Navigation: This page '' ---> Using Dooy ---> Main Page. HELP. Admin. Contact.

Artificial Intelligence Thought Experiement:
Two Chat-bots Conversing With Each Other

"Imagine a situation in which two chat-bots are connected with each other and active in producing 'texts', while there are no humans any more, would this not be a completely physical situation, no object-functions being actual? In other words, are chat-bots for their actual functioning in what they are designed for not completely dependent on actual functioning of humans with their subject functions?"

With this question, Henk Geertsema sparked off a long discussion about how we would view this artificial intelligence question philosophically. This page discusses how we might approach that, using Dooyeweerd's aspects.

Progeny: One participant, Richard, remarked that he liked Andrew Basden's take on AI. So Andrew wrote a long reply into the conversation, which explains his approach. The text of Andrew's reply is below, because it sets out a useful approach to understanding artificial intelligence as well as computers in general, and chat-bots in particular, which might be useful to some.

The text of the email exchange has been omitted, unfortunately, because we do not have permission to replicate it here. But in most references to it bg Andrew, he explains about it.

Dear Henk and all,

I promised to address Henk's question about two Chat-bots (LLMs large language models) conversing with each other without humans,

"Imagine a situation in which two chat-bots are connected with each other and active in producing 'texts', while there are no humans any more, would this not be a completely physical situation, no object-functions being actual? In other words, are chat-bots for their actual functioning in what they are designed for not completely dependent on actual functioning of humans with their subject functions?"

And, because Richard mentioned he likes "what Andrew is doing and has done", I think I should make a bit of effort to explain how I wuld approach this question, rather than just make a few responses. Much of the below is contained in Chapter V of my 2008 book, "Philosophical Frameworks for Understanding Information Systems", which you, Henk, reviewed and responded to at the time, if you recall. (Do you still have the copy the publishers sent?). But I'll try to make it shorter.

ChatGPT is the outcome of many hundreds of person-years of work, so there is a lot to understand about it before we can make a true judgement about it. Therefore this will be long, but hopefully cut up into 11 (bite-sized?) chunks, each with a heading.

I argue that the conversing chat-bots are functioning in the lingual aspect even when no humans are around, as well as all those aspects on which lingual foundationally depends, including the physical. To do this,

[I have merged two email threads because both say something germane.]

--- 1. Basic Points of the Question

In the second of your Maarssen lectures, on entities, for which Rudi uploaded the sound and I have just uploaded the video

audio: ""

video: ""

around the 40 minute point, your opinion is that the subject function of a computer is the physical. I agree.

You also make the point that that does not adequately account for the computer as computer. I agree.

You also say that the qualifying aspect of a computer is one of its object-functions. I agree, if we wish to use subject-object terminology.

To address your question I will employ 5 steps of thinking. Some go into other chapters of my 2008 book.

--- 2. The Nature Of Computer As Computer.

Computers as we know them today have (are meaningful in) at least the following aspects (I include the pre-physical aspects in the physical):

Those are the aspects that are necessary to *understand* computer as computer. Computers may be *described* validly at all these levels:

If preferred, we can view the computer as *functioning* in those aspects

Attached is a table that shows some of these things; see table below.

At the end is an Appendix in which I go through these aspects in detail, in case you want it.

(Roy asks below, "The [heated] wire begins to glow green. This gives every appearance of physically qualified heat causing a sensory qualified glow." Not so. What happens when we understand it physically is that the copper atoms are excited and emit certain wavelengths of light. It is only green when we understand what is happening via the psychical aspect. I think that this is what Henk meant in his reply.)

--- 3. Post-physical Aspects

When viewing or understanding the computer in post-physical aspect we are viewing or understanding it via one of its object functions. But such understanding is just as valid is thinking about its physical subject-functioning.

Even bit-manipulation. So Chris' statement about a computer doing bit processing and not knowing the meaning thereof is a statement from the perspective of the psychical aspect.

From the perspective of the lingual aspect we would say that the computer is ommunicating. Which implies that it is valid to think of the two chat-bots communicating.

Note that when swtiched off (no electric power) only the physical (and pre-physical) functioning is occurring, and it is nothing more than materials and the forces etc. within them. But when switched on and running, all those other aspects are active and meaningful.

