If we’re relying on a future superintelligence to reconstruct our brains, why not make it a little harder?
There’s no reason you couldn’t buy a wearable camera that recorded your inputs and outputs, and back everything up to hard disks in HD. Much cheaper to store than a frozen brain. After a few decades of video, there would have to be more than enough data to do the reconstruction. Then, when you die, you just stick the big stack-o-harddrives into a vault and wait for the future AI overlord to find them, scan them, and put them back together into a person again. Boom. Immortality on the cheap.
Again, it is sometimes suggested that recording enough of the sensory experiences and actions would be enough to produce brain emulation. This is unlikely to work simply because of the discrepancy in the number of degrees of
freedom between the brain (at least 10^14 synaptic strengths distributed in a 10^22 element connectivity matrix) and the number of bits recorded across a lifetime (less than 2 * 10^14 bits (Gemmell, Bell et al., 2006)).
I would really like to develop a good argument about when reconstructing a mind from its inputs and outputs is possible. Being a slice-and-dice favoring WBE thinker, I am suspicious of the feasibility. But am I wrong?
It is not too hard to construct “minds” that cannot be reconstructed easily from outputs. Consider a cryptographically secure pseudorandom number generator: watching the first k bits will not allow you to predict the k+1 bit with more than 50% probability, until you have run through the complete statespace (requires up to ~2^(number of state bits) output bits). This “mind” is not reconstructible from its output in any useful way.
However, this cryptographic analogy also suggests that some cryptographic approaches might be relevant. Browsing a paper like Cryptanalytic Attacks on Pseudorandom Number Generators by Kelsey, Schneier, Wagner and Hall (PDF) shows a few possibilities: input-based attacks would involve sending various inputs to the mind, and cryptoanalyzing the outputs. State compromise extension attacks make use of partially known states (maybe we have some partial brainscans). But it also describes ways the attacks might be made harder, and many of these seem to apply to minds: outputs are hashed (there are nontrivial transformations between the mindstate and the observable behavior), inputs are combined with a timestamp (there might be timekeeping or awareness that makes the same experience experienced twice feel different), occasionally generate new starting state (brain states might change due to random factors such as neuron misfiring, metabolism or death, sleep and comas, local brain temperature, head motion, cell growth etc). While the analogy is limited (PRNGs are very discrete systems where the update rule is simple rather than messy, more or less continuous systems with complex update rules—much harder to neatly cryptoanalyze) I think these caveats do carry over.
But this is not a conclusive argument. Some minds are likely nonreconstructible (imagine the “mind” that just stores a list of its future actions is another example: it can be reconstructed up until the point where the data runs out, and then becomes completely opaque), but other minds are likely trivially reconstructible (like the “mind” that just outputs 1 at every opportunity). A better kind of argument is to what extent our behavioural output constrains possible brain states. I think the answer is hidden in the theory of figuring out hidden Markov models.
Well, if I’m doing morningstar rhetoric, I’d best get my game face on.
First, in the paper below, their estimate of the information density of the brain is somewhat wrong. What you actually need is the number of neurons in the brain (10^11), squared, times two bytes for half-precision floating point storage of the strength of the synaptic connection, plus another two bytes to specify the class of neuron, times two, to add fudge factor for white matter doing whatever it is that white matter does, which all works out to 6.4 * 10^23.
Now that we’ve actually set up the problem, let’s see if we can find a way it might still be possible. First, let’s do the obvious thing and forget about the brain stem. Provided you’ve got enough other human brain templates on ice, that can probably be specified in a negligable number of terrabytes, to close enough accuracy that the cerebral cortex won’t miss it. What we really care about are the 2 10^10 neurons in the cerebral cortex. Which brings out overall data usage down to 1 10^22. Not a huge improvement, I grant you, but we’re working. Second, remember that our initial estimate was for the ammount of RAM needed, not the entropy. We’re storing slots in a two dimensional array for each neuron to connect to every other neuron, which will never happen. Assuming 5000 synapses per neuron, that means that 5000 / ( 2* 10^9) of our dataset is going to be zeroes. Let’s apply run-length encoding for zeroes only, and we should see a reduction by a factor of a hundred thousand, conservatively. That brings it down to 10^17 bits, or 11 petabytes.
Now let’s consider that the vast majority of connectomes will never occur inside a human brain. If you generate a random connectome from radio noise and simulate it as an upload, even within the constraints already specified, the result will not be able to walk, talk, reason, or breathe. This doesn’t happen to neurologically healthy adults, so we can deduce that human upbringings and neural topology tend to guide us into a relatively narrow section of connectome space. In fact, I suspect that there’s a good chance that uploading this way would be a process of starting with a generic human template, and tweaking it. Either way, if you took a thousand of those randomly generated minds, I would be very surprised if any of them was anything resembling a human being, so we can probably shave another three orders of magnitude off the number of bits required. That’s 10^15 bits of data, or 113 terrabytes. Not too shabby.
Based on this, and assuming that nights are a wash, and we get no data, in order to specify all those bits in ten years, we would need to capture something like 6.4 megabits a second of entropy, or a little less than a megabyte. This seems a little high. However, there are other tricks you could use to boost the gain. For example, if you have a large enough database of human brain images, you could meaningfully fill in gaps using statistics. For example: if, of the ten thousand people with these eight specific synaptic connections, 99% also have a particular ninth one, it’d be foolish not to include it.
In short, it seems somewhat infeasible, but not strictly impossible. You could augment by monitoring spinal activity, implanting electrodes under your scalp to directly record data on brain region activation, and by the future use of statistical analytics.
