34 Tony Veale and His Metaphor- and Story-Generating Programs
Computers that can create are not the kind that can turn us into slaves.
—Tony Veale126
Tony Veale has been hooked on computers ever since seeing a rerelease of Stanley Kubrick’s 2001: A Space Odyssey as a teenager in Ireland. Thereafter, he was obsessed with the computer HAL 9000. “AI was part of my life before I became a computer scientist,” he recalls.127 At the time, he didn’t fully grasp every nuance of the film, but he homed in on HAL as a thinking, feeling, and above all speaking presence who could play chess and plot the murder of astronauts. Thinking about this brought home to him the importance of the way people use language and the way we judge others by their use of tropes, “their ability to turn a phrase.” When he became a computer scientist, Veale chose to focus on natural-language processing, the way computers deal with human language, such as metaphors and similes, a career choice he encapsulates in the words “the first route in defines you.”
Veale brims with ideas and enthusiasm about computer science. From the start, he took an optimistic view of computers, he says, in contrast to the dystopian outlook of commentators such as Stephen Hawking, Elon Musk, and Nick Bostrom. Veale sees AI “as a place to explore passionate ideas like literature and creativity; it’s a way of doing creative work by another route because computer science touches on everything.”128
He points out that much human language revolves around metaphors. We “catch an aeroplane” and “catch a cold.” These, he says, are “conventional metaphors.” He is more interested in “conceptual metaphors,” complex cognitive structures that “provide conceptual mappings that shape our habitual thinking about such familiar ideas as Life, Love, Politics and War.”129 These are Shakespearean in scope.
As part of his exploration of metaphors, Veale has developed software that generates original metaphors. One such program is Metaphor Magnet.130 You put in a word—love, death, war, whatever—and a stream of similes, metaphors, and word associations pops out, too vast to begin to quote. Love, for example, spawns concepts from shining saint to shimmering soap bubble and millions more.
Another is Thesaurus Rex (illustrated with a picture of a baby tyrannosaurus rex bursting out of its egg).131 This provides synonyms and word associations for single words, while two words linked with and generate a variety of single words that are basically definitions.
Metaphor Magnet also has a Twitter account with which it tweets a metaphor every hour or so under the handle @MetaphorMagnet. Veale believes that Twitter offers more than just a means for social interaction. “In many ways it constitutes a whole new genre of text.”132 To create metaphorical and ironic tweets, @MetaphorMagnet taps into huge databases constructed by Veale and his coworkers, including Thesaurus Rex and Metaphor Eyes, another way of spinning ideas out of the relationships between concepts, such as scientists as artists, or writers as readers.133
To build up these databases, he uses n-grams, a huge database of short word chains that create conceptual metaphors. Take the four-word-grams, each made up of four words, like democracy is a cornerstone and democracy is a failure. The Metaphor Magnet website homes in on their properties and juxtaposes them “via resonant contrasts and norm contraventions.”134 It then informs the bot that the stereotype cornerstone should be qualified as important and the stereotype failure as worthless. The result is a tweet embodying two contrasting views on democracy:
To some voters, democracy is an important cornerstone
To others, it is a worthless failure
#Democracy = #Cornerstone #Democracy = #Failure
Veale contrasts the complex construction of this rational and figurative conceptual metaphor with the slapdash, mash-up tweets generated by the Twitterbot @Metaphor-a-Minute. This has a framework X is a Y, which it fills in every two minutes using random word choices, creating entries such as: “a macula is a result: blockheaded and refuse.” “Rather than random metaphors, they are random metaphor-shaped texts,” Veale says.135
He points out what he calls the placebo effect, whereby readers try to attribute meaning to a metaphor that is actually complete nonsense.136 “The meaning,” he says, “is created by the reader. The machine generates a provocation for the reader to create meaning.”137 The author (who in this case happens to be the computer) “is hacking your brain, creating the conditions for ideas to flourish. Twitter is a perfect medium because it values conciseness.”
Veale arranged a competition between the two Twitterbots, with human judges rating them on comprehensibility, tweet novelty, and retweetability. The judges agreed that @MetaphorMagnet outperformed @Metaphor-a-Minute on comprehensibility and retweetability, but was less strong on tweet novelty. This is no doubt due to @Metaphor-a-Minute’s bizarre juxtapositions of words. The assumption that groups of words are meaningful “can lead us to perceive (and enjoy) a creative meaning when none was ever intended,” Veale writes.138
I ask Veale whether Metaphor Magnet may be so loaded with enormous volumes of words and templates that it is bound to generate provocative and creative metaphors. “This is what humans do,” he replies. “We all use a toolbox of tricks. … All modern speakers instantiate tropes because they are the way to use language to persuade, to inspire, to surprise, to provoke. And that’s what computers are doing. … What computers are saying is that even the cleverest of humans are just bags of tricks, and they vary their tricks.”139
Veale’s work does not stop at metaphor. He has also produced a story-generating system, akin to those of Pablo Gervás and Rafael Pérez y Pérez, among others. “The challenge—and the opportunity—for a story generator,” he says, “is to do more than fill slots with matching fillers.”140 He looks for “a conceptual blend of characterization and plot,” which includes everything we already know about the character in question.141
Veale calls his story-generation system the Scéalextric Simulator and describes it as “a simulation-based approach to the generation of episodic stories in which stories are generated, evaluated and frequently discarded in a rapid, coarse-grained cycle of engagement and reflection.”142 The name blends together scéal, the Irish word for story, and Scalextric, a children’s racing car game in which segments of track are slotted together end to end. The simulator is online, ready to be played with.143 It offers the player—or the writer in search of inspiration—a choice of three sorts of character: from the “NOC list,” first names only, or one NOC character and the rest first names.
