The connection between language and thinking

Thank you very much! Sorry I didn't answer right away, I only get to my email in the evening and I didn't have time yesterday.

Indeed, my questions were written in a mixed way – I wrote it early in the morning and in a hurry, and I agree that there are many different questions within the same section. I think the main thing that bothered me beyond the questions you formulated was the element of language. For me, when I spoke of “analytic” I meant both the basic assumptions and the logical deductions, but not the way we came to formulate or knew the basic assumptions. In the context of what mainly concerned me – is, as mentioned, the matter of language. The thought became more acute for me when I read something by Roy Tsezna, who writes a lot on the subject, and he said that even the powerful engines are essentially language engines, all of them. And then it suddenly gives language a different status than I had attributed to it until now. The possibility arises that it is enough to teach the computer to speak properly for a real thought to emerge from it - or at least a process that accurately imitates real and conscious thought, or at least accurately enough for any practical need. And then I wonder what is it about language that makes this possible? The possibilities that came to my mind were:

1. Language is a representation of other logical or mental processes, and therefore it is enough that we manage to use language properly to create a logical process. The representation becomes the essence: either literally, assuming that the computer has consciousness, or only when there is a conscious person who reads the words and understands what is written in them. But in any case, it is impossible to distinguish between the two unless you are the person speaking/thinking the things. I think this is one of the things that motivated Russell to create a mathematical language that would not have paradoxes: to cleanse the language of its illogical elements so that language and logic become equivalent. It is indeed impossible, but it may be possible enough for practical needs.
2. Language itself is built on implicit concepts that we have and on relationships between objects, in space, time and causal relationships, and this is embedded, for example, in the natural syntax of the brain that Chomsky talks about, and therefore, as soon as we learn to speak, it also includes processes of thinking. Thinking is embedded within the syntax. This still means that from the computer's perspective, we may be in section 1, but from the perspective of the person speaking, language is no longer separate from logic - not as a representation, but because language itself is truly a logical process. Of course, it is not perfect because I can formulate absurd sentences in language.
3. Human language is also not really thinking, just as the language of the engine is not really thinking - it can be, but thinking is a separate process that runs parallel to putting things into words. Language can help this process or hinder it or simply reflect it, and it can run completely separate from it.

I will try to read the columns you sent me and watch the interview, but it will take me some time because I work from morning to night and I don't have a smartphone, but I will try and think about it a little more..

By the way, have you thought about maybe producing a podcast? It's a little more complex than recording lessons in my opinion, and if you find a partner with whom you can record and have an intelligent conversation about philosophical issues – it could be very successful, much more informative, because today people consume a lot of information in this way and not by reading.

Thank you very much!

I am confused about the language questions you raised. I think that behind language there are ideas, and the mechanics in which AI software operates do not mean that it deals with language. Even the name language is a representation, except that the things represented are not in the ”knowledge” of the software but in the programmer.
The software is trained through answers and feedback to cases, and these are not determined linguistically. For example, you train a software to recognize someone's face. The training is through examples and giving positive or negative feedback for its correct and incorrect identification. You understand that the feedback is determined by the question of whether it identified correctly or not, rather by the content. It is not just a formal syntax and that's it. Therefore, it shapes the network in a way that will respond to semantics. This is not a purely linguistic operation, although it is represented by language. With us, ideas are also represented by language, but there are ideas behind it. The person in the Chinese room receives feedback from Chinese speakers and not just syntactic connections.

As for the podcast, it's an interesting idea. I'll think about it, although lately I've had a lot of requests for existing podcasts and maybe this already meets that need.

I thought about your answer – I agree in principle because this was the way I thought about things until recently, that you can play with words using a computer, but the content – the meaning – remains with us and language is nothing more than a representation. But this perception is now being undermined for me because it seems that the artificial intelligence engines that are based on language are managing to imitate thinking so well that it seems that it is already difficult to distinguish between what they do and normal human thinking, and to the best of my knowledge all of these engines are essentially language-based (even the advanced engines). That is, even ten years ago we had neural networks that could solve problems using a huge mass of examples, but the turning point in the development of artificial intelligence engines began with language engines. At least to the best of my knowledge, it was not much. If this is true, I continue to wonder whether the picture you present – which is the picture that I also had until recently – Is it indeed true or is there something in language that is more than mere representation? It should be emphasized that I do not think that language is the same as thinking, and I raised this possibility only because it is a possibility that should be considered when trying to think about the subject, but it is not difficult to rule it out both conceptually and phenomenologically (I know from within myself that one can think without words and one can speak without thinking). But if it is true - then the question I started with two weeks ago returns for me: to what extent when I myself make an argument is there thinking behind it, or perhaps the thinking comes only after the argument has been made or not even then - that is, it provokes thinking about when language itself succeeds in "thinking on its own" and when it does not. As an example, I can give something similar but different - mathematics. If you give me an exercise in mathematics, let's say investigating a function, then I can observe the function and describe it quite well from an understanding of what it does and tell you what it will look like. That is thinking. But I can also sit with the algorithm and derive and build equations and transfer sides and place numbers, etc. ’ – a process that can be purely mechanical (and as long as we adhere to correct syntax, it seems that Lish will not make a mistake and we will always reach results) – and reach the same results even without understanding what I am doing, just as 3-unit students solve without knowing what they are doing. So in mathematics, the powerful language developed there “thinks” for them. Clearly, this is an example that fits what you wrote – the language does not really think but rather incorporates processes developed over hundreds of years of thinking and are only the bottom line, and the meaning given to them is only in the mind of the solver. But there is no doubt that mathematical language has great power that is largely equivalent to thinking, in terms of the results it produces – And in many cases it will also replace thinking, when the function is too complicated and I have no way of predicting its behavior, as in a chaotic dynamic situation, and then we rely on language alone if we want specific results. Then the question arises – What happens with natural language? Does it also have a certain degree of this power? I feel that language engines suggest that it may, and then it is perhaps very intriguing to think about what it is in language, in words – the natural language, not the artificial mathematical one that we created on purpose so that it would think for us – that makes this possible.

I saw that you started writing a few columns on the subject and I look forward to seeing if there will be a reference to this direction in the end, if at all you think it has any benefit. What is more, a small correction to the previous column – I am not a “former student”, I am a “student” 🙂

I intend to write the column after the next one (I added another one, beyond planning, before it) about this. In general, I do not see anything special in language engines beyond any other machine. The performance is of course much more powerful, but the whole argument is that use and meaning are not the same thing. The example of the student and the investigation of the mechanical function sharpens this well. Therefore, even where the use/syntax manages to reach the same results as humans, it does not mean that there is thinking there (as in the student). As mentioned, water also has tremendous abilities in solving Navier-Stokes equations (it is true that not in other fields, but it is a quantitative difference).
In my opinion, language engines do not really challenge human perception, and this is what I explained in these two columns. In the next column I plan to deal with the opposite question that you raised: not whether the machine is a person but whether a person is a machine. That is, whether our thinking is purely mechanical. Or, unfortunately, to ask whether language engines lead more to this conclusion (probably yes). But the question of whether a machine is a person is a different question. There is a difference between comparing performance (the Turing test) and deciding that it is a person and has thinking and awareness.

I think the discussion at the end of the last column about thinking in stages (dipsic) illustrates this distinction very well.

I haven't read the last column yet – I absolutely agree that the mere fact of arriving at results does not indicate thought, which is why I gave the example of the student and the function, but one of the things that interests me, besides the question of whether a person is a machine, is what is in language that makes this possible. I know what is in the language of mathematics that makes it possible to arrive at results, but it surprises me to discover that there is probably something like this in natural language as well. But I will wait to read what you wrote..

(I still don't know if it is possible to reach the conclusion that our thinking is purely mechanical, I still allow for the possibility that there is a certain component in thinking, which is expressed in rare situations and not in routine thinking, which is not mechanical. But I have no real basis for this claim other than the different experience – there is a difference in feeling between an insight that comes after organized thinking and an insight that comes from inspiration, perhaps a bit like the difference between standard science and a paradigm shift in Kuhn. But of course it is possible that this is just a different feeling and that in fact it is an unconscious mechanical mechanism).

I don't see this as a feature of language. It's about the content that language represents. The AI software is trained on content. The fact that it assimilates and processes them through language is also true for us. Language expresses content, and the correlations between words are not a feature of language, but of texts that express content and use language. The software does not arrive at its answers through syntactic considerations, but through correlations between words, and these are determined by the content (in the human texts on which the models are trained).
Think about it, if you were to train the model on texts that are syntactically correct but express incorrect and stupid content, you would get results that are worthless. Hence, what determines is the content (which is expressed in language) and not the linguistic structure and language itself.
As mentioned, I will get to the question of mechanical thinking in the fourth column.

On the sidelines of this discussion and perhaps as an appendix to it with certain implications: Will a person who speaks two or more languages, when the same thought occurs to him in one language, be the same - cognitively - as the same thought in another language? Intuition tells me no.

I didn't understand the question. Are you asking whether our thinking is done in language or the ideas themselves? It is commonly thought that this is a dispute between Toss and Rashba, but it is clear from the explanation that we usually think in language, but I suppose it is also possible without language. See the series from column 379 onwards.