So, Henk, when you ask, "In other words, are chat-bots for their actual functioning in what they are designed for not completely dependent on actual functioning of humans with their subject functions?" I would reply "Yes" and the above explains why. However, let us now go to the "designed for".

Chris, may I quibble? When you say "simply exchanging bits of information (1s and 0s)" I would disagree with your use of the word "simply", implying that that this is the basic way to understand computers and possibly what their subject-functioning is. They are indeed exchanging bits but that is the view of what they are doing at the psychical level, not the physical. You are correct, however, when you go on to say, involved above the "physical (micro-currents and magnetic orientations)". However I would agree that the exchange of bits is "communicate nothing" since we are viewing it at the psychical aspect. When we view what is happening from the lingual aspect, however, I would say that they are indeed communicating. The reason they are not communicating is not about "since there is no subjectivity" but about from which aspect we view things; when we ask about "communicating" we are directing our discussion to view things from the lingual aspect; when we ask about "bits", we direct ourselves to the psychical aspect, and so on. Hope that's clear.

Read the Appendix, to see how all this works out in detail.

--- 4. Program as Virtual Law-side

Henk below remarks about "reality in principle can be grasped by means of algorithms". Well, that is ultimately impossible, except perhaps for some mathematical reality. However, it is approximately or partially what occurs in a program.

To run, the computer is programmed. In my 2008 book I suggested that a program is a virtual law-side; you, Henk, liked that idea.

(The activity of programming is also validly describable in those five aspects, but I will not discuss that here; it is in one of the other chapters of my book.)

It is humans that program computers. So, in a running computer there is an inescapable human element (even when no humans are around, e.g. in a spacecraft).

--- 5. The Program Encapsulates the Laws of Aspects Chosen For It to Work In

For example, to do arithmetic calculations, requires the program to encapsulate the laws of the quantitative aspect. To analyse xray plates requires the program to encapsulate laws of the spatial and psychical aspects.

ChatGPT ecnapsulates laws of the lingual aspect and maybe a few others e.g. formative (to structure its written replies), analytical (to separate out what the user is asking), social (to offer appropriate replies), etc. Currently it is poor in the juridical aspect, in that it 'hallucinates' as Chris mentions i.e. writes things that are untrue. Notice that all these aspects are inherent in full lingual functioning by humans.

(How a computer is programmed, whether by a knowledge engineer or machine learning, makes no difference there, but I don't want to explain here because it will interrupt our flow. Ask me if you wish.)

GPT4, which responds to pictures too, encapsulates the lingual plus spatial-psychical and maybe aesthetic aspects.

Therefore the conversing chat-bots are functioning in the lingual aspect primarily with the other aspects as lingual functioning depends on them (foundationally for pre-lingual, antecipatorily for post-lingual).

Notice that I do not here say whether it is subject or object functioning. In my book I call if "meaningful functioning", which includes both, unspeficied.

--- 6. The Programming / Training of ChatGPT and Chat-bots

# As Chris Gousmett says below (read it at Apr 7, 2023, 7:55 PM), ChatGPT works by 'knowing' (having encapsulated within it knowledge of) which words and phrases typically follow which. In terms of my approach above, that is syntactic knowledge, encapsulating laws of the formative aspect, built on top of that of the analytical aspect of distinguishing words. However, in being trained on human text (167 billion such data) it also absorbs (encapsulates) some knowledge about semantic signification, i.e. of the lingual aspect. It also has some knowledge of the social aspect, such as it refuses to deal with certain topics that are unacceptable.

Now, we might raise the question, how is this 'knowledge' actually encapsulated within the chat-bot? There are three ways, manual programming and machine learning. Chat-bots use both, for different aspects:

--- 7. On the Human Element in Machine Learning (ML)

Machine learning seems to work without human activity. But in fact it depends on human choice of

(a) what to include in and exclude from the training set (167 billion in size for ChatGPT). That is decided by the chatbot designers

(b) what to look for in all those # For example, words, sequences, lengths of phrases, spelling errors, grammar errors, who the author was, which language it is in, which date or period in history, etc. # For example, When the training on pictures (e.g. whether GPT4 or the humble xray analyser), it would look for colours, blocks of colour, edges of those blocks, sharpness or fuzziness of such edges, shapes, relative positions, angles, etc. and also probably who the artist or photographer. # ChatGPT has millions of such parameters, I understand - and they are selected by the designers manually.

