2. はじめに
このたび、ロボット倫理学の重要文献、Wendell
Wallach and Colin Allen, Moral Machines:Teaching
Robot Right from Wrong, Oxford University Press
2009を久木田さんと一緒に翻訳した。(ウェンデル・
ウォラック、コリン・アレン『ロボットに倫理を教え
る:モラルマシーン』名古屋大学出版会、近刊)
ところが原書刊行から約10年も経ってしまった。ふ
つうの倫理学の研究書なら十年程度なんの問題もない
が、人工知能の場合、刊行後数年でブレイクスルーが
起こり、予想を超えた大発展が生じてしまった。(阪
大のロボット研で2013年頃に某氏が「ディープラーニ
ングはヤバイ」と言っていたのをよく覚えている。)
16. 書評
• Moral Machinesの刊行後、哲学の最新モノグラフ
の書評サイトNDPRや、情報倫理学や技術倫理に関
する有力ジャーナル各誌に書評が掲載された。
• どれもそれなりの長文で、重要な扱いがなされてい
る。
• Beavers, Anthony F.(2010) "Wendell Wallach and Colin Allen: Moral machines:
teaching robots right from wrong [Book Review]" Ethics and Information Technology 12
(4):357-358
• Danielson, Peter (2009) “Moral Machines: Teaching Robots Right from Wrong.” Notre
Dame Philosophical Reviews
• Pritchard, Michael S. (2012) “Moral Machines?” Science and Engineering Ethics 18
(2):411-417
• Tavani, Herman (2011)"Can we Develop Artificial Agents Capable of Making Good
Moral Decisions?" Minds and Machines 21 (3):465-474
20. 反応の一例
“Similarly, Wallach and Allen (2009) introduced a
diagram by which three classes of moral agency
(operational morality, functional morality, and full
moral agency) are distributed on a scale indicating
progressive degrees of moral sensitivity and
autonomy.” Fossa (2018)
“Wallach and Allen also contend AI must be installed
with a “functional morality” that empowers machines
with the capacity to assess and respond to moral
challenges” Grech et al. (2017)
22. 反応の一例
“It is for reasons like these, one should note, that
suggestions for a moral Turing Test (Wallach and Allen
2008) lean too heavily on imitation as an ethical
benchmark.” Arnold & Scheutz (2018)
“The Moral Turing Test developed by Wallach and Allen
(2009) is similar, but the behavior in question is in
relation to some ethical decision (or action)” Roff (2014)
“Allen and Wallach have broached this concept (coining
the acronym MTT for moral Turing test), rightly noting
some difficult problems an MTT would have to resolve
in order to be viable (Allen et al. 2006; Wallach and
Allen 2008). But while their discussion acknowledges
that the MTT is ‘‘inadequate’’, the idea is still not
wholly abandoned.” Arnold & Scheutz (2016)
24. 反応の一例
“Actually, for a robot to be an ethical learning robot
both top-down and bottom-up approaches are needed
(i.e., the robot should follow a suitable hybrid approach).
Typically, the robot builds its morality through
developmental learning similar to the way children
develop their conscience.” Tzafestas (2018)
“Top-down approaches involve turning explicit theories
of moral behavior into algorithms. Bottom-up
approaches involve attempts to train or evolve agents
whose behavior emulates morally praiseworthy human
behavior [12].” Azad-Manjiri (2016)
“Wallach and Allen contrast this top down approach to a
bottom up one, in which an emphasis is placed on
‘creating an environment where an agent explores
courses of action and learns and is rewarded for
behavior that is morally praiseworthy’” Sharkey (2017)
26. 反応の一例
“Allen and Wallach distinguish between a top-down
theoretically driven approach, a bottom-up developmental
or explorative approach and finally a hybrid (Wallach and
Allen, 2009, Chapter 6, 7 and 8), which combines both
top-down and bottom-up approaches and furthermore
includes a virtue ethical component.” Gerdes & Øhrstrøm
(2015)
“According to Wallach and Allen, virtue ethics is a
superior ethic to both deontology and consequentialism,
at least with respect to their computability. And since
this is what’s taken to matter at this stage in the game,
with this insight we are one step closer to the successful
development of AMAs. However, although it is important
to ask whether a virtue-ethical framework could be
successfully computed into machines, it must also be
asked whether the act of creating virtuous machines is
permitted by the tenets of virtue ethics.” Tonkens(2012)