Now, actually deducing the states of the brain based on its output, as you said, might be difficult or impossible enough to put an end to the whole game before it starts. Still, it might actually work.
What you actually need is the number of neurons in the brain (10^11), squared
But the vast majority of neuron-pairs is not connected at all, which suggests storing a list of connections instead of the full table of pairs which you propose. If every neuron can be specified in 1KB (location, all connections), we’re talking ~100 TB, about $10.000 in hard disks or less in e.g. tape media.
Of course, actually getting all this data is expensive, and you’d probably want a higher level of data security than “write it to a consumer hard drive and store that in a basement”.
1 KB seems very optimistic. Uniquely identifying each neuron would require the log of the number of neurons in the brain, or 36 bits. Figuring five thousand connections per neuron, that’s 36 5000 to store which synapse goes where, and (64 + 36) 5000 to store which synapse goes where, plus the signal intensity and metadata. In short, it’d actually be more like 500 KB per neuron, or 50,000 TB.
If we’re relying on a future superintelligence to reconstruct our brains, why not make it a little harder?
There’s no reason you couldn’t buy a wearable camera that recorded your inputs and outputs, and back everything up to hard disks in HD. Much cheaper to store than a frozen brain. After a few decades of video, there would have to be more than enough data to do the reconstruction. Then, when you die, you just stick the big stack-o-harddrives into a vault and wait for the future AI overlord to find them, scan them, and put them back together into a person again. Boom. Immortality on the cheap.
Sandberg & Bostrom are skeptical. Page 109:
Also, Sandberg in 2010:
Also http://www.aleph.se/andart/archives/2012/04/how_many_persons_can_there_be_brain_reconstruction_and_big_numbers.html
Ah, that’s the post I was primarily looking for, but couldn’t find for some reason.
Oh well, the last resort would be to hope that the future AI will just recreate most possible past humans.
Well, if I’m doing morningstar rhetoric, I’d best get my game face on.
First, in the paper below, their estimate of the information density of the brain is somewhat wrong. What you actually need is the number of neurons in the brain (10^11), squared, times two bytes for half-precision floating point storage of the strength of the synaptic connection, plus another two bytes to specify the class of neuron, times two, to add fudge factor for white matter doing whatever it is that white matter does, which all works out to 6.4 * 10^23.
Now that we’ve actually set up the problem, let’s see if we can find a way it might still be possible. First, let’s do the obvious thing and forget about the brain stem. Provided you’ve got enough other human brain templates on ice, that can probably be specified in a negligable number of terrabytes, to close enough accuracy that the cerebral cortex won’t miss it. What we really care about are the 2 10^10 neurons in the cerebral cortex. Which brings out overall data usage down to 1 10^22. Not a huge improvement, I grant you, but we’re working. Second, remember that our initial estimate was for the ammount of RAM needed, not the entropy. We’re storing slots in a two dimensional array for each neuron to connect to every other neuron, which will never happen. Assuming 5000 synapses per neuron, that means that 5000 / ( 2* 10^9) of our dataset is going to be zeroes. Let’s apply run-length encoding for zeroes only, and we should see a reduction by a factor of a hundred thousand, conservatively. That brings it down to 10^17 bits, or 11 petabytes.
Now let’s consider that the vast majority of connectomes will never occur inside a human brain. If you generate a random connectome from radio noise and simulate it as an upload, even within the constraints already specified, the result will not be able to walk, talk, reason, or breathe. This doesn’t happen to neurologically healthy adults, so we can deduce that human upbringings and neural topology tend to guide us into a relatively narrow section of connectome space. In fact, I suspect that there’s a good chance that uploading this way would be a process of starting with a generic human template, and tweaking it. Either way, if you took a thousand of those randomly generated minds, I would be very surprised if any of them was anything resembling a human being, so we can probably shave another three orders of magnitude off the number of bits required. That’s 10^15 bits of data, or 113 terrabytes. Not too shabby.
Based on this, and assuming that nights are a wash, and we get no data, in order to specify all those bits in ten years, we would need to capture something like 6.4 megabits a second of entropy, or a little less than a megabyte. This seems a little high. However, there are other tricks you could use to boost the gain. For example, if you have a large enough database of human brain images, you could meaningfully fill in gaps using statistics. For example: if, of the ten thousand people with these eight specific synaptic connections, 99% also have a particular ninth one, it’d be foolish not to include it.
In short, it seems somewhat infeasible, but not strictly impossible. You could augment by monitoring spinal activity, implanting electrodes under your scalp to directly record data on brain region activation, and by the future use of statistical analytics.
Now, actually deducing the states of the brain based on its output, as you said, might be difficult or impossible enough to put an end to the whole game before it starts. Still, it might actually work.
But the vast majority of neuron-pairs is not connected at all, which suggests storing a list of connections instead of the full table of pairs which you propose. If every neuron can be specified in 1KB (location, all connections), we’re talking ~100 TB, about $10.000 in hard disks or less in e.g. tape media.
Of course, actually getting all this data is expensive, and you’d probably want a higher level of data security than “write it to a consumer hard drive and store that in a basement”.
1 KB seems very optimistic. Uniquely identifying each neuron would require the log of the number of neurons in the brain, or 36 bits. Figuring five thousand connections per neuron, that’s 36 5000 to store which synapse goes where, and (64 + 36) 5000 to store which synapse goes where, plus the signal intensity and metadata. In short, it’d actually be more like 500 KB per neuron, or 50,000 TB.
Granted that’s before compression, but still.
Wearing a camera and mic at all times is costly in non-monetary ways (e.g., in convenience and signaling).
Some of them are pretty discrete, apparantly. I looked into it, and you can buy bluetooth headsets with cameras built into them.