Real or Fake Text?

How good are you (and other humans) at knowing when text has been written by a computer? How many sentences can a computer write before it no longer has you fooled? Is the computer better at generating convincing text if the topic is news, or short stories, or maybe presidential speeches? And lastly, when you notice text has a mistake in it that gives it away as computer-generated, how easy is it to fix that mistake?

In this research project, we're asking for YOUR help to answer these questions. If you're familiar with the Turing Test, you'll be participating in a similar experiment!

Play Now!

Why participate?

Ever see Blade Runner? In a world where artificial intelligence powers more and more of our interactions, it's good practice to stay vigilant online. RoFT lets you evaluate just how good you are at noticing whether text has been generated by a machine. Try your skills at different domains, and hone your detection skills! You can even earn yourself bragging points by vying for a spot on the leaderboard.

Plus, do all this while aiding in academic research! We especially interested in which domains are harder/easier for humans to notice generated text in, and how the amount of randomness in the generated text impacts how easy the game is. Your contributions will be used to answer these research questions.

If you want to use our tool in your research, please cite our EMNLP paper:

  title={RoFT: A Tool for Evaluating Human Detection of Machine-Generated Text},
  author={Dugan, Liam and Ippolito, Daphne and Kirubarajan, Arun and Callison-Burch, Chris},
  booktitle={Empirical Methods in Natural Language Processing, Demo Track},

Made by researchers at the University of Pennsylvania. Art credit to Trung Le.