Figure - PMC

Skip to main content

An official website of the United States government

Here's how you know

Here's how you know

Official websites use .gov
A .gov website belongs to an official government organization in the United States.

Secure .gov websites use HTTPS
A lock ( ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites.

View full-text article in PMC

. 2019 Mar 22;10:1334. doi: 10.1038/s41467-019-08931-6

Search in PMC
Search in PubMed
View in NLM Catalog
Add to search

© The Author(s) 2019

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

PMC Copyright notice

Fig. 2 — Forced-choice classification with indirectly encoded fooling images. a In Experiment 1, 200 subjects saw one fooling image at a time (48 images total), and chose between two candidate labels: The machine’s choice for that image, and a random label drawn from the other images in the imageset. b In Experiment 2, 200 subjects chose between the machine’s first-choice classification and its second-choice classification. c In Experiment 3a, 200 subjects saw the same images as before, but with all 48 labels visible at once. d In all 3 experiments, most subjects agreed with the machine more often than would be predicted by chance responding (yellow bars), and most images showed human-machine agreement more often than would be predicted by chance responding (red bars). Given that raw classification accuracy in human subjects will be modulated by factors such as attention, motivation, engagement with the task, time spent studying an image, etc., we report the percentage of subjects who agreed with the machine at above-chance rates, and the percentage of images that showed above-chance human-machine agreement. For Experiment 1, the 95% confidence interval for the % of subjects with above-chance classification was [94.6% 99.4%], and it was [82.8% 98.7%] for the % of images with above-chance classification. For Experiment 2, these intervals were [87.7% 95.7%] and [58.0% 83.7%], respectively. For Experiment 3, these intervals were [83.2% 93.2%] and [67.7% 90.7%], respectively. Across all three experiments, these outcomes were reliably different from chance at p < 0.001 (two-sided binomial probability test)