Policy ForumMachine Learning

Adversarial attacks on medical machine learning

See allHide authors and affiliations

Science  22 Mar 2019:
Vol. 363, Issue 6433, pp. 1287-1289
DOI: 10.1126/science.aaw4399

eLetters is an online forum for ongoing peer review. Submission of eLetters are open to all. eLetters are not edited, proofread, or indexed.  Please read our Terms of Service before submitting your own eLetter.

Compose eLetter

Plain text

  • Plain text
    No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.
Author Information
First or given name, e.g. 'Peter'.
Your last, or family, name, e.g. 'MacMoody'.
Your email address, e.g. higgs-boson@gmail.com
Your role and/or occupation, e.g. 'Orthopedic Surgeon'.
Your organization or institution (if applicable), e.g. 'Royal Free Hospital'.
Statement of Competing Interests

This question is for testing whether or not you are a human visitor and to prevent automated spam submissions.

Enter the characters shown in the image.

Vertical Tabs

  • Game theory with new ethical assumption for safety of AI in health care system
    • Yuichi Hirata, Associate Professor, Hokkaido University, Central Institute of Isotope Science, Graduate School of Biomedical Science and Engineering

    As explained in the POLICY FORUM (22 March, p. 1287), now two kinds of Artificial Intelligences (AIs) are appearing in the health care system (1).

    One is “Efficient AI”, and the other is “Adversarially attacked AI”.

    The Efficient AI had already been accurate, and surpassed human performance for visual recognition error rate of medical images (2).

    On the other hand, the Adversarially attacked AI of which machine learning systems confidently arrive at manifestly wrong conclusions, and intentionally misdiagnoses the medical images (1).

    By attacks based on ethical concerns, the Efficient AI with vulnerabilities might be altered to be the Adversarially attacked AI.

    The fights against the ethical concerns to ensure safety of the Efficient AI is considered to be a game to win or lose.

    Therefore, it might be analyzed by the game theory.

    Usually, the game theory is used for economical analysis to derive superior and cost-effective strategy by assuming that players of the game in the real world would be rational for costs and benefits.

    However, for the analysis of the game related with attacks for AI of the health care system, so as to ensure patient safety, the game theory should assume that reducing the ethical concerns causing attacks for AI has first and foremost priority.

    Such changes of the assumption in the game theory are revolutionary.

    However, new game theory with such new assumption is required for ens...

    Show More
    Competing Interests: None declared.