Overview

Adam explores Artificial Intelligence from a feminist perspective and critiques current AI projects as being heavily masculinist. She surveys the field of AI paying close attention to its history and its various developments and how the gradual growth of the field has come to reinforce this highly gendered perspective. The danger in the male dominated growth of AI is that the AI project attempts to capture human reason, and through the neglect of feminist perspective, the reflection of humanity that AI provides is very lacking. Adam also critiques the process and approach of AI, exposing some of its flaws as a representative system.

Notes

The key word in the application of masculinist ideology into AI is inscription: AI is inscribed with gender, and it must be read to be revealed. (p. 11) AI has a culture that is in conflict over certain issues: science vs engineering, cognitive vs computer science. AI is virtual: study of non-physical objects. Artifacts are non-present.

Keith Grint and Rosalind Gill 1995: Feminist analyses exist between serving masculinity and technology vs acknowledging the force of masculinity and patriarchy. Studies which assume terms of masculinity and patriarchy without explanation tend to essentialist claims. (p. 17) Pushing women into technology indicates sexual politics: assumes neutrality of institutions (p. 19) Exploring gender as female implies that male is norm: masculinity is neutral while gender is socially constructed. Harraway: Sex and gender are a complicated systems. (p. 22)

AI is inscribed with gender. AI concerned with knowledge and simulation of knowledge. Subject of knowledge teeters between visible and invisible. The subject is the one who actually knows and is doing the knowing, which is usually omitted from AI study, usually assumed as male standard. (p. 29) Nagel 1986 critiques epistemology and propositional knowledge: “view from nowhere”. In propositional logic (S knows p) ‘S’ is universal and perspectiveless. Gilbert Ryle 1963: Knowing what vs knowing how. (p. 30) WWK or Women’s Ways of Knowing contrasts with women’s epistemology, but forms an essentialist text.

Adam reviews AI in the context of its development. There is still the persistent perception of AI as creating artificial minds, fraught with conflicts of thinking vs doing, building vs understanding. The history here focuses on the original tasks of AI. (p. 34) The underlying tools used to build AI are: search, predicate logic, decision making, heuristics, state traversal, “bounded rationality”, and planning. These approaches relate to work done by researchers in the 1950s, and excludes affairs that might be taken up by women. It was not a deliberate choice to venerate male reason, but rather it was the natural choice for researchers. (p. 36)

AI is about framing: symbolic AI interlinks with Saussure’s semiotics. Investigate Zenon Pylyshyn’s “Computation and Cognition”: work supporting GOFAI (p. 38) Representational knowledge frames employ real world– or interpretation of it. Knowledge vs stereotype. AI confuses representation of simplified model with real world that supposedly informs model.

Adam investigates connectionist networks (neural networks) as opposed to symbolic systems. These seem less ideologically steeped in male reasoning, but still depend on simulation, and still must be told what to do. (p. 45)

Adam reviews philosophical critiques (and other critiques of AI. She is very careful to do so in context and focus of what AI is and is trying to accomplish. AI can never be framed as a philosophical test. (p. 48) Searle critiques AI on basis of “intentionality”, which is some sort of human response (Chinese Room). Dennett critiques this argument based on total impossibility/implausibility of argument. (pp. 51-52) Dennett: computer acts as an intentional way insofar as it may be interpreted to do so. Phenomenological critique of AI: Dreyfus: computer cannot know how. Rule following leads to infinite regress, depends on “what we are”. This is one perspective that can be linked to the feminist critique without too much extension. (p. 55)

AI issues are: representation, intentionality, agency, and culture.

