In philosophy of mind, the computational theory of mind (CTM), also known as computationalism, is a family of views that hold that the human mind is an information processing system and that cognition and consciousness together are a form of computation. Warren McCulloch and Walter Pitts (1943) were the first to suggest that neural activity is computational. They argued that neural computations explain cognition. The theory was proposed in its modern form by Hilary Putnam in 1967, and developed by his PhD student, philosopher and cognitive scientist Jerry Fodor in the 1960s, 1970s and 1980s. Despite being vigorously disputed in analytic philosophy in the 1990s due to work by Putnam himself, John Searle, and others, the view is common in modern cognitive psychology and is presumed by many theorists of evolutionary psychology. In the 2000s and 2010s the view has resurfaced in analytic philosophy (Scheutz 2003, Edelman 2008).
The computational theory of mind holds that the mind is a
computational system that is realized (i.e. physically implemented) by
neural activity in the brain. The theory can be elaborated in many ways
and varies largely based on how the term computation is understood.
Computation is commonly understood in terms of Turing machines
which manipulate symbols according to a rule, in combination with the
internal state of the machine. The critical aspect of such a
computational model is that we can abstract away from particular
physical details of the machine that is implementing the computation.
For example, the appropriate computation could be implemented either by
silicon chips or biological neural networks, so long as there is a
series of outputs based on manipulations of inputs and internal states,
performed according to a rule. CTM, therefore holds that the mind is not
simply analogous to a computer program, but that it is literally a
computational system.
Computational theories of mind are often said to require mental representation
because 'input' into a computation comes in the form of symbols or
representations of other objects. A computer cannot compute an actual
object, but must interpret and represent the object in some form and
then compute the representation. The computational theory of mind is
related to the representational theory of mind
in that they both require that mental states are representations.
However, the representational theory of mind shifts the focus to the
symbols being manipulated. This approach better accounts for
systematicity and productivity. In Fodor's original views, the computational theory of mind is also related to the language of thought.
The language of thought theory allows the mind to process more complex
representations with the help of semantics. (See below in semantics of
mental states).
Recent work has suggested that we make a distinction between the
mind and cognition. Building from the tradition of McCulloch and Pitts,
the computational theory of cognition (CTC) states that neural computations explain cognition. The computational theory of mind asserts that not only cognition, but also phenomenal consciousness or qualia,
are computational. That is to say, CTM entails CTC. While phenomenal
consciousness could fulfill some other functional role, computational
theory of cognition leaves open the possibility that some aspects of the
mind could be non-computational. CTC therefore provides an important
explanatory framework for understanding neural networks, while avoiding
counter-arguments that center around phenomenal consciousness.
"Computer metaphor"
Computational theory of mind is not the same as the computer metaphor, comparing the mind to a modern-day digital computer. Computational theory just uses some of the same principles as those found in digital computing.
While the computer metaphor draws an analogy between the mind as
software and the brain as hardware, CTM is the claim that the mind is a
computational system. More specifically, it states that a computational
simulation of a mind is sufficient for the actual presence of a mind,
and that a mind truly can be simulated computationally.
'Computational system' is not meant to mean a modern-day
electronic computer. Rather, a computational system is a symbol
manipulator that follows step by step functions to compute input and
form output. Alan Turing describes this type of computer in his concept of a Turing machine.
Early proponents
One of the earliest proponents of the computational theory of mind was Thomas Hobbes,
who said, "by reasoning, I understand computation. And to compute is to
collect the sum of many things added together at the same time, or to
know the remainder when one thing has been taken from another. To reason
therefore is the same as to add or to subtract."
Since Hobbes lived before the contemporary identification of computing
with instantiating effective procedures, he cannot be interpreted as
explicitly endorsing the computational theory of mind, in the
contemporary sense.
Causal picture of thoughts
At the heart of the computational theory of mind is the idea that
thoughts are a form of computation, and a computation is by definition a
systematic set of laws for the relations among representations. This
means that a mental state represents something if and only if there is
some causal correlation between the mental state and that particular
thing. An example would be seeing dark clouds and thinking "clouds mean
rain", where there is a correlation between the thought of the clouds
and rain, as the clouds causing rain. This is sometimes known as natural meaning.
Conversely, there is another side to the causality of thoughts and that
is the non-natural representation of thoughts. An example would be
seeing a red traffic light and thinking "red means stop", there is
nothing about the color red that indicates it represents stopping, and
thus is just a convention that has been invented, similar to languages
and their abilities to form representations.
Semantics of mental states
The computational theory of mind states that the mind functions as a
symbolic operator, and that mental representations are symbolic
representations; just as the semantics
of language are the features of words and sentences that relate to
their meaning, the semantics of mental states are those meanings of
representations, the definitions of the 'words' of the language of thought.
If these basic mental states can have a particular meaning just as
words in a language do, then this means that more complex mental states
(thoughts) can be created, even if they have never been encountered
before. Just as new sentences that are read can be understood even if
they have never been encountered before, as long as the basic components
are understood, and it is syntactically correct. For example: "I have
eaten plum pudding every day of this fortnight." While it's doubtful
many have seen this particular configuration of words, nonetheless most
readers should be able to glean an understanding of this sentence
because it is syntactically correct and the constituent parts are
understood.
Criticism
A range of arguments have been proposed against physicalist conceptions used in computational theories of mind.
An early, though indirect, criticism of the computational theory of mind comes from philosopher John Searle. In his thought experiment known as the Chinese room, Searle attempts to refute the claims that artificially intelligent agents can be said to have intentionality and understanding and that these systems, because they can be said to be minds themselves, are sufficient for the study of the human mind.
Searle asks us to imagine that there is a man in a room with no way of
communicating with anyone or anything outside of the room except for a
piece of paper with symbols written on it that is passed under the door.
With the paper, the man is to use a series of provided rule books to
return paper containing different symbols. Unknown to the man in the
room, these symbols are of a Chinese language, and this process
generates a conversation that a Chinese speaker outside of the room can
actually understand. Searle contends that the man in the room does not
understand the Chinese conversation. This is essentially what the
computational theory of mind presents us—a model in which the mind
simply decodes symbols and outputs more symbols. Searle argues that
this is not real understanding or intentionality. This was originally
written as a repudiation of the idea that computers work like minds.
Searle has further raised questions about what exactly constitutes a computation:
the wall behind my back is right now implementing the WordStar
program, because there is some pattern of molecule movements that is
isomorphic with the formal structure of WordStar. But if the wall is
implementing WordStar, if it is a big enough wall it is implementing any
program, including any program implemented in the brain.
Objections like Searle's might be called insufficiency objections.
They claim that computational theories of mind fail because computation
is insufficient to account for some capacity of the mind. Arguments from
qualia, such as Frank Jackson's knowledge argument,
can be understood as objections to computational theories of mind in
this way—though they take aim at physicalist conceptions of the mind in
general, and not computational theories specifically.
There are also objections which are directly tailored for computational theories of mind.
Putnam himself (see in particular Representation and Reality and the first part of Renewing Philosophy)
became a prominent critic of computationalism for a variety of reasons,
including ones related to Searle's Chinese room arguments, questions of
world-word reference relations, and thoughts about the mind-body
relationship. Regarding functionalism in particular, Putnam has claimed
along lines similar to, but more general than Searle's arguments, that
the question of whether the human mind can implement
computational states is not relevant to the question of the nature of
mind, because "every ordinary open system realizes every abstract finite
automaton." Computationalists have responded by aiming to develop criteria describing what exactly counts as an implementation.
Roger Penrose
has proposed the idea that the human mind does not use a knowably sound
calculation procedure to understand and discover mathematical
intricacies. This would mean that a normal Turing complete computer would not be able to ascertain certain mathematical truths that human minds can.
