鈴木プロジェクトでは、技術哲学の国際学会、Philosophy of Human-Technology Relations Conference (PHTR) 2020(オランダ、Twente大学)で、道具としてのAIをテーマとしたパネルセッションを実施した。プロジェクトでは、自律型エージェントとしてのAIと道具としてのAIの対比を軸として、道具としてのAIのさまざまな可能性を検討した。新型コロナウイルス感染症の感染拡大のため、学会はオンライン開催となったが、パネルセッションには20名ほどが参加し、提題に続く質疑応答の時間には、海外の研究者から(さらには進行役を担当してくれたオランダの学生からも!)多くの質問が提出され、活発な議論が展開された。一連の議論を通じて、道具としてのAIという視点の有効性があらためて確認された。以下はセッションの趣旨の和訳と各提題の要旨である。(報告:鈴木貴之)
*各提題の動画がこちらに公開されています。
セッションの趣旨:AIが現代の最も重要なテクノロジーの1つであることは間違いないだろう。近年、AIの倫理的・社会的問題が論じられる際に念頭に置かれているのは、人間のような汎用知能をもった自律型エージェントとしてのAIである。しかし、AIを道具の一種と見るならば、AIにはさらにさまざまな可能性があることがわかる。このセッションでは、道具としてのAIの可能性を理論的・哲学的に検討する。鈴木は、AIについての2つの見方、すなわち人間の代替物としてのAIと人間知能の補完物としてのAIという見方を比較する。柴田は、AI研究の歴史におけるこれら2つの見方のあいだの論争に検討を加える。上杉は、棒タイプの道具と器タイプの道具の比較を通じてAIを分析する。立花は、人間のよりよい道徳的判断を支える道具としてのAIの可能性を検討する。
Two Conceptions of Artificial Intelligence
Takayuki SUZUKI
Abstract: Artificial Intelligence research has been considered as a project to create an autonomous agent with general intelligence that is close to/equal to/superior to human intelligence. The attempts to create such an agent, especially ones based on a classical approach on AI, however, have faced various theoretical difficulties. Though recent AI researches based on deep neural networks succeed in overcoming some of the difficulties, it will take some time, it seems, to create an intelligent agent that is equal to human beings. There is another conception of AI, however, that AI is an intellectual tool to complement and enhance human intelligence. According to this conception, AI need not necessarily be general nor autonomous. It need not solve intellectual tasks in the same way as humans do. In this talk, I will argue that it is AI research based on this second conception that will be more fruitful at least in a short term and socially more important. I also will show concrete examples of AI research on the second conception.
AI vs. IA : The Real issues hidden in the struggle
Takashi SHIBATA
Abstract: There is a view that the development history of a computer could be considered as the rise and fall of two camps: autonomous intelligent machine (AI) vs. intelligence amplifier (IA) as a tool. In fact, there is a clear difference in the vocabulary used by the development leaders in early days. While John McCarthy and Marvin Minsky saw a computer as an automaton that “substitutes” for humans, Douglas Engelbert saw it as an instrument that “extends” human abilities. These two vocabularies, “substitution” and “extension”, were not limited to the development scenes of it, but were also introduced in various science fiction and future predictions. The former genealogy has given rise to the “singularity” theory, which discusses human “extinction”, and the latter genealogy to the “Homo Deus”, which means superhuman or “post human”. Before deciding which vision of the two is appropriate, I will point out defects common to both camps by clarifying that these vocabularies have the same root, and they are two parts of a whole. Based on the above, I pose real issues that should be addressed in the discussion about AI and IA.
Considerations on Analysing Relations Between Humans and AI technologies Based on Archetypes of Instruments ― Club-type and Pot-type
Shigeru WESUGI
Abstract: This talk considers how humans – AI technologies relations can be analysed from the perspective of instrument. Therefore, the speaker focuses on the archetypes of instruments – club-type and pot-type instrument (Kenzo SAKAMOTO, 1975, Kenji EKUAN, 2000 ).
The club-type instruments include from primitive stone knives and spears to modern machine tool and power shovel. The club-type instruments let users transform a shape of an object including straining and cutting off the object. The users can directly work on the object through reflecting an intention.
Meanwhile, the pot-type instruments include from stone dish and clay vessels to clothes and farming ground widely. The pot-type instruments preserve an object by reducing external influence, or mature an object by maintaining an appropriate environment. The users can work on the environment around the object not directly on the object. Therefore, the users often delegate the instruments to work on to the object.
Which of types are AI technologies? AI technologies analyse images, voices and texts by reading data in a sector and tuning variety of parameters. This working process indicates AI technologies are included in pot-type one inherently. Recently, such process has become a black box and the users can use it as if they utilized a club-type instrument.
As for pot-type, gradual cultivation in a frame and delegation of users are emphasized rather than shaping the object directly. The speaker analyses the humans-AI technologies relations based on this perspective and additionally considers a design approach of combining both types complementarily.
Artificial intelligence and Human Moral Virtue
Koji TACHIBANA
Abstract: Artificial intelligence has been equipped with various devices for supporting human social judgments. Although such equipment has rapidly been embedded into the different aspects of our society, such as medical care, industries and education, any discussion can hardly be found about the ELSI of AI for use in moral education. If cultivating citizens’ morality has a substantial contribution to their society, AI for use in moral education must be one of the genuine concerns for society. This presentation discusses possible types of AI systems for use in moral education and examines their ethical, legal, and social implications (ELSI). First, a brief survey will be provided about the current situation concerning the AI for use in education. Second, the notion of moral will be analysed and formalised by using the notion of virtue, based on virtue ethics. Third, some existing proposals for Al for use in moral education will be examined. Fourth, an alternative will be proposed, and their ELSIs will be examined. It will then conclude that AI can have a significant contribution to moral education, and their ELSI must cautiously be examined.