This presentation concerns a distinctly human
phenomenon: CREATIVITY.
This is a quality that is widely held to separate
us from most of the animals and every machine.
But I argue here that creative machines – made
in our image but not made to BE us – are the
key to magnifying human creativity and to
sustaining vibrant communities of human &
machine creators.
CREATIVITY is powerful but fragile. It is
lightning that strikes at the whim of a fickle
muse, and may only strike memorably just once
or twice in a person’s career.
So psychologists and creative practitioners have
long sought to bottle creativity, to put it on a
schedule so we may invoke it on demand.
Some approaches are a matter of faith, but may
have as much success at ending a creative dry
spell as rain dances have of bringing the rain!
Brainstorming is perhaps the most well-known
process for coaxing the creative muse into action.
Such processes – and there are many variations
on the brainstorming approach – encourage
people to riff on each other’s ideas while
reserving critical judgment.
But the most productive techniques to foster
creativity are those that drill deeper, into the
structure of ideas themselves.
Most new ideas are built from fragments of older
ideas that have already proven their value.
This is called combinatorial creativity, and it has
been formalized in a number of popular models.
Fritz Zwicky’s “Morphological Analysis” and
Genady Altshuler’s TRIZ each turn creativity
into something one does with the cross-indexed
wisdom of tables and charts.
Big companies embrace these methods whole-
heartedly.
But it is not just big companies or Hollywood
moguls that embrace systematic approaches.
As art became more abstract in the 20th century
it also became less mysterious, as artists from
Duchamp to Burroughs invented mechanical
methods to put their muses on a leash.
Artists turned to simple rules, tricks and random
methods to generate novel forms and ideas.
They wanted to be productive without being
repetitive, or without succumbing to cliché.
The CUT-UP Method of William Burroughs
allowed a writer to randomly chop up an existing
text to re-combine its parts into something new.
David Bowie is perhaps the most famous modern
exponent of Cut-Ups. Bowie uses the technique to
generate interesting juxtapositions for his lyrics.
Bowie even uses a special software tool he calls
The Verbasizer to generate random possibilities.
Of course, this is just a tool – Bowie is still the
master and sole creator of his own lyrics.
Picasso famously said “Computers are useless:
they can only give you answers”.
To go from the Verbasizer to something that is
truly creative in its own right, we need software
that can do more than retrieve existing answers
from its memory stores.
Our software must be able to invent its own
questions and to imagine new answers for them.
Good art is a question, not an answer.
The field we call Computational Creativity is
the next evolutionary step in AI and modern art.
Though William Burroughs – like Picasso – did
not recognize the creative potential of machines,
our field aims to pick up where they left off.
Computational Creativity adds intelligent self-
critiquing and filtering to the Cut-Up method
and its ilk, to turn the formal systematization of
creativity into the automation of creativity.
This may seem like hubris to some and horror to
others. Consider the cake on the previous slide.
This is a real cake, produced by a NY bakery that
has embraced the internet. Customers send the
bakery an image and/or text by email, and the
bakery prints them onto the cake with food dyes.
Here the customer used Microsoft Outlook to
mail their commission, but Outlook adds extra
HTML tags for fancy text effects. The bakery does
not use Outlook, so the tags were printed too!
Like the Sorcerer’s Apprentice, or the Golem of
old Prague, imagine the chaos caused by faulty
software that is allowed to create unchecked.
Except … well, this cake is the result of human
error, not machine error. A human baker cut-
and-pasted the customer’s email into the printer
app, and sent the resulting cake to the customer!
By training our software to recognize convention
and to flag any possible exceptions, a creative
machine can be more engaged in its task than
any bored or over-worked human.
But we do not want to over-train our software. It
must do more than critique. It must originate.
There is an important difference between a
clever and talented forger and an original artist.
So our creative software must do more than
appreciate the surface. It must also embody the
necessary mechanisms and internal processes to
go from an original idea to a final polished form.
If we train our machines to be expert players of
The Imitation Game, we should not be surprised
if all they can produce is imitations and pastiche.
We must understand the mechanisms of human
creativity before we embody them in a machine.
Yet our attempts at mechanical embodiment can
also be seen as experiments in human creativity,
or an engineering approach to psychology that
reveals the algorithmic basis of human ideation.
The media’s fixation on The Turing Test is thus
wrong-headed and unhelpful.