(c) by which criteria to tell the machine learner that it is successful or not - a similar choice.

(d) tweaking - such as adding knowledge of social acceptability.

Notice these are all about what is meaningful, some in one aspect, some in another. Hence Dooyeweerd's aspects could enormously assist the development of such ML.

Chat-bots would not come into existence without the functioning of humans in these aspects. If they keep on working after the last human has gone, they still function in the various aspects that are the structure of the Creation.

--- 8. Three Lingual Functionings Of Computers

I propose that the qualifying / leading aspect of any computer is lingual. I also see the leading aspect of chat-bots as lingual. But in different ways and for different reasons. Moreover, programming a computer is lingual functioning. Three lingual functionings going on, which should not be confused together.

1. In the application, e.g. chat-bots, they may be said to be conversing, as just mentioned. Not all computers function lingually in this way, that is, if their program encapsulates no laws of the lingual aspect (such as might be true of a simple calculator).

2. From earlier, the lingual aspect of the nature of computers, true of all computers even calculators. The make-up of a computer is threefold:

Each one exists and functions in all or most aspects - but only the first functions when the power is switched off. The attached [following] table (table 5-1 in my 2008 book) shows these three, the UI at the left, the innards in the middle and the device to the right ("beige colour" somewhat dates that table!).

Aspects of computer 1280,1500

e.g. when in a computer game I am fighting a dragon, I am fighting a dragon, not manipulating shapes nor watching colours, nor having eye neurons excited. Of course, all those things are happening, but as foundation for fighting a dragon.

3. The activity of programming a computer is lingual functioning, in 'telling' the computer what we want it to do. This can use various languages, even graphical ones, but it is all lingual functioning by the human programmer.

Note: 'programming' is the word used here for both manual knowledge representation and also machine learning; synonymous with 'encapsulating knowledge into computer'.

(Note: The thought experiment of 'what if we program a computer to program computers to program yet other computers, and so on, is that still lingual functioning?' I would still answer Yes. But that is not argued here.)

Those three are reasons why I believe the best aspect to lead or qualify a computer is lingual.

And I find that that is sufficient for a rich foundational understanding of computers, as I have expressed in my 2008 and 2018 books - and one that links well with other issues like the use of computers with benefit or harm, impact on society, and the development process.

--- 9. Subject and Object

I prefer not to think in terms of the philosophical ideas of subject and object, but take a broader view of the entire Creation, which is subject to the various law-spheres (for which I'll assume Dooyeweerd's 15 here). The chat-bots are part of that Creation, and humans used to be.

To be honest, in my view, it is not nearly so helpful to ask about subject and object in addressing your question. As your question and the subsequent conversation has shown, it leads into irresolvables. Not least because it depends on how we each understand and interpret the idea of subject and object. Roy is usually very clear and logical so probably has a clear definition of subject and object, but I don't, and I find other interpretations among others. Nick Breems introduced the intersting and useful idea of proxy-subject, which fits computers very well.

(Note: The nearly-unique thing about computers is that unlike other objects like pens and bridges, they can respond. They are active not passive - which implies that Roy's characterization of subject and object as active and passive might need some tweaking.)

I find that I can very adequately understand computers and chatbots without the notions of subject and object. So I prefer not to answer Henk's question directly. My view above is from the perspective of the entire Creation (all fact-side creatures, including chat-bots) being subject to the laws of the diverse aspects of Creation's law-side, of which Dooyeweerd delineated 15. The aspects pertain even when there are nothing functioning in them - and for example before humans were created.

(David points out that aspects are not things, and Roy puts it, "background conditions for causality to take place".)

Thus the chatbots are still functioning in the lingual aspect for which they were programmed even after all the humans have been taken away (as long as the electric power is still switched on).

--- 10. On Posing the Question to ChatGPT / Bing

Stephen put Henk's question to ChatGPT. Quite impressive, at first sight.