Finding the knowing subject/ knower in Cyc and Soar against a universalist perspectiveless viewpoint. Losing the knower preserves the conservative masculine “normality” and deflects responsibility. Knowing implies Responsibility! Considering traditional epistemologies: * implicit individualism, * absence of identity, * non-wierdness, * cultural imperialism. (p. 70) Subjective knowledge and consistency: Weirdness is non-beholden to standards in masculinist perception (ie, white, middle class, rational, academic culture) (p. 74)

Working from feminist epistemology: The anonymity of the inventor/scientist promotes technological imperialism “a technology appears”. Critiques great man theory, but grants anonymity of the author, who is given additional power through implicit assumption that technology is natural development. The subject “we” must be enacted, not given. (p. 77) Self knowledge and awareness demands a cultural/contextual knowledge/awareness. (p. 78) Responsibility and judgment: knower as participant vs total objectivity. Classical idealized knower is objective and detached, external to the world’s history: forms AI’s development as rooted in classical epistemology. Responsibility of agents (moral agents) is autonomous or by design?

Cyc project: Implies universality of subjects. Knowledge held is supposed to be common to everyone. Compare subject of Cyc with (for example) Wikipedia. How does Cyc cope with beliefs? Are beliefs necessary? How are they different from common sense? Mary Hesse: Observations themselves are mediated by other theories. (p. 85) Assumes universality, common denomination for decisions. Cyc implies authority, non-subjective nature. Implies normality: “healthy, sane, non-babies” who decides health and sanity? (p. 90)

Soar built from GPS (generalized problem solver). Soar meant as a candidate for a “unified theory of cognition”. The problem still is that it is a view from somewhere- that being male college students, aiming for “unnatural” problems. (p. 94)

Critique of objective reasoning: There should be plurality of voice in problem solving, a sharing of responsibility. Compare the independent Cartesian man of Reason with collectivist responsibility. (p. 98)

Language and AI: There is a symbolic order of language. Derrida identifies three isms in symbol-value systems relating to rationalist take on world: Logocentrism (supremacy of spoken word), Phallocentrism: “denotes a unitary drive toward a single, ostensibly reachable goal” (Tong 1994. p222), Dualism: way everything framed as binary oppositions. (p. 107) Knowing how vs knowing that underlies phenomenology. Knowing is inseparable from being. Attempting to separate them is Cyc’s failure: reduces the state of being into “somatic primitives” (p. 115) Focus of Cyc is knowledge, focus of Soar is architecture. In both, poor choice of problem leads to faulty theory. (p. 127)

Concerning embodiment, can assent with Andy Clark. Rationalist view denies the body. (p. 129) Lakoff critiques the objectivist perspective of AI and computation. The stance of experientialism wants to know why human conceptual system is why it is. The mind as machine perspective cannot cope with the manner in which different conceptual systems are organized: these assume equivalence occurs when one system may be translated into another. (p. 133) in Embodiment and A-Life: control is still limited by behaviorist notions of importance. Helmreich 1994 explores the subtext of A-Life: as an attempt to create life in-silico. Works like creation stories wherein masculine god (aka male programmer), breathes life into female program to create digital life. (p. 152) A-Life situation still has an absence of real human need or meaning, and an absence of social situatedness for populace of environments. Usually evolutionary programs/robots are aimed at competition, mating, creating some sort of economy or trade, but do not have any fun. These societies tend to be developed on basis of fitness and competition, and do not represent the larger social situation. (p. 154)

Adam questions what form a feminist AI project might have. Would it look any different? Built with the classical masculinist tools, could it work any differently? How could feminist programmers radically subvert AI? Ref to Mulvey and Film? (p. 157) The depth and personal situation in language and behavior is deep in human interaction. How to express reception or expectation or understanding, ambiguity or subtlety in AI or in simulated world? These are still lacking even from projects such as The Sims. (p. 162)

A distressing image here: Helmreich cites Hayles: “A male programmer mating with a female program to create a progeny whose biomorphic diversity surpasses the father’s imagination.” Adam continues: “The desires are to make the body obsolete, to play god in artificial worlds, and to download minds into robots. Such desires are predicated on the assumption that if a machine contains the contents of a person’s mind then it is that person. The body does not matter; it can be left behind.”