We are not in the business of building fake
humans, and creativity is not a test. Computers
can be creative and useful in ways that are
recognizably non-human, but just as useful for it.
The Turing Test is a race to the bottom that does
not advance AI. To make computers creative does
not mean making them pass for human.
How might a computer be usefully creative in a
recognizably non-human fashion?
It might compose a beautiful melody that no ten-
fingered pianist could every hope to play unaided.
Or paint a complex mathematical picture.
Or a creative computer could ask the questions
that most humans are too polite to ask, or consider
possibilities that humans do not put on the table.
Computers may talk about us as if they were
anthropologists from another planet, and open
doors to new ways of thinking about the world.
Computers do not need to mimic us, or try to fool
us, to engage meaningfully with us.
There is a value to being an outsider looking in at
the human experience.
A creative computer may be able to formulate
thought-provoking metaphors or analogies or
report apparent ironies in the way we behave.
As with the words of children, we might be
surprised by what our intellectual children have
to say about us.
Our creative computers are being designed to
complement us, not to mimic us or replace us.
Would The Beatles have been better with two
Lennons, or two McCartneys? The best partner-
ships have a measure of overlap and a measure of
contrast, or even conflict.
Our creative computers will not be mere tools, or
yes-men, or MINI-MEs. Neither will they be
autocrats. They will be co-creators that work
with us but argue for their own points of view.
So our creative machines will be social machines.
The myth of the lone creator has long impeded
our understanding of human creativity. We must
be careful not to succumb to the same myths
when working toward machine creativity.
We want our creative software to work well in
groups, of other humans and other machines.
Indeed, it is the friction of working together in a
mixed group of individuals that often sparks the
development of creative new ideas.
Living side-by-side with other people with their
own goals in the crush of the real world is a rich
source of unexpected friction and inspiration.
This is this friction that Burroughs and his
fellow modernists sought to reproduce with
random stimuli.
So wan we expose our creative machines to this
inspiring friction directly, by moving them from
our desktops into the virtual online world?
Cities have always been vital hubs of human
creativity. We obtain inspiration from each
other’s problems and each other’s solutions.
The Beat poets met in Bohemian cafes and recited
their experimental poems to a willing audience.
So where else would our Bot poets meet but on
Twitter, where a willing listener quickly
becomes an eager follower and an active
promoter.
Social spaces like are the virtual cities for new art
in the 21st century.
A wonderful bot named
@Pentametron uses the Cut-Up Technique to find an
accidental poetry in the most banal of tweets.
Indeed, more and more of the creative impulse on
Twitter is coming from autonomous machines.
This wonderful little bot called @Pentametron is
not very complicated but it achieves good results
with the Cut-Up method.
By splicing together random tweets from random
users on the basis of simple formal constraints,
@Pentametron allows an accidental but vibrant
poetry to emerge out of the digital cross-talk.
This has a democratizing effect on all creativity
and artistic experimentation on Twitter.
You do not need to be rich or famous, or be part
of an artistic elite to put your ideas out there and
attract followers. There is a level playing field
for humans and computers.
Anyone can build a Twitterbot and do subversive
experimentation in conceptual art. And anyone
can be an influential patron or critic, by choosing
what to retweet and favorite to others.
@MetaphorMagnet is just one example of a new
breed of creative Twitterbots that combine
knowledge of human behaviour with insights
from psychology about human creativity.
It uses this knowledge to formulate novel
metaphors, analogies and ironic observations
about the human condition.
@MetaphorMagnet is an effective test-bed for
theories about human analogy-making.
@MetaphorMirror is a related Twitterbot that
produces its metaphors and analogies in response
to breaking news stories.
More and more of our news on social media will
be written by computers with direct access to
the necessary financial, sports or seismic data.
@MetaphorMirror is an experiment in shaping
our view of the news via metaphor & analogy,
traditionally the role of a good human analyst.
AI has always raised fears about the unintended
consequences of our powerful new technologies.
Recently some very prominent voices in science
and in industry have joined the chorus of doom-
sayers. Can we put the genie back in the bottle?
While these concerns are worthy of debate, they
significantly over-estimate progress in AI and
significantly under-estimate human ingenuity
and the complexity of the human mind.
Our creative machines will not be the end of us,
but they are very likely to survive us.
Because their algorithmic workings will mirror
us at some important schematic level, they will
sustain the best of us after we humans are gone.