Notice what I have done, which Stephen's Chatbot didn't: I have questinoed the question, and suggested a different way of addressing the issue.

As we understand how we do that, I guess that someone will eventually be able to train an AI system to do similar. But that still does not mean that AI is taking over.

I hope to respond to Stephen's offering separately from this, and have OCR'd the pic in preparation for that.

--- 11. Conclusion

I have argued that two chatbots conversing without humans is still functioning in the lingual aspect - and simultaneously all the other aspects on which lingual functioning foundationally depends, including the physical. I have explained in detail what is going on in a general computer, what happens in programming (or encapsulating knowledge in a computer) including chatbots.

To me, ChatGPT could be a useful tool to help me gain material for my own thinking. But I would have to be careful that what it gives me is true and valid, and not an 'hallucination' - but that is my responsibility. (However, currently I am boycotting ChatGPT, for moral reasons to do with the cheating done by the owner of the company.)

Back to Richard's appeal to make use of research funding: Yes indeed. The above approach needs development, both philosophically and empirically. Could we work together on this? Who would be interested? Nick?

Hope this helps. I'm sorry about its length. Please forgive any spelling or other mistakes; I have not managed to read it through fully before having to send it off.

Andrew 17 April 2023

----- Appendix: A More Detailed Explanation Of The Levels Or Aspects Of Computer

In its PHYSICAL ASPECT, the computer memory is a spatial array of conductors and surrounded by insulating material so that electric charge remains and does not leak away. Each such conductor has next to it a special kind of material that can be conductor or insulator. That special material is semi-conductor and sometimes acts as conductor to allow charge in or to drain charge away. Which it does depends on the state of electric charge next to it somewhere else.

From the perspective of the PSYCHICAL ASPECT, each such conductor that stores charge is called a bit, and the bit or conductor is in a certain state. If it has sufficient electric charge (e.g. to have a voltage of 3 volts) then we say the bit is 'on' or '1'; if not, we say the bit is 'off' or '0'. (Those labels, on, off, 1, 0, are mere conventions and are not actual numbers etc.) Now, 8 bits taken together is called a byte, and a million such bytes is called a megabyte, etc. So, seen from the physical aspect, a megabyte refers to the charge-state of 8 million such conductors, each able to retain electric charge without leaking. Why 8? Convention that has proven useful. The state of bit-value of 8 bits is a bit pattern. For example, 00000000, 11111111. 01010101, 011000001, 01101010, etc. There are 256 different possible bit patterns in a byte. Four bytes together (32 bits) can have 4 billion different bit patterns.

From the perspective of the ANALYTICAL ASPECT, of disticnt concepts, we call the bit pattern 'data'. We see the bit pattern as standing for something. What it stands for depends on the coding system we choose to use. Let us think of three examples of these. They are all conventions, not necessary truths.

- Quantities, using binary coding, which is used to represent numbers. The first bit (right-hand end of the bit pattern above) is taken to represent the quanity 1, the next, the quantity 2, the next, 4, then 8, 16,32,64 and 128. (all powers of 2). The pattern of a byte represents various of these numbers added together if their bit is on or 1, but not counted if it is off or 0. Thus 00000000 = numeric 0, 11111111 = 255, 01100001 = 97 (64+32+1), and so on. An alternative coding convention reverses this, the left-most bit representing 1 and the right-most, 128, so 01100001 = 134 (128+4+2). To get quantities more than 255 we use more than one byte together; four bytes can represent 4 billion different quantities.

- Letters of alphabet and other characters. By the convention known as ASCII (American System for Coding Information Interchange), 01100001 = the letter 'a' and 00100001 the letter 'A', 01100010 = 'b' and 00100010 = 'B', and so on. 00010000 = . The digit '1' (as distinct from the quantity 1) is 00110001.

- Colours. Almost all colours can be seen as a combination of amounts of red, green and blue light. To represent a colour we use 3 bytes together in which the first byte is the quantity of red, the second, green and the third, blue. So in numeric terms, 255,0,0 is pure red, 128,0,0 is a darker pure red, 255,255,0 is a pure yellow (red and green), 0,0,128 is a darkish blue, 255,255,255 is white, 0,0,0 is black, and so on. Or, in bit patters, yellow would be 11111111,11111111,00000000.