Pancomputationalism
Supporters
of CTM are faced with a simple yet important question whose answer has
proved elusive and controversial: what does it take for a physical
system (such as a mind, or an artificial computer) to perform
computations? In other words, under what conditions does a physical
system implement a computation? A very straightforward account is based
on a simple mapping between abstract mathematical computations and
physical systems: a system performs computation C if and only if there
is a mapping between a sequence of states individuated by C and a
sequence of states individuated by a physical description of the system.
Putnam (1988) and Searle (1992) argue that this simple mapping account (SMA) trivializes the empirical import of computational descriptions. As Putnam put it, “everything is a Probabilistic Automaton under some Description”. Even rocks, walls, and buckets of water—contrary to appearances—are computing systems. Gualtiero Piccinini
identifies different versions of Pancomputationalism, depending on how
many computations—all, some, or just one—they attribute to each system. Among these various versions, unlimited Pancomputationalism—the
view that every physical system performs every computation—is most
worrisome. Because if it is true, then the claim that a system S
performs a certain computation becomes trivially true and vacuous or
nearly so; it fails to distinguish S from anything else.
In response to the trivialization criticism, and to restrict SMA,
philosophers of mind have offered different accounts of computational
systems. These typically include causal account, semantic account,
syntactic account, and mechanistic account.
Causal account: a physical system S performs computation C just
in case (i) there is a mapping from the states ascribed to S by a
physical description to the states defined by computational description
C, such that (ii) the state transitions between the physical states
mirror the state transitions between the computational states.
Semantic account: In addition to the causal restriction imposed
by the causal account, the semantic account imposes a semantic
restriction. Only physical states that qualify as representations may be
mapped onto computational descriptions, thereby qualifying as
computational states. If a state is not representational, it is not
computational either.
Syntactic account: Instead of a semantic restriction, the
syntactic account imposes a syntactic restriction: only physical states
that qualify as syntactic may be mapped onto computational descriptions,
thereby qualifying as computational states. If a state lacks syntactic
structure, it is not computational.
Mechanistic account: First introduced by Gualtiero Piccinini in 2007,
the mechanistic account of computational systems accounts for concrete
computation in terms of the mechanistic properties of a system.
According to the mechanistic account, concrete computing systems are
functional mechanisms of a special kind—mechanisms that perform concrete
computations.
Prominent scholars
Daniel Dennett proposed the multiple drafts model,
in which consciousness seems linear but is actually blurry and gappy,
distributed over space and time in the brain. Consciousness is the
computation, there is no extra step or "Cartesian theater" in which you become conscious of the computation.
Jerry Fodor
argues that mental states, such as beliefs and desires, are relations
between individuals and mental representations. He maintains that these
representations can only be correctly explained in terms of a language
of thought (LOT) in the mind. Further, this language of thought itself
is codified in the brain, not just a useful explanatory tool. Fodor
adheres to a species of functionalism, maintaining that thinking and
other mental processes consist primarily of computations operating on
the syntax of the representations that make up the language of thought.
In later work (Concepts and The Elm and the Expert), Fodor
has refined and even questioned some of his original computationalist
views, and adopted a highly modified version of LOT (see LOT2).
David Marr
proposed that cognitive processes have three levels of description: the
computational level (which describes that computational problem (i.e.,
input/output mapping) computed by the cognitive process); the
algorithmic level (which presents the algorithm used for computing the
problem postulated at the computational level); and the implementational
level (which describes the physical implementation of the algorithm
postulated at the algorithmic level in biological matter, e.g. the
brain). (Marr 1981)
Ulric Neisser coined the term 'cognitive psychology' in his book published in 1967 (Cognitive Psychology),
wherein Neisser characterizes people as dynamic information-processing
systems whose mental operations might be described in computational
terms.
Steven Pinker described a "language instinct," an evolved, built-in capacity to learn language (if not writing).
Hilary Putnam proposed functionalism
to describe consciousness, asserting that it is the computation that
equates to consciousness, regardless of whether the computation is
operating in a brain, in a computer, or in a "brain in a vat."
Georges Rey, professor at the University of Maryland,
builds on Jerry Fodor's representational theory of mind to produce his
own version of a Computational/Representational Theory of Thought.
In philosophy of mind, functionalism
is the thesis that mental states (beliefs, desires, being in pain,
etc.) are constituted solely by their functional role, which means,
their causal relations with other mental states, sensory inputs and
behavioral outputs. Functionalism developed largely as an alternative to the identity theory of mind and behaviorism.
Functionalism is a theoretical level between the physical implementation and behavioral output. Therefore, it is different from its predecessors of Cartesian dualism (advocating independent mental and physical substances) and Skinnerian behaviorism and physicalism
(declaring only physical substances) because it is only concerned with
the effective functions of the brain, through its organization or its
"software programs".
Since mental states are identified by a functional role, they are
said to be realized on multiple levels; in other words, they are able
to be manifested in various systems, even perhaps computers, so long as
the system performs the appropriate functions. While computers are
physical devices with electronic substrate that perform computations
on inputs to give outputs, so brains are physical devices with neural
substrate that perform computations on inputs which produce behaviors.
Multiple realizability
An important part of some arguments for functionalism is the idea of multiple realizability.
According to standard functionalist theories, mental states correspond
to functional roles. They are like valves; a valve can be made of
plastic or metal or other materials, as long as it performs the proper
function (controlling the flow of a liquid or gas). Similarly,
functionalists argue, mental states can be explained without considering
the states of the underlying physical medium (such as the brain) that
realizes them; one need only consider higher-level functions. Because
mental states are not limited to a particular medium, they can be
realized in multiple ways, including, theoretically, within
non-biological systems, such as computers. A silicon-based machine could
have the same sort of mental life that a human being has, provided that
its structure realized the proper functional roles.
However, there have been some functionalist theories that combine
with the identity theory of mind, which deny multiple realizability.
Such Functional Specification Theories (FSTs) (Levin, § 3.4), as they are called, were most notably developed by David Lewis and David Malet Armstrong.
According to FSTs, mental states are the particular "realizers" of the
functional role, not the functional role itself. The mental state of
belief, for example, just is whatever brain or neurological process that
realizes the appropriate belief function. Thus, unlike standard
versions of functionalism (often called Functional State Identity Theories),
FSTs do not allow for the multiple realizability of mental states,
because the fact that mental states are realized by brain states is
essential. What often drives this view is the belief that if we were to
encounter an alien race with a cognitive system composed of
significantly different material from humans' (e.g., silicon-based) but
performed the same functions as human mental states (for example, they
tend to yell "Ouch!" when poked with sharp objects), we would say that
their type of mental state might be similar to ours but it is not the
same. For some, this may be a disadvantage to FSTs. Indeed, one of Hilary Putnam's
arguments for his version of functionalism relied on the intuition that
such alien creatures would have the same mental states as humans do,
and that the multiple realizability of standard functionalism makes it a
better theory of mind.
The broad position of "functionalism" can be articulated in many
different varieties. The first formulation of a functionalist theory of
mind was put forth by Hilary Putnam in the 1960s. This formulation, which is now called machine-state functionalism, or just machine functionalism, was inspired by the analogies which Putnam and others noted between the mind and the theoretical "machines" or computers capable of computing any given algorithm which were developed by Alan Turing (called Turing machines).
Putnam himself, by the mid-1970s, had begun questioning this position.
The beginning of his opposition to machine-state functionalism can be
read about in his Twin Earth thought experiment.
In non-technical terms, a Turing machine is not a physical
object, but rather an abstract machine built upon a mathematical model.
Typically, a Turing Machine has a horizontal tape divided into
rectangular cells arranged from left to right. The tape itself is
infinite in length, and each cell may contain a symbol. The symbols used
for any given "machine" can vary. The machine has a read-write head
that scans cells and moves in left and right directions. The action of
the machine is determined by the symbol in the cell being scanned and a
table of transition rules that serve as the machine's programming.
Because of the infinite tape, a traditional Turing Machine has an
infinite amount of time to compute any particular function or any number
of functions. In the below example, each cell is either blank (B) or has a 1 written on it. These are the inputs to the machine. The possible outputs are:
Halt: Do nothing.
R: move one square to the right.
L: move one square to the left.
B: erase whatever is on the square.
1: erase whatever is on the square and print a '1.
An extremely simple example of a Turing machine which
writes out the sequence '111' after scanning three blank squares and then stops as specified by the following machine table:
State One
State Two
State Three
B
write 1; stay in state 1
write 1; stay in state 2
write 1; stay in state 3
1
go right; go to state 2
go right; go to state 3
[halt]
This table states that if the machine is in state one and scans a blank square (B), it will print a 1 and remain in state one. If it is in state one and reads a 1, it will move one square to the right and also go into state two. If it is in state two and reads a B, it will print a 1 and stay in state two. If it is in state two and reads a 1, it will move one square to the right and go into state three. If it is in state three and reads a B, it prints a 1 and remains in state three. Finally, if it is in state three and reads a 1, then it will stay in state three.
The essential point to consider here is the nature of the states
of the Turing machine. Each state can be defined exclusively in terms
of its relations to the other states as well as inputs and outputs.
State one, for example, is simply the state in which the machine, if it
reads a B, writes a 1 and stays in that state, and in which, if it reads a 1,
it moves one square to the right and goes into a different state. This
is the functional definition of state one; it is its causal role in the
overall system. The details of how it accomplishes what it accomplishes
and of its material constitution are completely irrelevant.
The above point is critical to an understanding of machine-state
functionalism. Since Turing machines are not required to be physical
systems, "anything capable of going through a succession of states in
time can be a Turing machine".
Because biological organisms “go through a succession of states in
time”, any such organisms could also be equivalent to Turing machines.
According to machine-state functionalism, the nature of a mental
state is just like the nature of the Turing machine states described
above. If one can show the rational functioning and computing skills of
these machines to be comparable to the rational functioning and
computing skills of human beings, it follows that Turing machine
behavior closely resembles that of human beings.
Therefore, it is not a particular physical-chemical composition
responsible for the particular machine or mental state, it is the
programming rules which produce the effects that are responsible. To put
it another way, any rational preference is due to the rules being
followed, not to the specific material composition of the agent.
Psycho-functionalism
A second form of functionalism is based on the rejection of behaviorist
theories in psychology and their replacement with empirical cognitive
models of the mind. This view is most closely associated with Jerry Fodor and Zenon Pylyshyn and has been labeled psycho-functionalism.
The fundamental idea of psycho-functionalism is that psychology
is an irreducibly complex science and that the terms that we use to
describe the entities and properties of the mind in our best
psychological theories cannot be redefined in terms of simple behavioral
dispositions, and further, that such a redefinition would not be
desirable or salient were it achievable. Psychofunctionalists view
psychology as employing the same sorts of irreducibly teleological
or purposive explanations as the biological sciences. Thus, for
example, the function or role of the heart is to pump blood, that of the
kidney is to filter it and to maintain certain chemical balances and so
on—this is what accounts for the purposes of scientific explanation and
taxonomy. There may be an infinite variety of physical realizations for
all of the mechanisms, but what is important is only their role in the
overall biological theory. In an analogous manner, the role of mental
states, such as belief and desire, is determined by the functional or
causal role that is designated for them within our best scientific psychological theory. If some mental state which is postulated by folk psychology
(e.g. hysteria) is determined not to have any fundamental role in
cognitive psychological explanation, then that particular state may be
considered not to exist .
On the other hand, if it turns out that there are states which
theoretical cognitive psychology posits as necessary for explanation of
human behavior but which are not foreseen by ordinary folk psychological
language, then these entities or states exist.
Analytic functionalism
A
third form of functionalism is concerned with the meanings of
theoretical terms in general. This view is most closely associated with David Lewis and is often referred to as analytic functionalism or conceptual functionalism.
The basic idea of analytic functionalism is that theoretical terms are
implicitly defined by the theories in whose formulation they occur and
not by intrinsic properties of the phonemes they comprise. In the case
of ordinary language terms, such as "belief", "desire", or "hunger", the
idea is that such terms get their meanings from our common-sense "folk
psychological" theories about them, but that such conceptualizations are
not sufficient to withstand the rigor imposed by materialistic theories
of reality and causality. Such terms are subject to conceptual analyses
which take something like the following form:
Mental state M is the state that is preconceived by P and causes Q.
For example, the state of pain is caused by sitting on a tack and causes
loud cries, and higher order mental states of anger and resentment
directed at the careless person who left a tack lying around. These
sorts of functional definitions in terms of causal roles are claimed to
be analytic and a priori
truths about the submental states and the (largely fictitious)
propositional attitudes they describe. Hence, its proponents are known
as analytic or conceptual functionalists. The essential
difference between analytic and psychofunctionalism is that the latter
emphasizes the importance of laboratory observation and experimentation
in the determination of which mental state terms and concepts are
genuine and which functional identifications may be considered to be
genuinely contingent and a posteriori identities. The former, on the other hand, claims that such identities are necessary and not subject to empirical scientific investigation.
Homuncular functionalism
Homuncular functionalism was developed largely by Daniel Dennett and has been advocated by William Lycan. It arose in response to the challenges that Ned Block's China Brain (a.k.a. Chinese nation) and John Searle's Chinese room
thought experiments presented for the more traditional forms of
functionalism (see below under "Criticism"). In attempting to overcome
the conceptual difficulties that arose from the idea of a nation full of
Chinese people wired together, each person working as a single neuron
to produce in the wired-together whole the functional mental states of
an individual mind, many functionalists simply bit the bullet, so to
speak, and argued that such a Chinese nation would indeed possess all of
the qualitative and intentional properties of a mind; i.e. it would
become a sort of systemic or collective mind with propositional
attitudes and other mental characteristics. Whatever the worth of this
latter hypothesis, it was immediately objected that it entailed an unacceptable sort of mind-mind supervenience: the systemic
mind which somehow emerged at the higher-level must necessarily
supervene on the individual minds of each individual member of the
Chinese nation, to stick to Block's formulation. But this would seem to
put into serious doubt, if not directly contradict, the fundamental idea
of the supervenience thesis: there can be no change in the mental realm
without some change in the underlying physical substratum. This can be
easily seen if we label the set of mental facts that occur at the higher-level M1 and the set of mental facts that occur at the lower-level M2. Given the transitivity of supervenience, if M1 supervenes on M2, and M2 supervenes on P (physical base), then M1 and M2 both supervene on P, even though they are (allegedly) totally different sets of mental facts.
Since mind-mind supervenience seemed to have become acceptable in
functionalist circles, it seemed to some that the only way to resolve
the puzzle was to postulate the existence of an entire hierarchical
series of mind levels (analogous to homunculi)
which became less and less sophisticated in terms of functional
organization and physical composition all the way down to the level of
the physico-mechanical neuron or group of neurons. The homunculi at each
level, on this view, have authentic mental properties but become
simpler and less intelligent as one works one's way down the hierarchy.
Mechanistic functionalism
Mechanistic functionalism, originally formulated and defended by Gualtiero Piccinini and Carl Gillett
independently, augments previous functionalist accounts of mental
states by maintaining that any psychological explanation must be
rendered in mechanistic terms. That is, instead of mental states
receiving a purely functional explanation in terms of their relations to
other mental states, like those listed above, functions are seen as
playing only a part—the other part being played by structures— of the
explanation of a given mental state.
