- #1
Cincinnatus
- 389
- 0
What would it mean to "transfer data directly from brain to brain"? We can assume that "data" in the brain amounts to some spatial-temporal pattern of activation across many neurons. This is a very reasonable assumption to make in light of what we know about neuroscience.
There's no way to reproduce the same spatial-temporal pattern of activation in a different brain since all brains are different, the cells are not all the in the same places and have different connections between them. So transferring information this way is a nonstarter.
How can we get around this problem. Well what we care about isn't the activity pattern itself, we care about the information. It seems likely that for any piece of information that we are capable of learning, then there will be a corresponding pattern of activity for each person, though this will likely not be the same pattern across people.
So what we need is a decoder. A black box that we can input "brain states" and will output "semantic information". This black box will need to be highly person specific. Maybe even situation specific in other ways as well.
How could we build such a machine? One route that people have been going has been to utilize techniques from machine learning. Say we have some machine that measures brain activity (like an fMRI scanner). We put a person in this machine and we have them do a large number of "training tasks". For example, say we want to build a machine that can always beat you in card games by cheating and "reading your brain" to see which cards you have. We would first individually have you look at each card and have our machine record the state of the relevant parts of your brain when you look at that particular card. Then we could have the machine read your brain later and return which card you are looking at. This would work as long as the particular "brain state to information mapping" is stable over time.
You might say that is kind of cheating. This machine can only decode what you're thinking if it's been trained on the exact same information that it will be decoding. Luckily, this difficulty can be overcome by techniques from machine learning. Recent work has shown that it is possible to categorize over a billion novel images from V1 activity measured using fMRI (see Gallant et al 2008).
So let's pretend now that we have a universal decoder that works for converting any brain-state to its corresponding semantic state. Let us also pretend that we have dealt with the issues of context that may cause the same brain state to correspond to different demantic states in different situations.
Now if we still want to transfer information from brain to brain we encounter the encoding problem. It is not clear if this is more difficult than decoding or not. The same machine learning procedure would work. We need to decode the mapping from brain states to semantic states for both participants and convert the information appropriately.
The final step is to bring about the activity pattern corresponding to the new knowledge in the recipient subject. This is a more difficult problem and it is one that is not well understood. The most similar device that I am aware of is the cochlear implant. Unlike a regular hearing aid which merely amplifies sounds. The cochlear implant directly stimulates the auditory nerve in a way that the brain can interpret as sound.
But what does this mean the brain "interprets the pattern of stimulation as a sound". The cochlear implant always does the same thing, there is a direct mapping from the frequency of its vibration due to sounds and the stimulation that it provides the auditory nerve. Inititially this is not sufficient for hearing. There is a learning period in which the brain comes to interpret this new pattern of stimulation as hearing sounds. That is, the brain learns the mapping from auditory nerve stimulation to sounds in the world. The brain is actually on its own performing a similar decoding procedure to what our decoding machine would have done to beat us in that card game.
It is possible that we could use a similar device to bring about any brain state in another person. IF (and this is a huge if) this is indeed doable it would then be in the realm of speculative possibility that we could link such a device to the appropriate decoding and encoding machines and only THEN could we transfer information "directly" from brain to brain.
There's no way to reproduce the same spatial-temporal pattern of activation in a different brain since all brains are different, the cells are not all the in the same places and have different connections between them. So transferring information this way is a nonstarter.
How can we get around this problem. Well what we care about isn't the activity pattern itself, we care about the information. It seems likely that for any piece of information that we are capable of learning, then there will be a corresponding pattern of activity for each person, though this will likely not be the same pattern across people.
So what we need is a decoder. A black box that we can input "brain states" and will output "semantic information". This black box will need to be highly person specific. Maybe even situation specific in other ways as well.
How could we build such a machine? One route that people have been going has been to utilize techniques from machine learning. Say we have some machine that measures brain activity (like an fMRI scanner). We put a person in this machine and we have them do a large number of "training tasks". For example, say we want to build a machine that can always beat you in card games by cheating and "reading your brain" to see which cards you have. We would first individually have you look at each card and have our machine record the state of the relevant parts of your brain when you look at that particular card. Then we could have the machine read your brain later and return which card you are looking at. This would work as long as the particular "brain state to information mapping" is stable over time.
You might say that is kind of cheating. This machine can only decode what you're thinking if it's been trained on the exact same information that it will be decoding. Luckily, this difficulty can be overcome by techniques from machine learning. Recent work has shown that it is possible to categorize over a billion novel images from V1 activity measured using fMRI (see Gallant et al 2008).
So let's pretend now that we have a universal decoder that works for converting any brain-state to its corresponding semantic state. Let us also pretend that we have dealt with the issues of context that may cause the same brain state to correspond to different demantic states in different situations.
Now if we still want to transfer information from brain to brain we encounter the encoding problem. It is not clear if this is more difficult than decoding or not. The same machine learning procedure would work. We need to decode the mapping from brain states to semantic states for both participants and convert the information appropriately.
The final step is to bring about the activity pattern corresponding to the new knowledge in the recipient subject. This is a more difficult problem and it is one that is not well understood. The most similar device that I am aware of is the cochlear implant. Unlike a regular hearing aid which merely amplifies sounds. The cochlear implant directly stimulates the auditory nerve in a way that the brain can interpret as sound.
But what does this mean the brain "interprets the pattern of stimulation as a sound". The cochlear implant always does the same thing, there is a direct mapping from the frequency of its vibration due to sounds and the stimulation that it provides the auditory nerve. Inititially this is not sufficient for hearing. There is a learning period in which the brain comes to interpret this new pattern of stimulation as hearing sounds. That is, the brain learns the mapping from auditory nerve stimulation to sounds in the world. The brain is actually on its own performing a similar decoding procedure to what our decoding machine would have done to beat us in that card game.
It is possible that we could use a similar device to bring about any brain state in another person. IF (and this is a huge if) this is indeed doable it would then be in the realm of speculative possibility that we could link such a device to the appropriate decoding and encoding machines and only THEN could we transfer information "directly" from brain to brain.
