Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
Machines we trust: perspectives on dependable AI
Pelillo M., Scantamburlo T., MIT Press, Boston, MA, 2021. 174 pp. Type: Book (978-0-262542-09-8)
Date Reviewed: May 24 2022

Skynet is here. This was my mindset when I undertook to review Machines we trust. The book comprises an introduction plus eight chapters divided into three parts: “Setting the Stage” (chapters 2 and 3); “Issues” (chapters 4 through 7); and “Prospects” (chapters 8 and 9). The individual and joint authors of these eight contributions are savvy, articulate, and widely published in intersecting areas of ethics, philosophy, and artificial intelligence (AI). Computer scientists will find fascinating material here, but none of the nuts and bolts. This book is about trust, intelligibility, explainabilty, augmentation, moral reasoning, and much more. A frequent position is that the commonly expressed goal of human-centered AI should be reversed to AI-decentered humanity.

Norbert Wiener set the stage, in 1954, when he warned:

Any machine constructed for the purpose of making decisions, if it does not possess the power of learning, will be completely literal-minded. Woe to us if we let it decide our conduct, unless we have previously examined the laws of its action, and know fully that its conduct will be carried out on principles acceptable to us! [1]

He could see Skynet.

The last time I taught AI was in 1994, 17 years before I retired. I learned Lisp and AI from the creators, John McCarthy and Marvin Minsky. I was an RA for Jim Slagle on SAINT and a member of the Chess Project. Concerning AI, one thing I observed repeatedly was that everything changed and nothing changed.

Then I read chapter 2, “Shortcuts to Artificial Intelligence.” According to author Nello Cristianini, although not in his words, there were three major sellouts starting two decades ago that have altered AI forever. First, the long quest for the perfect (or even adequate) algorithm was replaced by machine learning. Second, highly engineered and curated training data turned into found data, for example, scraped from the Internet. Third, data annotations by experts, or at least mechanical turks, became proxies such as click counts. Dedication to Skynet.

Two other chapters are striking. The first is chapter 6, “Cobra AI: Exploring Some Unintended Consequences of Our Most Powerful Technology,” by Federico Cabitza. This chapter outlines not the disasters of failed AI systems, but the threats and dangers of successful ones. Most examples are from healthcare--appropriate for a developer of medical AI systems.

The second is chapter 9, “The AI of Ethics,” by Robert C. Williamson. This chapter takes the frequently studied ethics of AI and turns it on its head. After listing seven theses justifiably held for the former and slanted toward moral reasoning as a human-only unaugmentable activity, Williamson gives seven anti-theses that show that the reality of AI in ethics is more graded, finer grained, and more nuanced. He stresses that technology has always supported philosophy, starting with the externalized memory of writing.

Several chapters deal with properties such as efficacy, effectiveness, intelligibility, convenience, and so on. They insist on the involvement of all stakeholders (commissioners, developers, deployers, users, those impacted by decisions, and others) in determining what should be considered and measured, how, and why.

Unfortunately, this book has many minor flaws that copyediting should have caught. Some appear to be a translation issue, but a lot of them are just carelessness. Chapter 7 is particularly riddled with errors.

As far as Skynet goes, a 1944 quote from Lewis Mumford can close this discussion as well as it ends the book: “Nothing that man has created is outside his capacity to change, to remold, to supplant, or to destroy: his machines are no more sacred or substantial than the dreams in which they originated.”

Here are four recent books worth comparing to the one under review. Trustworthy AI [2] is closest, covering almost the same topics but with a strong business orientation. After exposing the failure of today’s AI, Rebooting AI [3] offers a systematic revision to improve both outcome and trustworthiness. The luminaries writing [4] also offer a redo, but critics have suggested this is not the profile we need for a new AI--think Skynet. The centrality of the author in [5] is not as an expert but as the interviewer of 23 experts, obviously fans, but it is valuable to see what they do. Included is Cynthia Breazeal, whose lab I once made a late-night visit to for “midnight supplies.” It is also the least expensive with the most pages. The book under review is the opposite.

Reviewer:  Benjamin Wells Review #: CR147443 (2208-0111)
1) Wiener, N. The human use of human beings: cybernetics and society. Da Capo Press, New York, NY, 1954.
2) Ammanath, B. Trustworthy AI: a business guide for navigating trust and ethics in AI. Wiley, Hoboken, NJ, 2022.
3) Marcus, G.; Davis, E. Rebooting AI: building artificial intelligence we can trust. Pantheon Books, New York, NY, 2019.
4) Kissinger, H. A.; Schmidt, E.; Huttenlocher, D.; Schouten, S. The age of AI: and our human future. Little Brown and Company, New York, NY, 2021.
5) Ford, M. Architects of intelligence: the truth about AI from the people building it. Packt, Birmingham, UK, 2018.
Bookmark and Share
  Featured Reviewer  
 
Philosophical Foundations (I.2.0 ... )
 
 
Privacy (K.4.1 ... )
 
 
Sociology (J.4 ... )
 
 
General (K.4.0 )
 
 
Artificial Intelligence (I.2 )
 
Would you recommend this review?
yes
no
Other reviews under "Philosophical Foundations": Date
Handbook of personal data protection
Madsen W., Stockton Press, New York, NY, 1992. Type: Book (9780333569207)
Nov 1 1993
Privacy and security issues in information systems
Turn R., Ware W., Wadsworth Publ. Co., Belmont, CA, 1985. Type: Book (9780534042578)
Nov 1 1985
Data bases
Burnham D., Wadsworth Publ. Co., Belmont, CA, 1985. Type: Book (9780534042578)
Nov 1 1985
more...

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy