Martha the Symbol Transformer

Marchesk

I posted this in the other place, so I'll restate it a bit differently here.

As a twist on Searle's Chinese Room argument, let's say that Martha, who speaks !Kung, has been trained to meticulously follow whatever rules she is handed for transforming symbols from set A to set B. Sometimes the symbols are arbitrary and made up (with arbitrary rules for how to combine them), and sometimes they are human languages, but ones Martha is not familiar with, such as English and Chinese. And sometimes it's a mix, such that English is translated into B, or A is transformed into Chinese*.

Now I think this demonstrates clearly that symbol manipulation does not result in understanding, because the system (Martha, rules and room), does not require meaningful symbols in order to perform a translation from one set to another. As far as the system is concerned, there is no difference between English or Chinese and some arbitrary symbols, except probably in the complexity of the rules needed to compute the translation (for human languages).

As such, symbol manipulation cannot be what underpins understanding or meaning. Searle was correct about that. Whatever it is that we do, it's something more than symbol manipulation.

* Of course the generated Chinese doesn't stem from anything meaningful.

Hanover

I don't really see how this is as much a twist on the Chinese Room thought experiment as it is a restatement of it.

Marchesk

Maybe so, but a lot of criticisms I come across are system replies, so I thought maybe if it was stated differently, it would be clearer that the "systems reply" doesn't work.

Michael

Now I think this demonstrates clearly that symbol manipulation does not result in understanding — Marchesk

What about maths? Isn't that just symbol manipulation? Given the input 2² I've been told to output 4. This is even more evident when we start to use imaginary numbers and derivatives and whatnot. Surely this constitutes understanding?

The Great Whatever

↪Michael

A funny thing about this is that no one is actually taught math that way. Children aren't taught to output certain numbers when certain other numbers appear to the side of + -- no, they're taught to count, then learn addition on analogy to counting, then multiplication on analogy of multiple additions, then exponents on analogy to multiple multiplications.

Michael

↪The Great Whatever

When I was taught derivative functions I was taught to move the power to the left of the letter and then reduce the power by one such that x³ becomes 3x².

Seems to amount to symbol manipulation.

Pierre-Normand

Seems to amount to symbol manipulation. — Michael

Yes, but does this competence with symbol manipulation constitute understanding? That you are able to manipulate the symbols in accordance with rules immediately shows, at most, that you understand the rules; that is, that you understand what it is permissible or mandatory for you to do in order to correctly apply them in particular cases. This, however, isn't the understanding that is at issue in the debate regarding functionalism, or computationalism, in the philosophy of mind.

The understanding at issue rather is the understanding of the meaning of linguistic symbols (and hence also the understanding that grounds intentionality -- e.g. reference to extra linguistic items in the general case). Functionalists claim that competence in following the syntactic and/or logical (i.e inferential) rules that govern the use of the symbols is sufficient for constituting an understanding of their meanings. Searle disputes this. Hence, for him, an ability to "understanding" the rules (i.e. display an ability to comply with them) falls short from understanding the language.

Michael

Yes, but does this competence with symbol manipulation constitute understanding? — Pierre-Normand

That's the question I asked. When it comes to maths, doesn't understanding consist in knowing how to manipulate the symbols, or at least knowing what to do with the input (e.g. plot a graph)?

Michael

If we take a Wittgensteinian approach to language, knowing what a sentence means is knowing how to use that sentence. In the case of the symbol transformation, does Martha know how to use the sentence? Well, yes; she's been told what to do with it. The issue is that the rules she's been given are different to the rules that the Chinese and English speakers have learnt (e.g. use the sentence "it is raining" when water falls from the clouds). What's really going on here is that Martha doesn't understand Chinese and English in the way that the Chinese and English speakers understand them; she doesn't understand their rules for using the languages. But she does understand her rules for using the languages.

So she doesn't understand Chinese or English but she does understand Martha-Chinese and Martha-English.

Marchesk

That's the question I asked. When it comes to maths, doesn't understanding consist in knowing how to manipulate the symbols, or at least knowing what to do with the input (e.g. plot a graph)? — Michael

No, understanding math is like understanding programming. You can use both in situations you haven't encountered before.

When I was taught derivative functions I was taught to move the power to the left of the letter and then reduce the power by one such that x3 becomes 3x2 — Michael

And what does that get you? How do you use it to solve problems or accomplish tasks?

Marchesk

If we take a Wittgensteinian approach to language, knowing what a sentence means is knowing how to use that sentence — Michael

But by this, did Wittgenstein mean knowing how to transform one sentence into another, or did he mean knowing how to use it in the world?

