Artificial intelligence systems have produced everything from Friends screenplays to full-length musicals (the latter with a little human intervention). And now, an A.I. system has taken on politics. Created by a team of researchers at the University of Massachusetts, Amherst, the system uses a range of methods, including n-grams and recurrent neural networks to generate political speeches [PDF].
The researchers claim that the system can not only generate coherent speeches, but that it can be made to argue a supporting or opposing opinion, and adjusts its rhetoric depending on which political party it is writing for. “Many political speeches show the same structures and same characteristics regardless of the actual topic,” writes researcher Valentin Kassarnig. “The goal is to generate speeches where no one can tell the difference to hand-written speeches.”
To that end, the team had their system analyze more than 4000 speeches delivered during U.S. Congressional floor debates. Using the vocabulary, grammatical structures, and rhetorical flourishes it observed, the system began generating its own speeches. The results were surprisingly coherent: While many A.I. systems can mimic human writing for a few words or sentences at a time, their creations rarely make much sense when read as a whole. That the political speech generator created reasonably logical speeches is a testament to the Amherst team’s sophisticated design—and throws into sharp relief just how formulaic political speeches can be. Check out a sample from the report below:
Mr. Speaker, for years, honest but unfortunate consumers have had the ability to plead their case to come under bankruptcy protection and have their reasonable and valid debts discharged. The way the system is supposed to work, the bankruptcy court evaluates various factors including income, assets and debt to determine what debts can be paid and how consumers can get back on their feet. Stand up for growth and opportunity. Pass this legislation.