A mechanistic explanation
involves decomposing a given system, in this case a mental system, into
its component physical parts, their activities or functions, and their
combined organizational relations.
On this account the mind remains a functional system, but one that is
understood in mechanistic terms. This account remains a sort of
functionalism because functional relations are still essential to mental states,
but it is mechanistic because the functional relations are always
manifestations of concrete structures—albeit structures understood at a
certain level of abstraction. Functions are individuated and explained
either in terms of the contributions they make to the given system or in teleological terms. If the functions are understood in teleological terms, then they may be characterized either etiologically or non-etiologically.
Mechanistic functionalism leads functionalism away from the
traditional functionalist autonomy of psychology from neuroscience and
towards integrating psychology and neuroscience.
By providing an applicable framework for merging traditional
psychological models with neurological data, mechanistic functionalism
may be understood as reconciling the functionalist theory of mind with
neurological accounts of how the brain actually works. This is due to
the fact that mechanistic explanations of function attempt to provide an
account of how functional states (mental states) are physically
realized through neurological mechanisms.
Physicalism
There
is much confusion about the sort of relationship that is claimed to
exist (or not exist) between the general thesis of functionalism and physicalism. It has often been claimed that functionalism somehow "disproves" or falsifies physicalism tout court
(i.e. without further explanation or description). On the other hand,
most philosophers of mind who are functionalists claim to be
physicalists—indeed, some of them, such as David Lewis, have claimed to
be strict reductionist-type physicalists.
Functionalism is fundamentally what Ned Block has called a broadly metaphysical thesis as opposed to a narrowly ontological one. That is, functionalism is not so much concerned with what there is
than with what it is that characterizes a certain type of mental state,
e.g. pain, as the type of state that it is.
Previous attempts to answer
the mind-body problem have all tried to resolve it by answering both
questions: dualism says there are two substances and that mental states
are characterized by their immateriality; behaviorism claimed that
there was one substance and that mental states were behavioral
disposition; physicalism asserted the existence of just one substance
and characterized the mental states as physical states (as in "pain =
C-fiber firings").
On this understanding, type physicalism can be seen as
incompatible with functionalism, since it claims that what characterizes
mental states (e.g. pain) is that they are physical in nature, while
functionalism says that what characterizes pain is its functional/causal
role and its relationship with yelling "ouch", etc. However, any weaker
sort of physicalism which makes the simple ontological claim that
everything that exists is made up of physical matter is perfectly
compatible with functionalism. Moreover, most functionalists who are
physicalists require that the properties that are quantified over in
functional definitions be physical properties. Hence, they are physicalists, even though the general thesis of functionalism itself does not commit them to being so.
In the case of David Lewis, there is a distinction in the concepts of "having pain" (a rigid designator
true of the same things in all possible worlds) and just "pain" (a
non-rigid designator). Pain, for Lewis, stands for something like the
definite description "the state with the causal role x". The referent of
the description in humans is a type of brain state to be determined by
science. The referent among silicon-based life forms is something else.
The referent of the description among angels is some immaterial,
non-physical state. For Lewis, therefore, local type-physical reductions are possible and compatible with conceptual functionalism. (See also Lewis's mad pain and Martian pain.) There seems to be some confusion between types and tokens that needs to be cleared up in the functionalist analysis.
Criticism
China brain
Ned Block argues against the functionalist proposal of multiple realizability,
where hardware implementation is irrelevant because only the functional
level is important. The "China brain" or "Chinese nation" thought
experiment involves supposing that the entire nation of China
systematically organizes itself to operate just like a brain, with each
individual acting as a neuron. (The tremendous difference in speed of
operation of each unit is not addressed.). According to functionalism,
so long as the people are performing the proper functional roles, with
the proper causal relations between inputs and outputs, the system will
be a real mind, with mental states, consciousness, and so on. However,
Block argues, this is patently absurd, so there must be something wrong
with the thesis of functionalism since it would allow this to be a
legitimate description of a mind.
Some functionalists believe China would have qualia but that due to the size it is impossible to imagine China being conscious. Indeed, it may be the case that we are constrained by our theory of mind
and will never be able to understand what Chinese-nation consciousness
is like. Therefore, if functionalism is true either qualia will exist
across all hardware or will not exist at all but are illusory.
The Chinese room
The Chinese room argument by John Searle
is a direct attack on the claim that thought can be represented as a
set of functions. The thought experiment asserts that it is possible to
mimic intelligent action without any interpretation or understanding
through the use of a purely functional system. In short, Searle
describes a person who only speaks English who is in a room with only
Chinese symbols in baskets and a rule book in English for moving the
symbols around. The person is then ordered by people outside of the
room to follow the rule book for sending certain symbols out of the room
when given certain symbols. Further suppose that the people outside of
the room are Chinese speakers and are communicating with the person
inside via the Chinese symbols. According to Searle, it would be absurd
to claim that the English speaker inside knows Chinese simply based on
these syntactic processes. This thought experiment attempts to show
that systems which operate merely on syntactic processes (inputs and
outputs, based on algorithms) cannot realize any semantics (meaning) or
intentionality (aboutness). Thus, Searle attacks the idea that thought
can be equated with following a set of syntactic rules; that is,
functionalism is an insufficient theory of the mind.
In connection with Block's Chinese nation, many functionalists responded to Searle's thought experiment
by suggesting that there was a form of mental activity going on at a
higher level than the man in the Chinese room could comprehend (the
so-called "system reply"); that is, the system does know Chinese. Of
course, Searle responds that there is nothing more than syntax going on
at the higher-level as well, so this reply is subject to the same
initial problems. Furthermore, Searle suggests the man in the room could
simply memorize the rules and symbol relations. Again, though he would
convincingly mimic communication, he would be aware only of the symbols
and rules, not of the meaning behind them.
Inverted spectrum
Another main criticism of functionalism is the inverted spectrum or inverted qualia scenario, most specifically proposed as an objection to functionalism by Ned Block.
This thought experiment involves supposing that there is a person, call
her Jane, that is born with a condition which makes her see the
opposite spectrum of light that is normally perceived. Unlike normal
people, Jane sees the color violet as yellow, orange as blue, and so
forth. So, suppose, for example, that you and Jane are looking at the
same orange. While you perceive the fruit as colored orange, Jane sees
it as colored blue. However, when asked what color the piece of fruit
is, both you and Jane will report "orange". In fact, one can see that
all of your behavioral as well as functional relations to colors will be
the same. Jane will, for example, properly obey traffic signs just as
any other person would, even though this involves the color perception.
Therefore, the argument goes, since there can be two people who are
functionally identical, yet have different mental states (differing in
their qualitative or phenomenological aspects), functionalism is not
robust enough to explain individual differences in qualia.
David Chalmers tries to show that even though mental content cannot be fully accounted for in functional terms, there is nevertheless a nomological correlation
between mental states and functional states in this world. A
silicon-based robot, for example, whose functional profile matched our
own, would have to be fully conscious. His argument for this claim takes the form of a reductio ad absurdum.
The general idea is that since it would be very unlikely for a
conscious human being to experience a change in its qualia which it
utterly fails to notice, mental content and functional profile appear to
be inextricably bound together, at least in the human case. If the
subject's qualia were to change, we would expect the subject to notice,
and therefore his functional profile to follow suit. A similar argument
is applied to the notion of absent qualia.
In this case, Chalmers argues that it would be very unlikely for a
subject to experience a fading of his qualia which he fails to notice
and respond to. This, coupled with the independent assertion that a
conscious being's functional profile just could be maintained,
irrespective of its experiential state, leads to the conclusion that the
subject of these experiments would remain fully conscious. The problem
with this argument, however, as Brian G. Crabb (2005) has observed, is
that, while changing or fading qualia in a conscious subject might force
changes in its functional profile, this tells us nothing about the case
of a permanently inverted or unconscious robot. A subject with inverted
qualia from birth would have nothing to notice or adjust to. Similarly,
an unconscious functional simulacrum of ourselves (a zombie) would have
no experiential changes to notice or adjust to. Consequently, Crabb
argues, Chalmers' "fading qualia" and "dancing qualia" arguments fail to
establish that cases of permanently inverted or absent qualia are
nomologically impossible.
