Ascribing Mental Qualities to Machines | John McCarthy
This is a paper from the Darthmouth conference where the term "AI" was coined.
These notes are only for the introduction and section 2.
Main message:
Ascribing mental qualities like beliefs, intentions and wants to a machine is sometimes correct if done conservatively and is sometimes necessary to express what is known about its state. (1)
Introduction
Great opener:
To ascribe certain beliefs, knowledge, free will, intentions, consciousness, abilities or wants to a machine or computer program is legitimate when such an ascription expresses the same information about the machine that it expresses about a person. It is useful when the ascription helps us understand the structure of the machine, its past or future behavior, or how to repair or improve it. (1)
McCarthy initiated a debate with Searle around his Chinese Room argument that, essentially, aims to show that even though machines may display intelligent behavior, we cannot legitimately ascribe mental states such that this machines "really" has them.
However, McCarthy immediately moves more in the direction of what Dennett calls the intentional stance: perhaps it does not matter much if these machines "really" or "legitimately" have these beliefs and desires etc; it may still be a very efficient way of explaining the machine's behavior.
In a way, we can say the same about ascribing beliefs to humans:
It is perhaps never logically required even for humans, but expressing reasonably briefly what is actually known about the state of a machine in a particular situation may require ascribing mental qualities or qualities isomorphic to them. (1-2)
This is the most useful when we know very little about a machine's structure.
The start of AI, as a practical approach to tackling deep questions about the mind:
They can be taken as asserting that many of the philosophical problems of mind take a concrete form when one takes seriously the idea of making machines behave intelligently. (2)
AI highlights two things:
- We "need to determine the conditions under which specific mental and volitional terms are applicable" (2)
- general intelligence requires a metaphysics (general world-view) and epistemology (theory of knowledge) (3).
Currently machines do not have beliefs about beliefs.
What's also notable is that McCarthy is a "neat" in AI. He assumes that AI's "use of these concepts requires formal axiomatization" (4) The deductive approach to AI is nowadays less successful and in any case above statement is very controversial.
Why ascribe mental qualities
Because doing so is a more efficient way "to express available information about the machine and its current state" (4), instead of having long listings of computations that any machine did to reach that state.
So why? The reasons are epistemological i.e. "ascribing beliefs is needed to adapt to limitations on our ability to acquire knowledge, use it for prediction, and establish generalizations in terms of the elementary structure of the program" (5).
- Program states are not directly observable
- When interacting with a complex environment, regarding computer behavior on a detailed level would also require us to know and simulate the environment in a lot of detail. If this is even possible, this is too slow.
- "Ascribing beliefs may allow deriving general statements about the program's behavior that could not be obtained from any finite number of simulations." (5)
- They are easier to understand.
- The belief and goals structure may resemble the intended behavior as the designer had it in mind, and this may be useful in debugging efficiently.
- The difference between programs "may best be expressed as a difference in belief structure" (5)
N.B. the fact that these reasons are epistemological and not metaphysical underlines that McCarthy does not want to commit to saying that these machines really have these mental qualities.