It's the difference between looking up a word in a dictionary, and being able to use that word in various contexts, such as might come up in conversation.

Soylent

↪Marchesk

Forgive my ignorance, but my initial reaction was that the "systems reply" was still suitable. The symbol manipulation has to be meaningful to someone on the output end otherwise the rules are arbitrary. To the output recipient, the system understands the language even if the language is some hybrid of existing language(s).

Marchesk

↪Soylent

Humans understand the symbols the system is outputting. But does the system? Searle was objecting to a strong notion of AI in which computers could achieve real understanding. He was also objecting to the idea that our brains are basically symbol manipulators (computational theory of mind).

Soylent

Humans understand the symbols the system is outputting. — Marchesk

This seems question-begging. I just don't see how the Chinese Room demonstrates one way or the other that humans understand symbols in a different way than the aggregate of the system. Or if humans do understand symbols differently, why we should exclude the notion that a sophisticated system can also understand symbols, albeit differently.

Hanover

That's the question I asked. When it comes to maths, doesn't understanding consist in knowing how to manipulate the symbols, or at least knowing what to do with the input (e.g. plot a graph)? — Michael

If one performs mathematical processes without any understanding of the meaning of the symbols or of the reasons behind the processes, then that person would be like the English speaking person in Searle's Chinese Room. You've just restated the thought experiment using math as an example as opposed to the Chinese characters.

The question is whether such examples are akin to what we actually do when we speak and convey our thoughts to other people. I'd say it's pretty dissimilar, considering in the Chinese Room we have no idea what thought we're conveying, but when we say "I'm hungry" (for example) we do.

Marchesk

This seems question-begging. I just don't see how the Chinese Room demonstrates one way or the other that humans understand symbols in a different way than the aggregate of the system. Or if humans do understand symbols differently, why we should exclude the notion that a sophisticated system can also understand symbols, albeit differently. — Soylent

Let's put it a different way. Symbols stand in for whatever it is that we understand. They're an abstraction. The claim Searle is making is that no amount of symbol manipulation gets you to understanding, because understanding isn't in the symbols. The symbols represent or encode for some agreed upon meaning.

What I was trying to do with Martha the !Kung speaker, and the arbitrary sets of random symbols is to show that understanding is something other than symbolic computation (manipulation). To Martha, Chinese and English are the same as some random symbols that don't mean anything. And that is exactly what symbols are to a computer.

Marchesk

Since math is being brought up, let's take the symbols 1 and +. We can of course tell a computer to compute 1 + 1. It will give us back a new symbol, 2. And we could teach a young child to memorize or lookup the results of addition. Whenever they see 1 + 1, they say or write two. An idiot savant might be able to memorize adding (or multiplying) very large numbers, which would be computer like.

But that's not understanding. 1 + 1 means something. It means you can take one of any individual item and put it together with one of any other individual item and have two items. And that's how kids start off learning how to do basic math. They use beads or marbles or whatever. They don't just blindly memorize the rules for symbol manipulation.

That comes in later math classes, which has been a bone of contention and motivation for educational reforms in math (although it goes back to having kids use the multiplication table to get the result of say 3 times 7 without understanding what that means).

Janus

↪Michael

Wouldn't a deeper understanding of maths involve knowing why symbols are manipulated the way they are on the basis of the kind of understanding of the way math is built up from the primitive logic of counting that

↪The Great Whatever

pointed to?

Marchesk

Let's say we give a test subject a rulebook and then feed them sets of symbols. For example, whenever they see:

å ß ∂

Then they write down:

ç

Now do these symbols mean anything? Does the set of rules for computing the correct symbol (or symbols) result in some sort of understanding?

And I'll grant that there is an understanding (given a certain meaning of the word) in how to go from one set of symbols to another (based on the rules). But is that what we're doing when we speak? Searle's contention is that it is not, such that the Chinese room can't be said to understand Chinese.

Michael

But by this, did Wittgenstein mean knowing how to transform one sentence into another, or did he mean knowing how to use it in the world? — Marchesk

That depends on how the sentence is supposed to be used. If it's supposed to be used in the world then to understand it is to understand how to use it in the world. If it's supposed to be used in a transformation then to understand it is to understand how to use it in a transformation.

And as the sentences Martha is given are supposed to be used in a transformation then to understand them is to understand how to use them in a transformation.

Saying that Martha doesn't understand the sentences because she doesn't understand how to use them in the world is like saying that I don't understand the word "bite" because I don't understand how the French use it.

And what does that get you? How do you use it to solve problems or accomplish tasks?

It got me an A in Maths.

Marchesk

Saying that Martha doesn't understand the sentences because she doesn't understand how to use them in the world is like saying that I don't understand the word "bite" because I don't understand how the French use it. — Michael

Which is the same as saying that Martha doesn't understand Chinese, right? The point being that languages are used in context of a world, not a lookup table or by consulting a dictionary, or applying some bayesian algorithm.

Michael

Which is the same as saying that Martha doesn't understand Chinese, right? — Marchesk

Sure. But she understands the sentences in the Martha-specific language. Just as I understand the word "bite" in English.

Marchesk

But she understands the sentences in the Martha-specific language which uses Chinese characters. Just as I understand the word "bite" in English. — Michael

We can agree on that. Searle's contention is stronger. He was arguing against the notion that a computer could understand Chinese like a human being does. Applying a different meaning to "understand" and then claiming that Searle has it wrong is to miss what he was arguing against.

If we want to say that the Chinese room understands Chinese in a rule-following or symbol manipulation manner, then okay. That would be like saying that Siri understands when it's cold outside because she says, "Brrrr, it's 15 degrees outside". But of course she doesn't know what it means to be cold.

Michael

He was arguing against the notion that a computer could understand Chinese like a human being does — Marchesk

True. But only because the machine is being taught to use the words in a different way to how a human does. So obviously it won't understand it like a human does. But what if the input is sensory information rather than sentences? The machine detects water falling from the clouds and so outputs "it is raining". This would be a proper way to consider computer understanding.

The problem with Searle's argument is that if a human was put under the same conditions as a computer then the human wouldn't understand (in the same way as a human in a traditional situation). But a human is still conscious. So that a computer wouldn't understand (in the same way we would) under those same conditions is not that it is not conscious. He needs to put the computer under the same conditions that a human would be under to understand the sentences.

Marchesk

The machine detects water falling from the clouds and so outputs "it is raining". This would be a proper way to consider computer understanding. — Michael

The computer understands it in a propositional sense. Let's make this more complex. Let's say the computer has been programmed to read faces and emotion at a funeral. It then tells a grieving person that it's very sorry for their loss.

Does the computer understand what it means to lose someone?

Michael

↪Marchesk

I could ask the same about you. You read faces and emotions and tell a grieving person that you're sorry for their loss. How would I determine that you understand what it means to lose someone? What sort of evidence would verify or falsify the claim that you do (or don't)? Surely the evidence must be public. And so I can look for that same evidence with a computer.

Michael

Wouldn't a deeper understanding of maths involve knowing why symbols are manipulated the way they are on the basis of the kind of understanding of the way math is built up from the primitive logic of counting that ↪The Great Whatever pointed to? — John

What does fully understanding maths consist of? Knowing the axioms, the rules of inference, and then being able to apply the latter to the former? So I'm provided with some input sentences, told what to do with them, and then output the result.

Marchesk

↪Michael

Perhaps I have yet to lose someone close and therefore am just being polite.

So let's say you stub your toe. I say that looks painful. Are you going to doubt that I understand what being in pain is? Or if you tell me about a strange dream. Do you doubt that I will understand having a strange dream?

Michael

So let's say you stub your toe. I say that looks painful. Are you going to doubt that I understand what being in pain is? Or if I say that it really sucks not sleeping the night before a big day. Are you going to doubt that I know what it's like to be fatigued? — Marchesk

I might or I might not. And I might or might not doubt the computer that says the same thing. I'm not really sure what this question is supposed to highlight. Perhaps that we dogmatically believe that people understand but computers don't?

Marchesk

The problem with Searle's argument is that if a human was put under the same conditions as a computer then the human wouldn't understand (in the same way as a human in a traditional situation). — Michael

Right, because he was attacking symbol manipulation as a form of understanding.

But a human is still conscious. So that a computer wouldn't understand (in the same way we would) under those same conditions is not that it is not conscious. He needs to put the computer under the same conditions that a human would be under to understand the sentences. — Michael

Okay, so there's consciousness-based understanding where the words, "I'm sorry for your loss" don't imply understanding unless symbol producer has experienced loss or can empathize with losing someone.

And then there's the question of intentionality. How do symbols refer? How is it that 1 stands for any single individual item?

Marchesk

I might or I might not. — Michael

How would you not? Are you supposing that I have some condition where I can't experience pain or fatigue (I'm not aware that there are any humans immune to fatigue).

Martha the Symbol Transformer

Welcome to The Philosophy Forum!

Categories

More Discussions