A related critique of the inverted spectrum argument is that it
assumes that mental states (differing in their qualitative or
phenomenological aspects) can be independent of the functional relations
in the brain. Thus, it begs the question
of functional mental states: its assumption denies the possibility of
functionalism itself, without offering any independent justification for
doing so. (Functionalism says that mental states are produced by the
functional relations in the brain.) This same type of problem—that there
is no argument, just an antithetical assumption at their base—can also
be said of both the Chinese room and the Chinese nation arguments.
Notice, however, that Crabb's response to Chalmers does not commit this
fallacy: His point is the more restricted observation that even if
inverted or absent qualia turn out to be nomologically impossible, and
it is perfectly possible that we might subsequently discover this fact
by other means, Chalmers' argument fails to demonstrate that they are
impossible.
Twin Earth
The Twin Earth thought experiment, introduced by Hilary Putnam,
is responsible for one of the main arguments used against
functionalism, although it was originally intended as an argument
against semantic internalism.
The thought experiment is simple and runs as follows. Imagine a Twin
Earth which is identical to Earth in every way but one: water does not
have the chemical structure H₂O, but rather some other structure, say
XYZ. It is critical, however, to note that XYZ on Twin Earth is still
called "water" and exhibits all the same macro-level properties that H₂O
exhibits on Earth (i.e., XYZ is also a clear drinkable liquid that is
in lakes, rivers, and so on). Since these worlds are identical in every
way except in the underlying chemical structure of water, you and your
Twin Earth doppelgänger
see exactly the same things, meet exactly the same people, have exactly
the same jobs, behave exactly the same way, and so on. In other words,
since you share the same inputs, outputs, and relations between other
mental states, you are functional duplicates. So, for example, you both
believe that water is wet. However, the content of your mental state of
believing that water is wet differs from your duplicate's because your
belief is of H₂O, while your duplicate's is of XYZ. Therefore, so the
argument goes, since two people can be functionally identical, yet have
different mental states, functionalism cannot sufficiently account for
all mental states.
Most defenders of functionalism initially responded to this
argument by attempting to maintain a sharp distinction between internal
and external content. The internal contents of propositional attitudes,
for example, would consist exclusively in those aspects of them which
have no relation with the external world and which bear the
necessary functional/causal properties that allow for relations with
other internal mental states. Since no one has yet been able to
formulate a clear basis or justification for the existence of such a
distinction in mental contents, however, this idea has generally been
abandoned in favor of externalist causal theories of mental contents (also known as informational semantics). Such a position is represented, for example, by Jerry Fodor's
account of an "asymmetric causal theory" of mental content. This view
simply entails the modification of functionalism to include within its
scope a very broad interpretation of input and outputs to include the
objects that are the causes of mental representations in the external
world.
The twin earth argument hinges on the assumption that experience
with an imitation water would cause a different mental state than
experience with natural water. However, since no one would notice the
difference between the two waters, this assumption is likely false.
Further, this basic assumption is directly antithetical to
functionalism; and, thereby, the twin earth argument does not constitute
a genuine argument: as this assumption entails a flat denial of
functionalism itself (which would say that the two waters would not
produce different mental states, because the functional relationships
would remain unchanged).
Meaning holism
Another common criticism of functionalism is that it implies a radical form of semantic holism. Block and Fodor referred to this as the damn/darn problem.
The difference between saying "damn" or "darn" when one smashes one's
finger with a hammer can be mentally significant. But since these
outputs are, according to functionalism, related to many (if not all)
internal mental states, two people who experience the same pain and
react with different outputs must share little (perhaps nothing) in
common in any of their mental states. But this is counterintuitive; it
seems clear that two people share something significant in their mental
states of being in pain if they both smash their finger with a hammer,
whether or not they utter the same word when they cry out in pain.
Another possible solution to this problem is to adopt a moderate
(or molecularist) form of holism. But even if this succeeds in the case
of pain, in the case of beliefs and meaning, it faces the difficulty of
formulating a distinction between relevant and non-relevant contents
(which can be difficult to do without invoking an analytic–synthetic distinction, as many seek to avoid).
Triviality arguments
According
to Ned Block, if functionalism is to avoid the chauvinism of
type-physicalism, it becomes overly liberal in "ascribing mental
properties to things that do not in fact have them".
As an example, he proposes that the economy of Bolivia might be
organized such that the economic states, inputs, and outputs would be
isomorphic to a person under some bizarre mapping from mental to
economic variables.
Hilary Putnam, John Searle, and others
have offered further arguments that functionalism is trivial, i.e. that
the internal structures functionalism tries to discuss turn out to be
present everywhere, so that either functionalism turns out to reduce to behaviorism, or to complete triviality and therefore a form of panpsychism.
These arguments typically use the assumption that physics leads to a
progression of unique states, and that functionalist realization is
present whenever there is a mapping from the proposed set of mental
states to physical states of the system. Given that the states of a
physical system are always at least slightly unique, such a mapping will
always exist, so any system is a mind. Formulations of functionalism
which stipulate absolute requirements on interaction with external
objects (external to the functional account, meaning not defined
functionally) are reduced to behaviorism instead of absolute triviality,
because the input-output behavior is still required.
Peter Godfrey-Smith has argued further
that such formulations can still be reduced to triviality if they
accept a somewhat innocent-seeming additional assumption. The assumption
is that adding a transducer layer, that is, an input-output
system, to an object should not change whether that object has mental
states. The transducer layer is restricted to producing behavior
according to a simple mapping, such as a lookup table, from inputs to
actions on the system, and from the state of the system to outputs.
However, since the system will be in unique states at each moment and at
each possible input, such a mapping will always exist so there will be a
transducer layer which will produce whatever physical behavior is
desired.
Godfrey-Smith believes that these problems can be addressed using causality,
but that it may be necessary to posit a continuum between objects being
minds and not being minds rather than an absolute distinction.
Furthermore, constraining the mappings seems to require either
consideration of the external behavior as in behaviorism, or discussion
of the internal structure of the realization as in identity theory; and
though multiple realizability does not seem to be lost, the
functionalist claim of the autonomy of high-level functional description
becomes questionable.
The AI effect occurs when onlookers discount the behavior of an artificial intelligence program by arguing that it is not real intelligence.
Author Pamela McCorduck
writes: "It's part of the history of the field of artificial
intelligence that every time somebody figured out how to make a computer
do something—play good checkers, solve simple but relatively informal
problems—there was a chorus of critics to say, 'that's not thinking'." Researcher Rodney Brooks complains: "Every time we figure out a piece of it, it stops being magical; we say, 'Oh, that's just a computation.'"
"The AI effect" tries to redefine AI to mean: AI is anything that has not been done yet
A
view taken by some people trying to promulgate the AI effect is: As
soon as AI successfully solves a problem, the problem is no longer a
part of AI.
Pamela McCorduck calls it an "odd paradox"
that "practical AI successes, computational programs that actually
achieved intelligent behavior, were soon assimilated into whatever
application domain they were found to be useful in, and became silent
partners alongside other problem-solving approaches, which left AI
researchers to deal only with the "failures", the tough nuts that
couldn't yet be cracked."
When IBM's chess playing computer Deep Blue succeeded in defeating Garry Kasparov in 1997, people complained that it had only used "brute force methods" and it wasn't real intelligence. Fred Reed writes:
"A
problem that proponents of AI regularly face is this: When we know how a
machine does something 'intelligent,' it ceases to be regarded as
intelligent. If I beat the world's chess champion, I'd be regarded as
highly bright."