- Similarly for sounds, logical values, encryptions, ledger entries, web links, and so on - many different kinds of data.

How bit patterns are treated as concepts or pieces of data depends on the convention we use. It is made possible by the analytical aspect that we can (a) choose which convention for interpreting a bit pattern as a piece of data, (b) allow bit patterns to represent distinct values of these data. Data is not reducible to bit patterns, nor vice versa; the link between them is human convention.

From, the perspective of the FORMATIVE ASPECT, of human shaping, we see the structure that links pieces of data together, and the processing that goes on. So, several basic pieces of data representing letters can represent a word. The five basic pieces of data with character values h,r,t,t,u arranged together can represent the word "truth". (However, that ordering is so simple that we could treat words like "truth" as a whole piece of data rather than as a composite of letters, that is, of the ANALYTICAL ASPECT; thus enters the idea of vocabulary, a set of pieces of data that are valid.)

The FORMATIVE ASPECT comes into its own when we have more complex structuring. Several of words arranged in order give a sentence, e.g. "I am the truth." And so on. Such a sentence, seen from the analytical aspect is merely a collection of 15 letters or punctuations, and from the psychical aspects of bits would be 15 byte-sized bit patterns - the bit pattern for the whole sentence would consist of 8*15 = 120 bits. Simple text like that is very simple; less simple is an array or table, and even less simple is a family tree or a novel. Diagrams are yet another structure, in which the pieces of data are not letters but shapes. A map is a very complex kind of diagram.

All these are data structures, and it is the formative aspect at which we 'see' the structure among all the pieces of data.

From the perspective of the LINGUAL ASPECT, we take account of the signification of the words and sentences. So, "I am the truth." would be seen as something to do with truth and something to do with a person, and "the" would indicate one not many truths.

I believe that it is at this level, this aspect, that most programs operate, for which the laws of an aspect are encapsulated. It is at this level that Google search is able to see when one has made a spelling mistake and correct it and suggest the correction. For example, I asked

"What port fpt?"

and Google replied

"Do you mean 'What port ftp?'?"

It found out that fpt was not a valid piece of data in its vocabulary (analytical aspect), reckoned there was a mistake, such as two letters wrong way found, and found which would fit (ftp), when "port" is taken to mean its computer meaning of something by which a computer receives stuff over the internet. However, if "port" is place for shipping, then "fpt" might have a different error, such as an abbreviation of the name of such a port. So sort out such things requires operation at the lingual aspect, not just formative, though it is also operating at the other aspects too.

At the physical perspective, all that is seen as electric charges.

----- end of appendix

[Then followed the email conversation]


* Basden A. 2008. Philosophical Frameworks for Understanding Information Systems. Hershey, PA, USA: IGI Global (IDEA Group Inc.). ISBN: 978-1-59904-036-3 (hbk), 978-1-59904-038-3 (ebk).

* Basden A. 2017/2018. Foundations of Information Systems: Research and Practice. Routledge. ISBN: 978-0-367-870-300 (pbk), 978-1-138-79701-7 (hbk), 978-1-138-75748-3 (ebk)

Significance: A monograph that brings together five major areas of concern in the field of digital or information systems, so they may be considered together. To do this it employs Dooyeweerd's philosophy. Taking the initial ideas in Basden (2008), it sets them in the context of extant ideas - over 50 extant discourses are discussed, showing how Dooyeweerd might help resolve their more challenging problems. The approach has been to listen to the discourses, affirm their core meaningfulness, critique their presuppositions and enrich them. It is found that Dooyeweerd's philosophy has different resonances in the five areas, but that his idea of aspects pervades and enriches all. At the end are collected together over 100 specific research projects that may be carried out in the field of information systems - which may be found within The Dooyeweerd Pages. Its uniqueness and importance is as follows:

This page, "", is part of a collection that discusses application of Herman Dooyeweerd's ideas, within The Dooyeweerd Pages, which explain, explore and discuss Dooyeweerd's interesting philosophy. Email questions or comments are welcome.

Written on the Amiga and Protext in the style of classic HTML.

You may use this material subject to conditions. Compiled by Andrew Basden.

Created: 18 April 2023 from email exchange. Last updated: