How Many Bytes in Human Memory?
\nby Ralph C. Merkle
\n(appeared in Foresight Update No. 4, 1988)
\n(merkle.pa@xerox.com)<\/p>\n
Today it is commonplace to compare the human brain to a
\ncomputer, and the human mind to a program running on that
\ncomputer. Once seen as just a poetic metaphore, this viewpoint
\nis now supported by most philosophers of human consciousness and
\nmost researchers in artificial intelligence. If we take this view
\nliterally, then just as we can ask how many megabytes of RAM a PC
\nhas we should be able to ask how many megabytes (or gigabytes, or
\nterabytes, or whatever) of memory the human brain has.<\/p>\n
Several approximations to this number have already appeared in the
\nliterature based on ‘hardware’ considerations (though in the case
\nof the human brain perhaps the term ‘wetware’ is more
\nappropriate). One estimate of 10**20 bits is actually an early
\nestimate (by Von Neumann in ‘The Computer and the Brain’) of all
\nthe neural impulses conducted by the brain during a lifetime. This
\nnumber is almost certainly larger than the true answer. Another
\nmethod is to estimate the total number of synapses, and then
\npresume that each synapse can hold a few bits. Estimates of the
\nnumber of synapses have been made in the range from 10**13 to 10**15
\n— with corresponding estimates of memory capacity.<\/p>\n
A fundamental problem with these approaches is that they rely on
\nrather poor estimates of the raw hardware in the system. The
\nbrain is highly redundant and not well understood: the mere fact
\nthat a great mass of synapses exists does not imply that they are
\nin fact contributing to the memory capacity. This makes the work
\nof Thomas K. Landauer very interesting for he has entirely avoided
\nthis hardware guessing game by measuring the actual functional
\ncapacity of human memory directly (‘How Much Do People
\nRemember? Some Estimates of the Quantity of Learned
\nInformation in Long-term Memory’ in Cognitive Science 10, 477-
\n493, 1986).<\/p>\n
Landauer works at Bell Communications Research — closely
\naffiliated with Bell Labs where the modern study of information
\ntheory was begun by C. E. Shannon to analyze the information
\ncarrying capacity of telephone lines (a subject of great interest to
\na telephone company). Landauer naturally used these tools by
\nviewing human memory as a novel ‘telephone line’ that carries
\ninformation from the past to the future. The capacity of this \ufffd\ufffd3\/\ufffd\ufffd\ufffd\ufffd3\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd \ufffd
\n\u598fhone line’ can be determined by measuring the information
\nthat goes in and the information that comes out — the great power
\nof modern information theory can be applied.<\/p>\n
Landauer reviewed and quantitatively analyzed experiments by
\nhimself and others in which people were asked to read text; look at
\npictures; hear words, short passages of music, sentences and
\nnonsense syllables. After delays ranging from minutes to days the
\nsubjects were then tested to determine how much they had
\nretained. The tests were quite sensitive (they did not merely ask
\n‘What do you remember?’) often using true\/false or multiple choice
\nquestions, in which even a vague memory of the material would
\nallow selection of the correct choice. Often, the differential
\nabilities of a group that had been exposed to the material and
\nanother group that had not been exposed to the material were used.
\nThe difference in the scores between the two groups was used to
\nestimate the amount actually remembered (to control for the
\nnumber of correct answers an intelligent human could guess
\nwithout ever having seen the material). Because experiments by
\nmany different experimenters were summarized and analyzed, the
\nresults of the analysis are fairly robust; they are insensitive to
\nfine details or specific conditions of one or another experiment.
\nFinally, the amount remembered was divided by the time alloted to
\nmemorization to determine the number of bits remembered per
\nsecond.<\/p>\n
The remarkable result of this work was that human beings
\nremembered very nearly two bits per second under ALL the
\nexperimental conditions. Visual, verbal, musical, or whatever —
\ntwo bits per second. Continued over a lifetime, this rate of
\nmemorization would produce somewhat over 10**9 bits, or a few
\nhundred megabytes.<\/p>\n
While this estimate is probably only accurate to within an order of
\nmagnitude, Landauer says ‘We need answers at this level of
\naccuracy to think about such questions as: What sort of storage
\nand retrieval capacities will computers need to mimic human
\nperformance? What sort of physical unit should we expect to
\nconstitute the elements of information storage in the brain:
\nmolecular parts, synaptic junctions, whole cells, or cell-circuits?
\nWhat kinds of coding and storage methods are reasonable to
\npostulate for the neural support of human capabilities? In
\nmodeling or mimicking human intelligence, what size of memory
\nand what efficiencies of use should we imagine we are copying?
\nHow much would a robot need to know to match a person?’<\/p>\n
What is interesting about Landauer’s estimate is its small size.
\nPerhaps more interesting is the trend — from Von Neumann’s early
\nand very high estimate, to the high estimates based on rough
\nsynapse counts, to a better supported and more modest estimate
\nbased on information theoretic considerations. While Landauer
\ndoesn’t measure everything (he did not measure, for example, the
\nbit rate in learning to ride a bicycle nor does his estimate even
\nconsider the size of ‘working memory’) his estimate of memory
\ncapacity suggests that the capabilities of the human brain are \ufffd\ufffd3f\ufffd\ufffd\ufffd\ufffd3\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd\ufffd \ufffd
\n\u5971pproachable than we had thought. While this might come as
\na blow to our egos, it suggests that we could build a device with
\nthe skills and abilities of a human being with little more hardware
\nthan we now have — if only we knew the correct way to organize
\nthat hardware.<\/p>\n
X-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-X
\n Another file downloaded from: The NIRVANAnet(tm) Seven<\/p>\n
& the Temple of the Screaming Electron Taipan Enigma 510\/935-5845
\n Burn This Flag Zardoz 408\/363-9766
\n realitycheck Poindexter Fortran 510\/527-1662
\n Lies Unlimited Mick Freen 801\/278-2699
\n The New Dork Sublime Biffnix 415\/864-DORK
\n The Shrine Rif Raf 206\/794-6674
\n Planet Mirth Simon Jester 510\/786-6560<\/p>\n
“Raw Data for Raw Nerves”
\nX-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-X<\/p>\n