When problems have not yet been formalised, they can still be characterised by a model of computation that includes human computation.
The computational burden of a problem is split between a computer and a
human: one part is solved by computer and the other part solved by a
human. This formalisation is referred to as human-assisted Turing machine.
AI applications become mainstream
Software
and algorithms developed by AI researchers are now integrated into many
applications throughout the world, without really being called AI.
Michael Swaine
reports "AI advances are not trumpeted as artificial intelligence so
much these days, but are often seen as advances in some other field".
"AI has become more important as it has become less conspicuous", Patrick Winston
says. "These days, it is hard to find a big system that does not work,
in part, because of ideas developed or matured in the AI world."
According to Stottler Henke, "The great practical benefits of AI
applications and even the existence of AI in many software products go
largely unnoticed by many despite the already widespread use of AI
techniques in software. This is the AI effect. Many marketing people
don't use the term 'artificial intelligence' even when their company's
products rely on some AI techniques. Why not?"
Marvin Minsky
writes "This paradox resulted from the fact that whenever an AI
research project made a useful new discovery, that product usually
quickly spun off to form a new scientific or commercial specialty with
its own distinctive name. These changes in name led outsiders to ask,
Why do we see so little progress in the central field of artificial
intelligence?"
Nick Bostrom
observes that "A lot of cutting edge AI has filtered into general
applications, often without being called AI because once something
becomes useful enough and common enough it's not labelled AI anymore."
Legacy of the AI winter
Many AI researchers find that they can procure more funding and sell
more software if they avoid the tarnished name of "artificial
intelligence" and instead pretend their work has nothing to do with
intelligence at all. This was especially true in the early 1990s, during
the second "AI winter".
Patty Tascarella writes "Some believe the word 'robotics' actually carries a stigma that hurts a company's chances at funding"
Saving a place for humanity at the top of the chain of being
Michael Kearns suggests that "people subconsciously are trying to preserve for themselves some special role in the universe".
By discounting artificial intelligence people can continue to feel
unique and special. Kearns argues that the change in perception known as
the AI effect can be traced to the mystery being removed from
the system. In being able to trace the cause of events implies that it's
a form of automation rather than intelligence.
A related effect has been noted in the history of animal cognition and in consciousness studies, where every time a capacity formerly thought as uniquely human is discovered in animals, (e.g. the ability to make tools, or passing the mirror test), the overall importance of that capacity is deprecated.
Herbert A. Simon,
when asked about the lack of AI's press coverage at the time, said,
"What made AI different was that the very idea of it arouses a real fear
and hostility in some human breasts. So you are getting very strong
emotional reactions. But that's okay. We'll live with that."
In the history of artificial intelligence, an AI winter is a period of reduced funding and interest in artificial intelligence research. The term was coined by analogy to the idea of a nuclear winter. The field has experienced several hype cycles, followed by disappointment and criticism, followed by funding cuts, followed by renewed interest years or decades later.
The term first appeared in 1984 as the topic of a public debate at the annual meeting of AAAI
(then called the "American Association of Artificial Intelligence"). It
is a chain reaction that begins with pessimism in the AI community,
followed by pessimism in the press, followed by a severe cutback in
funding, followed by the end of serious research. At the meeting, Roger Schank and Marvin Minsky—two
leading AI researchers who had survived the "winter" of the
1970s—warned the business community that enthusiasm for AI had spiraled
out of control in the 1980s and that disappointment would certainly
follow. Three years later, the billion-dollar AI industry began to
collapse.
Hype is common in many emerging technologies, such as the railway mania or the dot-com bubble.
The AI winter was a result of such hype, due to over-inflated promises
by developers, unnaturally high expectations from end-users, and
extensive promotion in the media. Despite the rise and fall of AI's reputation, it has continued to develop new and successful technologies. AI researcher Rodney Brooks
would complain in 2002 that "there's this stupid myth out there that AI
has failed, but AI is around you every second of the day." In 2005, Ray Kurzweil
agreed: "Many observers still think that the AI winter was the end of
the story and that nothing since has come of the AI field. Yet today
many thousands of AI applications are deeply embedded in the
infrastructure of every industry."
Enthusiasm and optimism about AI has generally increased since
its low point in the early 1990s. Beginning about 2012, interest in
artificial intelligence (and especially the sub-field of machine learning) from the research and corporate communities led to a dramatic increase in funding and investment.
Overview
There were two major winters in 1974–1980 and 1987–1993 and several smaller episodes, including the following:
During the Cold War,
the US government was particularly interested in the automatic, instant
translation of Russian documents and scientific reports. The government
aggressively supported efforts at machine translation starting in 1954.
At the outset, the researchers were optimistic. Noam Chomsky's new work in grammar was streamlining the translation process and there were "many predictions of imminent 'breakthroughs'".
However, researchers had underestimated the profound difficulty of word-sense disambiguation.
In order to translate a sentence, a machine needed to have some idea
what the sentence was about, otherwise it made mistakes. An apocryphal
example is "the spirit is willing but the flesh is weak." Translated
back and forth with Russian, it became "the vodka is good but the meat
is rotten." Similarly, "out of sight, out of mind" became "blind idiot". Later researchers would call this the commonsense knowledge problem.
By 1964, the National Research Council had become concerned about the lack of progress and formed the Automatic Language Processing Advisory Committee (ALPAC)
to look into the problem. They concluded, in a famous 1966 report, that
machine translation was more expensive, less accurate and slower than
human translation. After spending some 20 million dollars, the NRC ended
all support. Careers were destroyed and research ended.
Machine translation is still an open research problem in the 21st century, which has met with some success.
The abandonment of connectionism in 1969
Some of the earliest work in AI used networks or circuits of
connected units to simulate intelligent behavior. Examples of this kind
of work, called "connectionism", include Walter Pitts and Warren McCullough's first description of a neural network for logic and Marvin Minsky's work on the SNARC system. In the late 1950s, most of these approaches were abandoned when researchers began to explore symbolic reasoning as the essence of intelligence, following the success of programs like the Logic Theorist and the General Problem Solver.
However, one type of connectionist work continued: the study of perceptrons, invented by Frank Rosenblatt, who kept the field alive with his salesmanship and the sheer force of his personality.
He optimistically predicted that the perceptron "may eventually be able to learn, make decisions, and translate languages".
Mainstream research into perceptrons came to an abrupt end in 1969, when Marvin Minsky and Seymour Papert published the book Perceptrons, which was perceived as outlining the limits of what perceptrons could do.
Connectionist approaches were abandoned for the next decade or so. While important work, such as Paul Werbos' discovery of backpropagation, continued in a limited way, major funding for connectionist projects was difficult to find in the 1970s and early 1980s.
The "winter" of connectionist research came to an end in the middle 1980s, when the work of John Hopfield, David Rumelhart and others revived large scale interest in neural networks. Rosenblatt did not live to see this, however, as he died in a boating accident shortly after Perceptrons was published.
The setbacks of 1974
The Lighthill report
In 1973, professor Sir James Lighthill was asked by the UK Parliament
to evaluate the state of AI research in the United Kingdom. His report,
now called the Lighthill report, criticized the utter failure of AI to
achieve its "grandiose objectives." He concluded that nothing being done
in AI couldn't be done in other sciences. He specifically mentioned the
problem of "combinatorial explosion" or "intractability",
which implied that many of AI's most successful algorithms would grind
to a halt on real world problems and were only suitable for solving
"toy" versions.
The report was contested in a debate broadcast in the BBC
"Controversy" series in 1973. The debate "The general purpose robot is a
mirage" from the Royal Institution was Lighthill versus the team of Donald Michie, John McCarthy and Richard Gregory. McCarthy later wrote that "the combinatorial explosion problem has been recognized in AI from the beginning".
The report led to the complete dismantling of AI research in England.
AI research continued in only a few universities (Edinburgh, Essex and
Sussex). Research would not revive on a large scale until 1983, when Alvey
(a research project of the British Government) began to fund AI again
from a war chest of £350 million in response to the Japanese Fifth
Generation Project (see below). Alvey had a number of UK-only
requirements which did not sit well internationally, especially with US
partners, and lost Phase 2 funding.
DARPA's early 1970s funding cuts
During the 1960s, the Defense Advanced Research Projects Agency (then known as "ARPA", now known as "DARPA") provided millions of dollars for AI research with few strings attached. J. C. R. Licklider, the founding director of DARPA's computing division, believed in "funding people, not projects" and he and several successors allowed AI's leaders (such as Marvin Minsky, John McCarthy, Herbert A. Simon or Allen Newell) to spend it almost any way they liked.
This attitude changed after the passage of Mansfield Amendment in 1969, which required DARPA to fund "mission-oriented direct research, rather than basic undirected research".
Pure undirected research of the kind that had gone on in the 1960s
would no longer be funded by DARPA. Researchers now had to show that
their work would soon produce some useful military technology. AI
research proposals were held to a very high standard. The situation was
not helped when the Lighthill report and DARPA's own study (the American Study Group)
suggested that most AI research was unlikely to produce anything truly
useful in the foreseeable future. DARPA's money was directed at specific
projects with identifiable goals, such as autonomous tanks and battle
management systems. By 1974, funding for AI projects was hard to find.
AI researcher Hans Moravec
blamed the crisis on the unrealistic predictions of his colleagues:
"Many researchers were caught up in a web of increasing exaggeration.
Their initial promises to DARPA had been much too optimistic. Of course,
what they delivered stopped considerably short of that. But they felt
they couldn't in their next proposal promise less than in the first one,
so they promised more."
The result, Moravec claims, is that some of the staff at DARPA had lost
patience with AI research. "It was literally phrased at DARPA that
'some of these people were going to be taught a lesson [by] having their
two-million-dollar-a-year contracts cut to almost nothing!'" Moravec
told Daniel Crevier.
While the autonomous tank project was a failure, the battle management system (the Dynamic Analysis and Replanning Tool) proved to be enormously successful, saving billions in the first Gulf War, repaying all of DARPAs investment in AI and justifying DARPA's pragmatic policy.
The SUR debacle
DARPA
was deeply disappointed with researchers working on the Speech
Understanding Research program at Carnegie Mellon University. DARPA had
hoped for, and felt it had been promised, a system that could respond to
voice commands from a pilot. The SUR team had developed a system which
could recognize spoken English, but only if the words were spoken in a particular order. DARPA felt it had been duped and, in 1974, they cancelled a three million dollar a year contract.
Many years later, several successful commercial speech recognition systems would use the technology developed by the Carnegie Mellon team (such as hidden Markov models) and the market for speech recognition systems would reach $4 billion by 2001.
The setbacks of the late 1980s and early 1990s
The collapse of the LISP machine market
In the 1980s, a form of AI program called an "expert system" was adopted by corporations around the world. The first commercial expert system was XCON, developed at Carnegie Mellon for Digital Equipment Corporation,
and it was an enormous success: it was estimated to have saved the
company 40 million dollars over just six years of operation.
Corporations around the world began to develop and deploy expert systems
and by 1985 they were spending over a billion dollars on AI, most of it
to in-house AI departments. An industry grew up to support them,
including software companies like Teknowledge and Intellicorp (KEE), and hardware companies like Symbolics and LISP Machines Inc. who built specialized computers, called LISP machines, that were optimized to process the programming language LISP, the preferred language for AI.
In 1987, three years after Minsky and Schank's prediction, the market for specialized LISP-based AI hardware collapsed. Workstations by companies like Sun Microsystems
offered a powerful alternative to LISP machines and companies like
Lucid offered a LISP environment for this new class of workstations. The
performance of these general workstations became an increasingly
difficult challenge for LISP Machines. Companies like Lucid and Franz LISP
offered increasingly powerful versions of LISP that were portable to
all UNIX systems. For example, benchmarks were published showing
workstations maintaining a performance advantage over LISP machines. Later desktop computers built by Apple and IBM
would also offer a simpler and more popular architecture to run LISP
applications on. By 1987, some of them had become as powerful as the
more expensive LISP machines. The desktop computers had rule-based
engines such as CLIPS available.
These alternatives left consumers with no reason to buy an expensive
machine specialized for running LISP. An entire industry worth half a
billion dollars was replaced in a single year.
By the early 1990s, most commercial LISP companies had failed, including Symbolics, LISP Machines Inc., Lucid Inc., etc. Other companies, like Texas Instruments and Xerox,
abandoned the field. A small number of customer companies (that is,
companies using systems written in LISP and developed on LISP machine
platforms) continued to maintain systems. In some cases, this
maintenance involved the assumption of the resulting support work.
Slowdown in deployment of expert systems
By
the early 1990s, the earliest successful expert systems, such as XCON,
proved too expensive to maintain. They were difficult to update, they
could not learn, they were "brittle" (i.e., they could make grotesque
mistakes when given unusual inputs), and they fell prey to problems
(such as the qualification problem) that had been identified years earlier in research in nonmonotonic logic. Expert systems proved useful, but only in a few special contexts. Another problem dealt with the computational hardness of truth maintenance efforts for general knowledge. KEE used an assumption-based approach (see NASA, TEXSYS) supporting multiple-world scenarios that was difficult to understand and apply.
The few remaining expert system shell companies were eventually
forced to downsize and search for new markets and software paradigms,
like case-based reasoning or universal database access. The maturation of Common Lisp saved many systems such as ICAD which found application in knowledge-based engineering. Other systems, such as Intellicorp's KEE, moved from LISP to a C++ (variant) on the PC and helped establish object-oriented technology (including providing major support for the development of UML.
The end of the Fifth Generation project
In 1981, the Japanese Ministry of International Trade and Industry set aside $850 million for the Fifth Generation computer
project. Their objectives were to write programs and build machines
that could carry on conversations, translate languages, interpret
pictures, and reason like human beings. By 1991, the impressive list of
goals penned in 1981 had not been met. According to HP Newquist in The Brain Makers, "On June 1, 1992, The Fifth Generation Project ended not with a successful roar, but with a whimper." As with other AI projects, expectations had run much higher than what was actually possible.
Strategic Computing Initiative cutbacks
In 1983, in response to the fifth generation project, DARPA again
began to fund AI research through the Strategic Computing Initiative. As
originally proposed the project would begin with practical, achievable
goals, which even included artificial general intelligence as long-term
objective. The program was under the direction of the Information Processing Technology Office (IPTO) and was also directed at supercomputing and microelectronics.
By 1985 it had spent $100 million and 92 projects were underway at 60
institutions, half in industry, half in universities and government
labs. AI research was generously funded by the SCI.
Jack Schwarz, who ascended to the leadership of IPTO in 1987,
dismissed expert systems as "clever programming" and cut funding to AI
"deeply and brutally", "eviscerating" SCI. Schwarz felt that DARPA
should focus its funding only on those technologies which showed the
most promise, in his words, DARPA should "surf", rather than "dog
paddle", and he felt strongly AI was not "the next wave".
Insiders in the program cited problems in communication, organization
and integration. A few projects survived the funding cuts, including
pilot's assistant and an autonomous land vehicle (which were never
delivered) and the DART battle management system, which (as noted above)
was successful.
Developments post-AI winter
A survey of reports from the early 2000s suggests that AI's reputation was still less than stellar:
Alex Castro, quoted in The Economist, 7 June 2007:
"[Investors] were put off by the term 'voice recognition' which, like
'artificial intelligence', is associated with systems that have all too
often failed to live up to their promises."
Patty Tascarella in Pittsburgh Business Times, 2006: "Some believe the word 'robotics' actually carries a stigma that hurts a company's chances at funding."
John Markoff in the New York Times,
2005: "At its low point, some computer scientists and software
engineers avoided the term artificial intelligence for fear of being
viewed as wild-eyed dreamers."
In the late 1990s and early 21st century, AI technology became widely used as elements of larger systems, but the field is rarely credited for these successes. In 2006, Nick Bostrom
explained that "a lot of cutting edge AI has filtered into general
applications, often without being called AI because once something
becomes useful enough and common enough it's not labeled AI anymore." Rodney Brooks
stated around the same time that "there's this stupid myth out there
that AI has failed, but AI is around you every second of the day."
Technologies developed by AI researchers have achieved commercial success in a number of domains, such as machine translation, data mining, industrial robotics, logistics, speech recognition, banking software, medical diagnosis, and Google's search engine.
Fuzzy logic controllers have been developed for automatic gearboxes in automobiles (the 2006 Audi TT, VW Touareg and VW Caravelle feature the DSP transmission which utilizes fuzzy logic, a number of Škoda variants (Škoda Fabia) also currently include a fuzzy logic-based controller). Camera sensors widely utilize fuzzy logic to enable focus.
Heuristic search and data analytics are both technologies that have developed from the evolutionary computing
and machine learning subdivision of the AI research community. Again,
these techniques have been applied to a wide range of real world
problems with considerable commercial success.
Data analytics technology utilizing algorithms for the automated
formation of classifiers that were developed in the supervised machine
learning community in the 1990s (for example, TDIDT, Support Vector
Machines, Neural Nets, IBL) are now used pervasively by companies for marketing survey targeting and discovery of trends and features in data sets.
AI funding
Researchers
and economists frequently judged the status of an AI winter by
reviewing which AI projects were being funded, how much and by whom.
Trends in funding are often set by major funding agencies in the
developed world. Currently, DARPA and a civilian funding program called EU-FP7 provide much of the funding for AI research in the US and European Union.
As of 2007, DARPA was soliciting AI research proposals under a number of programs including The Grand Challenge Program, Cognitive Technology Threat Warning System (CT2WS), "Human Assisted Neural Devices (SN07-43)", "Autonomous Real-Time Ground Ubiquitous Surveillance-Imaging System (ARGUS-IS)" and "Urban Reasoning and Geospatial Exploitation Technology (URGENT)"
Perhaps best known is DARPA's Grand Challenge Program which has developed fully automated road vehicles that can successfully navigate real world terrain in a fully autonomous fashion.
DARPA has also supported programs on the Semantic Web with a great deal of emphasis on intelligent management of content and automated understanding. However James Hendler,
the manager of the DARPA program at the time, expressed some
disappointment with the government's ability to create rapid change, and
moved to working with the World Wide Web Consortium to transition the technologies to the private sector.
The EU-FP7 funding program provides financial support to
researchers within the European Union. In 2007–2008, it was funding AI
research under the Cognitive Systems: Interaction and Robotics Programme
(€193m), the Digital Libraries and Content Programme (€203m) and the
FET programme (€185m).
Current "AI spring"
A
marked increase in AI funding, development, deployment, and commercial
use has led to the idea of the AI winter being long over.
Concerns are occasionally raised that a new AI winter could be
triggered by overly ambitious or unrealistic promises by prominent AI
scientists or overpromising on the part of commercial vendors.
The successes of the current "AI spring" are advances in language translation (in particular, Google Translate), image recognition (spurred by the ImageNet training database) as commercialized by Google Image Search, and in game-playing systems such as AlphaZero (chess champion) and AlphaGo (go champion), and Watson (Jeopardy champion). Most of these advances have occurred since 2010.
Underlying causes behind AI winters
Several
explanations have been put forth for the cause of AI winters in
general. As AI progressed from government-funded applications to
commercial ones, new dynamics came into play. While hype is the most commonly cited cause, the explanations are not necessarily mutually exclusive.
Hype
The AI winters can be partly understood as a sequence of over-inflated expectations and subsequent crash seen in stock-markets and exemplified by the railway mania and dotcom bubble.
In a common pattern in the development of new technology (known as hype
cycle), an event, typically a technological breakthrough, creates
publicity which feeds on itself to create a "peak of inflated
expectations" followed by a "trough of disillusionment". Since
scientific and technological progress can't keep pace with the
publicity-fueled increase in expectations among investors and other
stakeholders, a crash must follow. AI technology seems to be no
exception to this rule.
For example, in the 1960s the realization that computers could
simulate 1-layer neural networks led to a neural-network hype cycle that
lasted until the 1969 publication of the book Perceptrons
which severely limited the set of problems that could be optimally
solved by 1-layer networks. In 1985 the realization that neural
networks could be used to solve optimization problems, as a result of
famous papers by Hopfield and Tank, together with the threat of Japan's 5th-generation project, led to renewed interest and application.
Institutional factors
Another factor is AI's place in the organisation of universities. Research on AI often takes the form of interdisciplinary research.
AI is therefore prone to the same problems other types of
interdisciplinary research face. Funding is channeled through the
established departments and during budget cuts, there will be a tendency
to shield the "core contents" of each department, at the expense of
interdisciplinary and less traditional research projects.
Economic factors
Downturns
in a country's national economy cause budget cuts in universities. The
"core contents" tendency worsens the effect on AI research and investors
in the market are likely to put their money into less risky ventures
during a crisis. Together this may amplify an economic downturn into an
AI winter. It is worth noting that the Lighthill report came at a time
of economic crisis in the UK, when universities had to make cuts and the question was only which programs should go.
Insufficient computing capability
Early
in the computing history the potential for neural networks was
understood but it has never been realized. Fairly simple networks
require significant computing capacity even by today's standards.
Empty pipeline
It
is common to see the relationship between basic research and technology
as a pipeline. Advances in basic research give birth to advances in
applied research, which in turn leads to new commercial applications.
From this it is often argued that a lack of basic research will lead to a
drop in marketable technology some years down the line. This view was
advanced by James Hendler in 2008,
when he claimed that the fall of expert systems in the late '80s was
not due to an inherent and unavoidable brittleness of expert systems,
but to funding cuts in basic research in the 1970s. These expert systems
advanced in the 1980s through applied research and product development,
but, by the end of the decade, the pipeline had run dry and expert
systems were unable to produce improvements that could have overcome
this brittleness and secured further funding.
Failure to adapt
The
fall of the LISP machine market and the failure of the fifth generation
computers were cases of expensive advanced products being overtaken by
simpler and cheaper alternatives. This fits the definition of a low-end disruptive technology,
with the LISP machine makers being marginalized. Expert systems were
carried over to the new desktop computers by for instance CLIPS,
so the fall of the LISP machine market and the fall of expert systems
are strictly speaking two separate events. Still, the failure to adapt
to such a change in the outside computing milieu is cited as one reason
for the 1980s AI winter.
Arguments and debates on past and future of AI
Several
philosophers, cognitive scientists and computer scientists have
speculated on where AI might have failed and what lies in its future. Hubert Dreyfus highlighted flawed assumptions of AI research in the past
and, as early as 1966, correctly predicted that the first wave of AI
research would fail to fulfill the very public promises it was making.
Other critics like Noam Chomsky have argued that AI is headed in the wrong direction, in part because of its heavy reliance on statistical techniques. Chomsky's comments fit into a larger debate with Peter Norvig,
centered around the role of statistical methods in AI. The exchange
between the two started with comments made by Chomsky at a symposium at
MIT to which Norvig wrote a response.
