“Incompressible Patterns: CRISPR vs Dennett” (3/30/17)

Katie Creel

Abstract: Dennett’s classic paper defines “Real Patterns” as present in data if
“there is a description of the data that is more efficient than the
bit map, whether or not anyone can concoct it.” (Dennett 1991, 34)
However, compressibility is not the right criterion for pattern
realism. A better pattern ontology is one based on informational
relationships between the pattern and the perceiver of the pattern,
whether human, biological, or machine.

A simple compression algorithm such as Huffman coding should be
perfect for Dennett’s purposes. It can compress text into an efficient
lossless binary tree in which letters are assigned unique codes based
on their frequency. Instead, Huffman coding illustrates the problem
with compression as a metric: lack of generalizability. If the
algorithm were only used once, it could compress a novel into one
character: “W” for all of War and Peace. But this would be no
informational savings. What makes compression work is that the “cost”
of the compression algorithm is amortized over many uses.

Further, any discrete chunk of randomness can be recognized as a
pattern if it has the right informational relationship with its
recognizer. Such recognition relationships between random sequences
and detectors occur in genetic material. New tools for genetic
manipulation such as CRISPR use a recognition relationship with
sequences of base pairs to snip and replace precise segments of DNA.
Using case studies, I suggest that we should think of patterns as
representing the informational relationship between pattern and
recognizer.