1

This question already has an answer here:

I'm a relatively new student studying philosophy and became interested in the computer models of mind. I've read John Searle's CR experiment and had a quick thought. How can CR experiment refute to the below?

If a computer is fed with bunch of data that it actually learns or correctly answers untaught problem, is it not being intelligent?

Say for instance, a system is given a multiplication table from 1 to 100. It is taught formally and knows how to read the table and give correct outputs of multiplication. Let's say the computer "learns" how to multiply (say the data above is enough for the computer to actually learn multiplication).

If a user asks "101 x 101?" and the computer correctly answers it, this displays learning and thus capability to think?

marked as duplicate by Conifold, Thomas Klimpel, virmaior, Mozibur Ullah, John Am Feb 23 '17 at 16:00

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.

  • 2
    Hi, welcome to philosophy SE. On Searle's view neither Chinese room nor computer "actually learns" anything because it lacks "understanding". Similar question appears to have been asked before, see Does Searle's Chinese Room model computers correctly? and already has answers. – Conifold Feb 21 '17 at 19:01
  • 1
    Question. If I get into an elevator, press "3", and the elevator goes up (or down) to floor 3, stops, and opens its doors, is that an act of intelligence? Serious question, since what computers do is no different in principle. – user4894 Feb 21 '17 at 19:16
  • If the computer used the incomplete data it was given to perform a more complete version of the task, one would likely say it was thinking. But what are the odds of this actually happening? It would have to extrapolate the pattern in what it sees and derive the additional fact. It has not been given information about how to do this extrapolation. So what is the question? – jobermark Feb 21 '17 at 19:35
  • 1
    @Vinci Since even an advanced machine learning neural network reduces to a physical instance of a Turing machine, no argument along these lines can work. A four-function calculator has a "memory." If it "remembers" an intermediate result, is it intelligent? I can program a computer to "remember" your preferences, just as Amazon "remembers" everything I've ever bought on their site. That's not intelligence, it's programming. Storing bit patterns for later retrieval. Of course some think (without the slightest evidence) that this is what brains do. – user4894 Feb 21 '17 at 20:15
  • 2
    I think the word "learns" is used in two different senses when it is applied to humans and to computers. If students in a math class memorized how to use the power, product, chain, etc., rules mindlessly to arrive at correct answers we do not say that they "learned" calculus. Yet if a neural network did the same we would call it "machine learning". There is a higher threshold for learning when it comes to humans, some intangible extra. It could just be that it is an "illusion" and deep deep down it reduces to neural memorization, but at this point this is a wild speculation with no specifics. – Conifold Feb 21 '17 at 21:08
1

If the instructions in the Chinese room were so abstruse that one could actually follow them without indirectly acquiring the ability to read Chinese, the person in there would not learn Chinese. The problem here is that odds are that over time he actually would learn Chinese, to the point that he only occasionally had to refer to the instructions. We all know this, because the actual process of speaking Chinese is simpler in human terms, than any description of it. We are information-based, meaning-seeking animals. Our language performance ability outstrips our ability to form lexicons and usable grammars by many orders of magnitude. From the point of view of someone like Winograd in his simplification of Maturana, it is this kind of built-in process that renders finding meaning more natural to us than performing tasks that forms the content of understanding.

But that is not true for a machine. Instructions are instructions and following them is what the mechanism does. If those instructions are a heuristic for learning the pattern behind a process, they may uncover that pattern. But then that heuristic is the program. The data about multiplication is not the program in your example, and the machine is not learning multiplication, it is displaying its program's ability to recognize and reproduce patterns of behavior. It has no bias toward finding meaning. It may borrow ours through simulations of ourselves that we create, but it does not have and understanding (in the above sense) of its own.

Basically, you are cheating: you are giving the machine a great deal more information than a human starts with, and then ignoring the fact it is there. So this is not a challenge to the argument, once you include the setup of the machine. And you are assigning that borrowed power to the machine itself, as if it did not come from outside.

-1

Keep in mind that Searle's "Chinese Room Argument" is a refutation of the computational theory of mind and not a refutation of the possibility of machine learning or non-biological consciousness (as it is often mis-read that way). Nor is Searle's theory positing that the computer is not a useful metaphor for consciousness or tool for examining consciousness - both, however, have their limits. What Searle is doing with the Chinese Room argument is showing the limits of syntactical manipulation to achieve semantic content. His aim in doing so is to explain and one day demonstrate how to get from "the physics" to "the semantics."

If a computer is fed with bunch of data that it actually learns or correctly answers untaught problem, is it not being intelligent?

Doesn't your question presume the consequent? If a computer actually learns, i.e. demonstrates intelligence, then yes, the computer demonstrates intelligence but you are presuming that the computer actually learns.

Clarification is needed as to what it would mean for a computer to actually learn in something other than a jejune sense of actual learning. I.e. you could say that the computer has "actually learned arithmetic" after I enter numbers into the computer and it returns their sum, however, after entering into the computer data for the values of two and three as well as programming the means by which to concatenate numerical values, when I then instruct the computer to return their sum (i.e. ask it "what is two plus three?") and the number 5 is printed to the screen - this is not evidence that the computer has learned anything at all. Note here that the computer does not know how to answer "what is two plus three?" it merely operates upon the symbols for 2, 3, and + such that 5 is returned.

enter image description here

In the case of what computers (i.e. Turing machines) are actually doing, they are only literally learning in the sense that "literal" is used to mean "metaphorical" or "figurative" and not actual. To the point tho, what is meant here by your question: "is [the computer] not being intelligent?" - "intelligent" according to whom? "Learn" according to whom?

While the field of machine learning has made impressive advances with what can be called learning (see the work of Jürgen Schmidhuber), this kind of machine learning is distinct from the first-person subjective ontology of biological learning - whether human or some other animal.

For example, in computation what are called "neuronal networks" are an analog of how we can usefully model what is happening in the brain (and presumably causing consciousness). A metaphor, however, is all it is. We simply do not know enough about the physical processes in the brain and the body which cause consciousness to simulate anything other than artificial intelligence - "weak AI" in Searle's parlance (something he is not arguing against)

Lastly, to address your question so that you may formulate one which has heuristic and not merely interpretive value, consider as well in which sense do you mean "intelligent"? Or "Artificially intelligent"?

From "Brain, Mind, and Consciousness: A Conversation with Philosopher John Searle" March 3, 2015 by Dan Turello

The expression “Artificial Intelligence” is multiply ambiguous, and it is unfortunate that this ambiguity has not been sorted out. An artificial x can be either a real x produced artificially or a fake x. For example, artificial dyes are real dyes produced artificially, it is just they are not produced from vegetables. Artificial cream, on the other hand, is not real cream but fake cream. So “Artificial Intelligence” can mean either something that is not intelligent and is produced artificially, or it can mean something that is actually intelligent produced artificially. “Intelligence” is also ambiguous because it is ambiguous between real, honest to John observer-independent intelligence – as, for example, when a human being is thinking about something – and the observer-relative, derivative, metaphorical sense of intelligence – for instance, when we speak of my pocket calculator or computer as displaying intelligence. This is unfortunate because in the sense in which humans have real observer-independent intelligence, commercial computers have nothing like that. The sense in which the computer is intelligent is entirely observer relative or metaphorical. The reason the “intelligence” of the commercial computer is entirely observer relative is that we do not know how to make a conscious computer. Existing computers work entirely by having complex electronic circuits, and what we think of as “computation” is a series of programmed transitions between the states of the complex electrical circuitry.

As to whether or not machines will be conscious, it is important to remember that we are machines. We are biological machines and we are conscious. I do not see any reason, in principle, why we could not build an artificial machine that was conscious, but we are unable to do that now because we do not know how the brain does it. The question, “Can you build an artificial machine that is conscious?” is just like the question “Can you build an artificial heart that pumps blood?” We know how to build artificial hearts because we know how the biological heart works. We do not know how to build an artificial brain because we do not know how the brain works. But assuming we knew how the brain worked, I see no obstacle in principle to building an artificial conscious machine. The important thing to see is that the human brain is a machine, a biological machine, and it produces consciousness by biological processes. We will not be able to do that artificially until we know how the brain does it and we can then duplicate the causal powers of the brain. Perhaps we can do it in some completely different medium as we build artificial hearts in a completely different medium from muscle tissue, but at present we do not know enough about the brain to build an artificial brain.

Not the answer you're looking for? Browse other questions tagged or ask your own question.