Artificially intelligent systems are capable of many
human-like behaviors. At least on a surface-level, they can think, learn,
speak, read, and simulate emotions. The convergence of human and machine
behavior has led to the well-worn ideas of the Singularity and possible machine
superintelligence.
Those uncomfortable sharing their headspace with a machine
purport that computers will likely never be capable of one of our most human
traits: creativity. Creativity seems random, messy, and subjective, three attributes
not easily grasped by computers. Creative AI research, while interesting and
promising, has provided more questions than answers; namely, "What is
creativity, and who's capable of it?"
Creativity is "the ability to transcend traditional ideas,
rules, patterns, relationships, or the like, and to create meaningful new
ideas, forms, methods, [and] interpretations," according to Random House
Dictionary. Given that computers essentially analyze fairly standard rules,
patterns, and relationships, a truly creative one seems unlikely. Creative
works are often described as "novel" and somehow valuable or useful; without
direct human input, this is a monumental challenge for AI.
These hurdles haven't stopped researchers from trying,
though. Lior Shamir, a computer science professor at Lawrence Technological
University, developed algorithms that identified
similarities between works of Jackson Pollock and Van Gogh and correctly
ordered all thirteen Beatles LPs chronologically based on audio and visual
samples. Because computers aren't adept at handling discrete pixel and
frequency data, Shamir converted each visual or audio sample into thousands of
numerical values that were fed into a pattern recognition system. Shamir's
music understanding programs are based on an earlier project that analyzed
15,000 whale songs to identify that whales communicate using different dialects
depending on their geographical origins.
While analyzing creative works might seem much simpler than
actually producing them, automated painting and composing software programs
have been used for decades. AARON,
a program created by Harold Cohen in the mid-'70s,
is hand-coded to produce artworks (like the one below) in a specific style. Even Cohen is quick to
point out that AARON isn't creative, though: it simply follows procedures
outlined by its programmer, who is the true artist. Returning to the definition
of creativity, AARON follows the rules rather than transcending them. More complex painting
robots are becoming common, although these still seem to be merely going
through the motions.
If true artistic creativity seems outside the reach of AI,
what about the experimental kind? Assuming Pasteur's assertion that "in the
experimental fields, chance favors the prepared mind," a data-loaded processor
would be much more likely to stumble upon a scientific breakthrough than a
messy human brain.
AI research seems peculiarly unconcerned with the philosophy
of the work, with practical solutions trumping abstract ideas: an application
that works will naturally form the basis for an abstract theory. Many in the
field point to aviation as a good example: it took building a working flying
machine to answer the century-long debate asking "Can we, and should we, fly?"
So is there a useful application for creative AI? Music
recommendation engines used by iTunes and streaming music services would
certainly benefit from creativity breakthroughs. Although in a subjective
sense, I wouldn't find an hour of identical-sounding music very appealing.
Image credits: The Academic Minute | Stanford University
|
Comments rated to be "almost" Good Answers: