousands to millions of microscopic components organized into a massive, complex, and convoluted signal-transmitting and signal-processing network. The goal is similar: to deduce the underlying functions performed by the physical network. The ICs are sliced apart, layer by layer, and then imaged with a metallurgical microscope. This initial step is very similar to slicing and scanning brain tissue of course. The captured images must be registered (aligned and overlapped) and then analyzed. Once components are identified and labeled, nascent comprehension of parts of the chip begins to emerge. Established abstract models (known basic models of transistor circuits) are then applied to further the image comprehension and reverse engineering of the specific chip under investigation.

Shirriff even extended this work to the next logical step: modeling of the deduced structure and function in a software emulation. This step confirmed the correctness of the deduced model and offered yet additional understanding of the chip’s observed behavior, which could have been elusive and confusing without the detailed knowledge provided by the reverse engineering efforts. Specifically, he determined that the reason the chip in question performed more slowly on larger numbers was that it would iteratively accumulate sums, such that summing to a greater total took more iterations to complete. This illumination might have been speculated from external behavior, but could only be solidly confirmed through low-level physical analysis, i.e., reverse engineering.

Smallwood and Shirriff concluded with extensions beyond digital ICs, namely to analog chips such as the 555 timer. This talk gave great insight into the process of reverse engineering a neural and synaptic network to deduce algorithmic functions of the brain. While differences between ICs and brains can easily be listed, the principle is nevertheless similar and the presentation was fascinating in that regard.

Jonathan Gornet, from New York University, then presented Dynamical Modeling of Extracted Connectomes. How do we discover and determine the Drosophila (fruit fly) connectome, and then how to do simulate it on a computer (a fly WBE). Gornet and others have been slicing and imaging fly brains with electron microscopy, layer by layer, to reconstruct whole neurons (a process frequently known as neurite tracing). Gornet discussed the task of correlating reconstructed morphology with dynamic measurements taken prior to the sectioning, imaging, and emulation process. Gornet walked through a particular neural circuit model in which a neuron is represented by its branching structure, with each branch represented by an action potential function. A neuron then comprises a branching series of such functions, and we can follow an action potential along some sequence if these functions as it traverse a neuron.

Gornet addressed the concern of validation too. He described a motion selectivity experiment in which light bulbs are illuminated from left to right and the membrane potential of a single neuron is observed and modeled. At this stage he simplified his model, representing whole neurons as single functions instead of individual branches. The model’s behavior was confirmed to be virtually unaltered by this simplification while reducing the model running time by a factor of forty. An implication of this result was that synapses are the critical feature of signal transmission and that population structure is more important than individual neuron geometry.

Randal Koene returned with a second presentation titled Model Validation and System (De)composition at Multiple Scales. In Koene’s second talk, he asked what information is needed to insure that a WBE works. What information is available from a preserved brain and what additional information (if any) is required beyond that provided by a preservation, such as dynamic recordings made prior to preservation? Koene compared building a WBE with reserve engineering an IC, as had been presented earlier by Alicia Smallwood. He emphasized the shared trait of large nonlinear dynamic models, which is the underlying feature of such models. Koene asked whether there might be a dynamic fingerprint that can be compared with an emulation for the purpose of validation. He prescribed an iterative process of parameter estimation, model validation, corrective modification and model selection, and repeated parameterization and validation, ultimately converging on a successful model. Where does this validation data come from? If it is recorded dynamically prior to preservation, then what spatial and temporal scales are required of such data to capture the necessary information for later model validation?

Prof. Dong Song, from the University of Southern California, presented Towards a Clinical Hippocampal Memory Prosthesis. Song’s talk presented research into dementia, the biggest neurological disease in the U.S., and for which there is currently no cure. Techniques that have helped some diseases, such as direct brain stimulation (DBS), applied toward Parkinson’s with great success, do not work at all on other maladies, such as hippocampal deficits. In fact, DBS can make the hippocampal function even worse. Song described a biomimetic device that mimics hippocampal function and and bypasses the damaged hippocampal brain region, a topic that Ted Berger has presented on in previous Carboncopies workshops.

Song presented a classic variant on animal (rat) lever-pushing experiments. In this case, neural firing patterns were recorded in conjunction with these behavioral experiments and then those patterns were computationally modeled as spatiotemporal action potential firing patterns. This recorded code was then used to restore neural function after the memory had been forgotten. Restoration of the memory proceeded by directly stimulating neurons with the previously recorded firing pattern. The rats successfully reproduced behavior associated with the lost memory, demonstrating the successful recording and later external stimulation of memories as neural circuits and firing patterns.

Song has moved on to human trials with clear applications to dementia. Early experiments with humans attempting to remember abstract visual stimuli (visual patterns) have been encouraging and confirm the approach of applying model-based memory restoration.

The workshop concluded with a panel discussion titled Restoring Jane Doe from her Preserved Brain, hosted by Dr. Diana Deca from the University of Southern California and Dr. Stephen Larson of the OpenWorm project. Deca and Larson led other participants through a discussion of whether an emulation of a preserved brain should be interpreted as a preservation of identity. One outcome of this discussion was the suggestion that we be open to the interpretation of a partial chance of identity preservation instead of insisting on judging the matter in the binary terms of success and failure.

The April Carboncopies workshop asked a specific question. Given the recent BPF announcement that we now have a viable long term brain preservation method, what is the subsequent path forward for developing a WBE method to emulate a preserved brain, and relatedly to philosophically interpret a WBE as a preservation of personal identity. We will not be able to perform WBE of a human brain for a long time, but the path forward seems relatively clear. By extrapolating and extending current research such as hippocampal prosthesis, and by considering related challenges of reverse engineering complex networks, such as integrated circuits, and by carefully considering which neural and cognitive parameters should be deemed salient and required, we can lay out a roadmap for research and development toward the eventual emulation of a whole human brain from a long term preservation.

Carboncopies’ next workshop is currently scheduled for August 2018 with the topic of delving more deeply into the philosophical questions of consciousness and personal identity.

 

Keith wiley serves on the board of Carboncopies as director of communications. His book, A Taxonomy and Metaphysics of Mind-Uploading, is available on Amazon.

 

The workshop’s archived URL is https://www.carboncopies.org/workshop-2018-apr and includes links to the videos. The same URLs are offered below in chronological order: