DARPA Dreams: Cortical Modems and Neural RAMplants for Restoring Active Memory
A compelling encore to inventing the Internet appears more elusive for DARPA with each passing day. But what if they could put the Internet in your brain? That might turn some heads. Before anyone gets around to that, it would seem prudent to try to put just a single idea into a mind — a single memory, skill, or replay of some notable event. With a news release announced yesterday, DARPA intends to do just that.
The solicitation is for a two-year research program to develop what they are calling RAM Replay, or implants for ‘Restoring Active Memory’. This concept has not been hatched in full form out of thin air, but rather builds on a succession of recent projects to build an interface that is active at the level of the conscious mind — a cognitive implant, if you will. Just this February DARPA sponsored a meeting in the Silicon Valley area where they discussed futuristic conceptions for what they call a cortical modem.
In its rawest form, they imagine such a cortical modem might be a $10 device the size of a couple of coins. It would enable a visual impression to be perceived by appropriate stimulation of the visual cortex. That’s a tall order, but perhaps a bit more modest then their previous REMIND (Restorative Encoding Memory Integration Neural Device) project, which would be nothing short of a full-blown memory prosthesis that jacks into the hippocampus.
For various reasons, the idea of an implant to restore memories in the sense of actually reconstructing lost memories is not just premature, but downright preposterous. Once a memory is gone, anything put back could hardly be reliable. The more modest goal — restoring the ability to store memories — would be difficult enough even if one had access to the entire brain, to say nothing of the difficulty of trying to do it just from the hippocampus alone.
On the other hand, a mere replay of some event, in the form of a visual overlay with the resolution of, say, a digital alarm clock, might be something one could work with. At this point we would need a more descriptive term for exactly what it is we are talking about. If we restrict ourselves out of convenience to deal with vision alone, and reserve the term virtual reality to mean those artificially presented external simulations that replace our normal visual world, then the combination of normal vision and virtual reality is already something we are familiar with. That would simply be a visual overlay or a ‘HUD’, a heads up display.
But here we are not talking about an overlay for the eye, but rather an internal display manufactured from the wetware of the brain itself. That totality, i.e. normal vision, plus any virtual reality overlay, plus internally generated visual perceptions, we would propose for our purposes here to call the ‘visuosphere’. How does one then create a visuosphere, and do it safely, reversibly, and in a way that retains some measure of coherence with reality? Furthermore, how does one localize that window to a specific place in the mind’s eye, like one would a ‘toast’ or notification on their smartphone? Those are the questions I think DARPA is asking.
Our purpose here is to describe something that is no less than an open-ended solicitation for an impossibly futuristic transhumanist device. What better way to do that then to submit such a proposal ourselves? To that end, we would start with the premise that any implant with the capability to generate part or all of a visuosphere should first and foremost ‘do no harm’ to the user. To be consistent with that ideal, one would have to be crazy to use anything like the currently-envisioned devices that are simply plunged into the cortex proper itself.
The most advanced cortex implants money can now buy are made by a company called Blackrock. Their devices are behind some of the most successful brain-computer interfaces that have been used to date. They can do once-fantastical things like get paralyzed people to command a robotic arm to give them food or drink. But at the end of the day, these are not forever devices; when these pincushion arrays are eventually removed from the cortex, they leave their mark.
What we would propose instead is not to stimulate the cortical gray matter, but rather the white matter, the axons that project out of the cortex. The reasons for this would be as follows:
1. Safety. Axonal overstimulation is more survivable for the cell. You may kill the axon, or just that axon branch, but generally not the whole neuron. In principle and practice, the so-called initial segments and periodically spaced nodes of axons are much more excitable and therefore controlled by lower currents. The new ‘optocapacitance’ technique we recently discussed is a good example of a way to stimulate axons without high currents or questionable optogenetic manipulations.
2. Access. A neuron’s activity is accessible at each point along the axon, rather than just at one point near the cell body. That means you can potentially get two or more redundant reads on a cell to determine the direction and potential connectivity of its projection. To actually target axons relevant for vision, the place you want to put your hardware is inside the lateral ventricles of the brain. The axons that bring information from the thalamus (and ultimately the retina) to the visual cortex line the walls of this area. Furthermore, the return projections, which is what you may really want to target to keep the primary visual inputs unmolested, is actually 10x numerically enriched relative to the upstream projection.
3. Reversibility. If companies have learned anything in the neural implant business, it’s that there is no real market for brain implants for just the few handicapped persons that need them most. Just ask anyone walking around today with a useless orphan brain implant manufactured by a now-defunct company. A successful implant business needs to make implants that are for everyman. That means that when they are taken out, it’s back to business as usual for the brain. An implant that targets white matter axons from inside the ventricles can not only be easily removed, but also the cell it accesses doesn’t even have to know it is there when it is being used. The neuron can go on integrating local visual cortex information on its dendrites and sending out its own pulses. But the implant can intercept them, commandeer them, and change them to whatever it desires — all with the cell body being no worse for the wear.
Obviously we have made several simplifying assumptions in all this. But the idea of a white-matter access point for visual, memory, or any other implants is an idea whose time has come. The return visual projections we mentioned specifically above do have their own important unique role in the visual system that would need to be taken into consideration in the design of any implant. However, when compared with the alternative of blasting away at the visual cortex itself, targeting these anatomically defined byways first would seem like a good initial strategy. We would claim that any research effort that does not recognize and begin with these obvious truths, is devoid of any real focus, and ultimately, perhaps even bogus.