2022

Two-Turn Debate Doesn't Help Humans Answer Hard Reading Comprehension Questions

Parrish, Alicia, Trivedi, Harsh, Nangia, Nikita et al.

Understand

The use of language-model-based question-answering systems to aid humans in completing difficult tasks is limited, in part, by the unreliability of the text these systems generate.

  • Using hard multiple-choice reading comprehension questions as a testbed, we assess whether presenting humans with arguments for two competing answer options, where one is correct and the other is incorrect, allows human judges to perform more accurately, even when one of the arguments is unreliable and deceptive.
  • If this is helpful, we may be able to increase our justified trust in language-model-based systems by asking them to produce these arguments where needed.
  • Previous research has shown that just a single turn of arguments in this format is not helpful to humans.

Built on

  • Quantifying interpretability and trust in machine learning systems

    Original

    Philipp Schmidt and Felix Biessmann. 2019 · 1901

    Earlier work this paper cites.

  • AI safety via debate

    Original

    Geoffrey Irving, Paul Christiano, and Dario Amodei. 2018 · 2018

    Earlier work this paper cites.

  • Explainable machine-learning predictions for the prevention of hypoxaemia during surgery

    Scott M. Lundberg, B. Nair, M. Vavilala, M. Horibe, M. Eisses, Trevor Adams, D. Liston, Daniel King-Wai Low, Shu-Fang Newman, J. Kim, and Su-In Lee. 2018 · 2018

    Earlier work this paper cites.

Similar

  • "Hello AI": Uncovering the onboarding needs of medical practitioners for human-AI collaborative decision-making

    Carrie J. Cai, Samantha Winter, David Steiner, Lauren Wilcox, and Michael Terry. 2019 · 2019

    Cited alongside, same era.

  • On human predictions with explanations and predictions of machine learning models: A case study on deception detection

    Vivian Lai and Chenhao Tan. 2019 · 2019

    Cited alongside, same era.

  • Does the whole exceed its parts? the effect of ai explanations on complementary team performance

    Gagan Bansal, Tongshuang Wu, Joyce Zhou, Raymond Fok, Besmira Nushi, Ece Kamar, Marco Tulio Ribeiro, and Daniel Weld. 2021 · 2021

    Cited alongside, same era.

Then

  • WebGPT: Browser-assisted question-answering with human feedback

    Original

    Reiichiro Nakano, Jacob Hilton, Suchir Balaji, Jeff Wu, Long Ouyang, Christina Kim, Christopher Hesse, Shantanu Jain, Vineet Kosaraju, William Saunders, et al. 2021 · 2021

    Later among the works it cites.

  • Quality: Question answering with long input texts, yes!

    Richard Yuanzhe Pang, Alicia Parrish, Nitish Joshi, Nikita Nangia, Jason Phang, Angelica Chen, Vishakh Padmakumar, Johnny Ma, Jana Thompson, He He, and Samuel R. Bowman. 2022 · 2022

    Closest in time.

  • Single-turn debate does not help humans answer hard reading-comprehension questions

    Alicia Parrish, Harsh Trivedi, Ethan Perez, Angelica Chen, Nikita Nangia, Jason Phang, and Samuel Bowman. 2022 · 2022

    Closest in time.

Beyond the bibliography

alphaXiv searches the wider corpus for related work and actual follow-ups.

Open on alphaXiv

alphaXiv is searching for related work…