In DepthComputer Science

Has artificial intelligence become alchemy?

See allHide authors and affiliations

Science  04 May 2018:
Vol. 360, Issue 6388, pp. 478
DOI: 10.1126/science.360.6388.478

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
CAPTCHA

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

Image CAPTCHA
Enter the characters shown in the image.

Vertical Tabs

  • Reproducibility problems of artificial intelligence inherently come from random numbers

    Matthew Hutson wrote an article entitled “Has artificial intelligence become alchemy?” (1). Matthew only focused on deep learning as artificial intelligence. There are two kinds of artificial intelligence: deductive reasoning and inductive reasoning. Deep learning is classified into inductive reasoning, stochastic reasoning, or statistical reasoning. Since stochastic (statistical) reasoning schemes are all based on random numbers, generating random numbers are to change the result of deep learning. Deductive reasoning schemes do not use random numbers so that the reproducibility problem does not exist. Many of artificial intelligence researchers are not aware of the importance of a random number seed. Before running an artificial intelligence (deep learning) program, the random number seed should be fixed. Without fixing the random number seed, the result may be changed. In other words, the reproducibility problems of the artificial intelligence (deep learning) can be fixed.

    References:
    1. Matthew Hutson, Has artificial intelligence become alchemy?, Science 04 May 2018: Vol. 360, Issue 6388, pp. 478

    Competing Interests: None declared.
  • Advancing Artificial Intelligence

    The characterization of artificial intelligence (AI) as alchemy, albeit emotive, is merely an acknowledgment of the lack of understanding of the workings of the underlying algorithms (1, see also 2). In developing an explicit theoretical understanding of intelligence, following in the footsteps of physics makes perfect sense. However, the toy problem (black-and-white images instead of color photos) prescribed for gaining insights into recognition algorithms appears to be diagnostic of a deeper problem with contemporary AI, which is getting lost in the here & now of immediate solutions and losing sight of the big picture: development of the science of intelligence (3). Here we suggest a different approach: mathematical knowing as a Bohr atom of knowing, which can serve as a solid foundation to build the science of human intelligence (subsuming reasoning, perception, and cognition, among others).

    But how, and why?

    First, science is a hallmark of human intelligence. As such, a scientific account of science constitutes the science of intelligence. Although we do not have the science of the development of scientific theories and models, we have functorial semantics: a mathematical account of abstracting theories and building models of various categories of mathematical objects (4, 5). Taking a cue from Galileo's investigation of a simple motion—motion of falling bodies—that served as a foundation for the science of motion, we suggest investigating t...

    Show More
    Competing Interests: None declared.