Monday, February 16, 2009

The AI debate and 20th century philosophy

Philosophy is one of the most important sources for getting alternative perspectives on a
problem. In their critique of the Artificial Intelligence (AI) tradition, Winograd and Flores
were inspired by the work of the German philosopher Martin Heidegger. Winograd & Flores
found Heidegger particularly interesting in that they found his ideas to be in direct opposition
to most of the implicit assumptions of the AI field at that time.

Winograd and Flores argued that Heidegger´s understanding of the human condition is a
better foundation for understanding and designing computer technology than the ruling
paradigm in AI at that time. As they saw it, the cognitivist approach to understanding
computers in use must be rejected if we take Heidegger seriously. In this critique they
followed up the early work of Dreyfus (“What computers can’t do”, 1972).

The AI debate as such has limited relevance for a study of interactivity. The reason is that
its research question is totally different. The AI debate has centered around the question “Can
we build intelligent computers?”. This was a reaction to early AI research that mainly asked
the question “How do we build intelligent computers?”. The debate led to the question “What
is intelligence?”, which again led to the question “In what ways are computers different from
people?”. The latter is the question Dreyfus discussed in (1986) with reference to philosophy.
We are then very close to the philosophical question “What does it mean to be human?”.

Dreyfus entered the AI debate from philosophy. He soon realized that the early AI
researchers were in many respects doing philosophy of mind, but with little or no knowledge
or reference to two and a half thousand years of philosophical research on the subject. The
main differences between the AI researchers and the philosophers were their choice of
medium and their choice of research methodology. For the philosophers, the medium has
always been language, and the methodology has always been the philosophical discourse. For
the early AI researchers, the medium was the computer, and the research methodology was
systems construction. Dreyfus showed how the AI research, despite these differences,
repeated ancient discussions in the philosophy of mind. The strength of Dreyfus’ analogy is
that it enabled him to make predictions about the results of these “discourses” based on his
knowledge of the similar philosophical discourses. His predictions have so far to a large
degree been correct. I find this to give a strong credibility to his argument.

The strongest relevance of the AI-debate for the current study of interactivity is in its use
of philosophy. It showed to many in the computer-science community that philosophy can be
used as a resource and inspiration without having to become a philosopher, much in the same
way as researchers in computer science have always used mathematics without becoming
mathematicians. Dreyfus (1972) draws mainly on three philosophers: Heidegger, Merleau-
Ponty, and Wittgenstein. As the work of these philosophers have relevance for the current
work, a short introduction is appropriate.

Martin Heidegger (1889-1976) belongs to the phenomenological tradition in Continental
philosophy. One of its most influential proponents was his teacher Edmund Husserl (1859-
1938). For Heidegger it was important to move philosophy back from the realm of the spheres
to the reality of everyday human life. This meant, as he saw it, a definite break with 2000
years of philosophical tradition. In his "Being and Time" from 1927 (Heidegger, 1997), he
breaks with the tradition of exploring ideas without reference to our factual existence as
human beings. He departed from his teacher concerning the possibility of making explicit this
“background” of everyday practices that gives meaning to the world. In trying to develop a
philosophy starting out with our factual human existence, he found himself trapped in the web
of meaning produced by the basic assumptions of Western civilization. He found it necessary
to develop a set of new concepts better suited for the task. Reading Heidegger consequently
becomes a difficult task, as one first has to acquire his new "language". The problem is that
this language can not be fully understood purely through definitions referring back to our
"ordinary" language. The meaning of his concepts slowly emerge through the reading of the
work. The reading of Heidegger thus becomes an iterative process, or what in philosophy is
called a hermeneutic circle.

The French philosopher Maurice Merleau-Ponty (1908-1961) was heavily influenced by
both Husserl and Heidegger. Put simply, Heidegger brought philosophy back to everyday
human life, while Merleau-Ponty took it all the way back to the human body. In Merleau-
Ponty´s most important work The Phenomenology of Perception from 1945 (Merleau-Ponty,
1962) he explored the implicit assumptions about perception at that time. He ended up with an
understanding of perception that is totally different from the naive idea of perception as
stimuli reception. The latter view can still be found in part of the current literature on Human-
Computer Interaction (HCI). To Merleau-Ponty, perception is a process where an active body
enters into a "communion" with its surroundings. Perception is a continuos interaction
involving the subject's intentions, expectations, and physical actions. From this perspective,
every attempt at applying some variation of Shannon and Wiever's information theory (see
Reddy, 1993) to Human-Computer Interaction becomes an absurdity. There is clearly no purely
active "sender" or purely passive "receiver", nor any well-defined "information" or "point in
time". The fact that his understanding is in direct opposition to some of the most influential
theoretical foundations of the HCI field, makes a study of Merleau-Ponty an interesting
starting point for an exploration of human-computer interaction.
Since Merleau-Ponty published Phenomenology of Perception in 1945, phenomenology
as a philosophical discipline has developed further. The most complete attempt to date at
building a complete analysis of human existence based on the phenomenological insights is
done by Schutz and Luckman in their The structure of the life-world (1973). Luckman uses
this framework as a foundation for his current empirical study of everyday social interaction.
As a sociologist, he makes use of light-weight video equipment and films long sequences of
everyday interaction between people in their natural surroundings. He then analyzes these
sequences in search of levels of meaning.

In British philosophy we find a similar questioning of the limits of analytical approaches
in the late work of Ludwig Wittgenstein (1889-1951). Wittgenstein started out with a pureanalytical approach to philosophy. His main interest was the philosophy of language. In his
most important early work “Tractatus” (Wittgenstein, 1923), he argued for the logical nature
of language and worked out a complete system for determining the “truth value” of sentences.
Referring to the AI debate, his early position would have placed him among the first
generation of AI researchers with their trust in an analytical, symbolic, and de-contextualized

After publishing “Tractatus”, Wittgenstein found no interest in philosophy, as he though
its problems to be “solved”. After some years as a school teacher in Austria, he started
questioning the foundations of his early work, and returned to Oxford. He struggled until his
death with all the paradoxes he found in his early approach. He never developed his
ideas into a coherent philosophy, but published his thoughts on the subject in
“Philosophical Investigations” (post-hume, 1953). He found one of the most important
problems of the analytical approach to an understanding of language to be its lack of attention
to context. This led him to develop the concepts language game and life form. To the late
Wittgenstein, the meaning of a sentence is given by its use. Language is primarily a means of
communication. In a certain use situation, there is a context of language users, physical
objects, and practices that give meaning to the words. He described these local uses of
language as language games. He further argues that all use of language is done within a
certain language game, whether it is involving only two people coordinating a specialized
task, or a discourse about the language of philosophy itself. For language users to be able
to comprehend the words of another language user, they need a shared background of
experience. This includes culture, corporeality, sensory system, social life etc. Wittgenstein
uses the term life form for this. To him, language users of different life forms can never truly
We see strong similarities between Heidegger, Merleau-Ponty, and Wittgenstein in their
critique of a purely analytical approach to philosophy. They all ended up with a focus on
everyday life, and on our factual existence as human beings. This is why Dreyfus found them
relevant for the AI debate, and this is why they are relevant for a discussion of interactivity.

No comments: