117 SMITH JOURNALIN 1949, A BRITISH BRAIN SURGEON
NAMED GEOFFREY JEFFERSON
MADE A ROUSING SPEECH AT THE
LONDON SURGICAL SOCIETY.
..........................................
Taking the stage, Je erson railed against
the idea that the recently invented “digital
programmable computer” would ever be
capable of human-like thought. Yes, he
conceded, these new machines could perform
impressive mathematical calculations. But
they would never be able to “write a sonnet...
because of thoughts and emotions felt.”
Faced with a new and powerful technology,
Jefferson used poetry to draw a boundary
around humanness. And in defining poetry
as the yardstick of our humanity, he was in
good company. Philosopher John Stuart Mill
once said that to write poetry was to “paint
the human soul truly”, and the art form’s
elevated position in our culture has hardly
dipped since. To this day, the very idea of a
computer writing verse triggers reactions
similar to Jefferson’s. Sure, computers might
be able to beat us at chess and drive trucks
by themselves, but they can’t do that.
Well actually, yes. They can.
In 2014, I made ‘Bot or Not’, a website that
functioned like a Turing test for poetry.
Visitors to this website were presented with
a poem and asked to guess whether it was
generated by a computer or written by a
human. Tens of thousands of people took
the test, and the results were conclusive:
computers, it turned out, could indeed
write poems indistinguishable from
ones penned by humans.For some, the results were a troubling
sign about the future. Not only was AI
coming for our jobs, it was also encroaching
on the core of our humanity. (The result of
my stupid website even led one journalist
to write a story titled, ‘We’re Ready for Our
Robot Poet Overlords’.) In reality, getting
a computer to write a poem was a lot less
impressive than people thought. All it took
was a program that had a dictionary of
words and a bunch of basic rules that
dictated how they could be combined
and recombined — a database and an
algorithm, in other words.But using algorithmic processes to generate
poetry is nothing new. In medieval Spain,
Jewish Kabbalists would combine and permute
the letters of the Hebrew alphabet according
to a strict set of rules, hoping to discover some
secret attributes of the divine. In 1846, Edgar
Allan Poe claimed that his most famous poem,
‘The Raven’, was written according to a
formula that he executed with “the precision
and rigid consequences of a mathematicalproblem”, as if he were some sort of
poetry-automaton. In the 20th century,
avant-garde poets like Jackson Mac Low
and Bernadette Mayer came up with
mechanistic rules to create poems, such
as replacing every noun in a sentence
with another seven below it in the
dictionary, to generate a new text.While these poets didn’t use machines to
write poetry, they were, in essence, creating
computational art: using algorithms and
databases of words to generate poetic texts.
For these poets, following generative rules
did not make the practice any less creative
or human. In fact, by giving up some
creative agency, they were able to find new
combinations of words and ideas, and even
chart unexplored parts of language.Thinking about AI in light of this history
might help temper some of the Elon Musk-
esque fantasies and fears around automation
that are so prevalent right now. Looked at in
the right way, a poetry-generating machine
can offer a new type of collaborator for human
poets. As the programmer and poet Allison
Parrish once put it, using computers to
generate poetry is like sending a robot probe
into space. The probe can travel into realms
of language that humans find inhospitable
and report back on what it finds. But the
task of finding meaning in the telemetry
remains a deeply human process. •bard in the machine
DO YOU HAVE TO BE A LIVING, BREATHING, QUILL-WAVING
HUMAN TO WRITE POETRY? ARTIFICIAL INTELLIGENCE
SAYS NO – AND IT’S GOT THE STANZAS TO PROVE IT.Writer Oscar Schwartz Illustrator Timothy Rodgers