There is no reason to believe that 1 bit or 1 synaptic connection corresponds to a single pattern of a memory (such as a word or utterance). Not even sure where that idea originated beyond some article in Scientific comparing individual neurons or even isolated synaptic connections to memories and bits. More evidence points to the specific path a stimulus takes in the lower neocortical layers becoming generalized in the higher layers, and it's these high layer similarities that give us the impression of a shared pattern. Two stimuli can be completely different in the lower layers, but look the same in the higher layers that are used by the brain to form the predictive model, thus 1 qualia corresponds to possibly hundreds, even thousands of potential stimuli (synaptic connections in the lower layers) that can result in that higher layer pattern. In this way, the amount of qualia that can be stored are quite low, order of magnitudes lower than the total number of synaptic connections, while the number of memories, the combinations of those qualia, can be quite high.
Think closer to I-II-III-IV in music theory, same chord progression but the instrument is different, tones can be different, etc. The instruments are the stimuli, the tones they produce are the qualia, and the generalized chord progression is the memory (note that this isn't an example, just a way to envision it - humans can remember actual musical tones and instruments just fine) On Sat, Sep 9, 2023, 4:22 PM Matt Mahoney <[email protected]> wrote: > There are several thousand languages for programming people. The most > common is English. Since the target is a neural network and not a register > based von Neumann machine, the primary difference is that the language is > learned by example rather than described by a formal grammar and semantics. > The learning is hierarchical, with semantics learned before grammar. Human > language is not compiled or interpreted. Rather, knowledge is stored in the > weighted connections between neurons representing phonemes, letters, or > words, or some linear combination of these. > > At the cognitive level, memory capacity is fixed at about 10^9 bits > representing associations between 10^4.5 words and a similar number of > nonverbal features such as faces and objects. For reasons not fully > understood, this is much smaller than the 10^14 bits suggested by the > Hopfield limit of 0.15 bits per connection for the brain's 6 x 10^14 > synapses. > > Memory management in neural networks consists of updating connections to > produce the desired output. This has the effect of gradually replacing old, > unused memories with new. Humans can learn at a rate of 5 to 10 bits per > second at an information rate of 1 bit per written character or 0.5 bits > per speech phoneme. About 50% is forgotten after a day and 90% after a > year. The learning rate drops roughly inversely with age. > > Because of this rate limit, effectively programming humans requires > holding their attention for long periods. It takes about 8 hours per day > for 2% of your age to change your political views. Education and religious > indoctrination take even longer, perhaps 5% to 25%. > > The classical methods of holding your attention using emotions like humor, > anger, and fear have been known since the time of Shakespeare. AI can do > this effectively by learning your interests through social media > interactions, and soon, by monitoring your facial expressions and eye > movements. Soon, most of your experiences will be AI generated rather than > just AI filtered. Censorship will be invisible. > > We have already shown that political channels can hold your attention for > hours each day just by telling you all the bad things that the other party > is doing and there is nothing you can do about it. This makes you angry, > afraid, and addicted to their propaganda. It doesn't matter left or right. > It is the same technique. > > But this is only the beginning of what is possible. > > *Artificial General Intelligence List <https://agi.topicbox.com/latest>* > / AGI / see discussions <https://agi.topicbox.com/groups/agi> + > participants <https://agi.topicbox.com/groups/agi/members> + > delivery options <https://agi.topicbox.com/groups/agi/subscription> > Permalink > <https://agi.topicbox.com/groups/agi/Tc1bcda5fdb4147f4-M1f50130b13d76c656a5d6cdd> > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/Tc1bcda5fdb4147f4-Mf95ee58f5e4b9ef77d8c110a Delivery options: https://agi.topicbox.com/groups/agi/subscription