The NOC list is a kind of insider joke. It stands for “nonofficial cover” and is espionage lingo for agents working covertly for organizations but with no connection to the government for which they work, and who therefore have no diplomatic immunity. The characters on the NOC list are familiar and can be real or fictional—Jesse Jackson, Armand Richelieu, Superman, Lex Luthor, and so on. The first names are just names—Violetta, Brady, Gwendolyn, whatever.
Having chosen what type of character you want, next you tailor the action by ticking or unticking any of the three following boxes:
- □ Causal references to previous actions
- □ Strict linking of arcs (last arc action to next)
- □ Information transfer between characters
The simulator then generates a story that fulfils your chosen criteria and is different every time. Of the story generators we’ve discussed thus far, this one generates probably the most sophisticated plots.
Scéalextric has over three thousand prefabricated plot sections. It may choose characters who are well suited but impossibly matched, such as Steve Jobs and Leonardo Da Vinci, Mahatma Gandhi and Obiwan Kenobi, or Cicero and Barack Obama, taking vivid details of their lives and personalities from the NOC database. Scéalextric pairs them on a metaphorical basis, homing in on their similarities, even though they may exist in different domains, genres, or periods.
One story Veale’s system created is a confrontation between (a) Richard Nixon and (b) Frank Underwood from the House of Cards series, until recently played by Kevin Spacey. The first of Scéalextric’s prefabricated segments read:
—campaign_against—but → are humiliated_by144
The simulator fleshes this out, creating the following prologue and first two lines:
0. Richard Nixon and Frank Underwood were driven by very different political agendas.
1. So at first, Frank campaigned vigorously against Richard.
2. But Richard humiliated Frank by calling the sociopathic and ruthless Frank the Keyser Söze of wielding political power. 145
The story goes on for ten more scenes.
“The system jumped outside itself,” Veale recalls excitedly of these first two lines.146 It invented—on the fly—an exchange of insults that went outside the storyline, bringing in Keyser Söze, who was also played by Spacey in the film The Usual Suspects. “I don’t want to suggest that that shows the machine to be more clever than it is. Partly it’s an accident,” Veale says. But, he continues, “it has a lot of knowledge to play with to acquire this postmodern irony in the way it manipulates its characters. … A storytelling system has to be generative on multiple levels.”
I ask Veale whether his storytelling system has been criticized for being loaded. He replies, “It’s like a Chinese buffet, where one customer decides to make his own soup with some of the ingredients. … The more ingredients you have in the program, the wider the scope of combinatorial creativity; that’s what we mean by … fully loaded programs that exist in a knowledge context.”147 In other words, with such a vast amount of information, the possibility exists for the system to pull out unexpected similarities.
His “beef,” Veale says, with neural networks is that they are “somewhat accidental and don’t use knowledge deliberately or knowingly.”148 Perhaps it was pure accident that his symbolic system, programmed with enormous amounts of information about people and things together with ways to manipulate all this data, happened to compare Frank Underwood with Keyser Söze, both of whom were played by Kevin Spacey. But Veale feels comfortable with the way this happened. He is happy to assert that the system uses its knowledge deliberately and knowingly.
Veale believes that, at present, fully loaded symbolic systems are the best way to enhance and understand human creativity, spurring “the ability to generate and to own the results—that is, to appreciate and to critique and say, ‘here it is, hope you like it.’”149
Currently Veale is installing his Scéalextric system in a Nao robot. At the moment, these small, cute robots are just expensive toys. To use a robot as a storyteller will provide Veale with an embodied system that can tell stories accompanied by gestures. The human-like appearance and gestures of Gil Weinberg’s music-making robots, Shimi and Shimon, make people look beyond the simplicity of their forms into something primal. This will happen with Veale’s Nao robots too as audiences become more and more enraptured by the unfolding story.
Veale agrees that computers show glimmers of creativity and gives AlphaGo as an example. AlphaGo has certainly done things that can be assessed as novel (the thirty-seventh move in its second game of Go had not been made before) and useful (it achieved the goal of advancing its game), both of which are often taken as criteria of creativity. Can AlphaGo therefore be called creative?
Says Veale, it neither questions its knowledge as to whether a move is conventional or not, nor has any conception of having made a spectacular move. “The machine lacks a social dimension.”150 He believes it will be able to attain this by being inserted into a robot, by embodiment. In this way, it will be able to be “out there”: bombarded by stimuli, entertained by culture, maybe even falling in love, while generating interesting metaphors and creating literature, art, and music. And it will be able to do this without being organic.
Veale concludes optimistically, “The question of whether computers are truly creative will fade away as people value their results.”151