10.2) Storage and Associative Memory :

10.2) Storage and Associative Memory :

DNA might also be used to mirror, and even improve upon, the associative capabilities of the human brain. In [4] Baum proposed a method for making a large content addressable memory using DNA. A truly content addressable memory occurs when a data entry can be directly retrieved from storage by entering an input that most closely resembles it over other entries in memory. This input may be very incomplete, with a number of wildcards, and in an associative memory might even contain bits that do not actually occur within the closest match. This contrasts with a conventional computer memory, where the specific address of a word must be known to retrieve it. Rather, the use of this technique would replicate what is thought by many to be a key factor in human intelligence.

Baum's models for a DNA associative memory are quite simple, and build from the techniques used in other areas of DNA computation. Storing a word could be done by assigning a specific DNA subsequence to each component value pair and building a fixed length word from these subsequences. To then retrieve the word closest to the input, one would introduce marked complementary subsequences into the storage medium and chose the molecule that has the most matches to this input. This technique could be further refined to more closely approximate the brain by appending words to only store attributes that an object has, rather than wasting space using '0's to represent attributes that an object does not have.
Baum has further speculated that a memory could be constructed where only portions of the data are content-addressable and associative, with other information on an object compactly stored in addresses relative to the associative portion of the entry. To save on operating costs and reduce error frequency, this portion of the memory could be kept in double-stranded form.

Considering the brain's limit of about 10^15 synapses and Feynman's low-end estimate that the brain can distinguish about 10^6 concepts, such a DNA based associative memory could have certain advantages over the brain. Without accounting for redundant molecules, Baum estimates that a large bath tub of DNA, about 50g in 1000L, could hold over 10^20 words. In attempting to come up with practical uses for this memory scheme one will have to weigh the massive size and ability to retrieve in an associative manner against the slow retrieval times necessitated by current biomolecular engineering techniques. And, although Baum's simplistic approach has accounted for some error rates, his initial paper remains quite speculative.