Unpredictability and Unintelligibility

Einstein's famous dictum that 'God does not play dice' might be a projection of a problem with the semantic grounding of scientific (or any...) theories.

In order to make sense to one another, we must (at least) be able to predict, to some extent, how honest and competent interlocutors will use the words we use to do this. For Wittgenstein, this predictability of use was what we really meant by 'meaning'.

If the world is arranged in such a way that two participants in a conversation might be using a word correctly but still semantically diverge, then the consequent confusion cannot be resolved. We might imagine that a fundamentally stochastic universe could be like this - that two experimenters might correctly give conflicting answers to a question about, for example, when a uranium atom emitted an alpha particle.

This is not, of course, what happens. While they might only be able to agree on a probability distribution beforehand, we do not expect them to disagree on the actual timing of the event afterward. The 'Many Worlds' interpretation of quantum mechanics, and possibly other holistic approaches to it, avoids the problem by ensuring that interlocutors who might give different accounts of the actual event can never converse, so no semantic incoherence can arise.

We might wonder whether any solution of this kind to the problem can really be satisfactory. My instinct is that Einstein would have thought not.

But we need to be wary of instinct here - both mine and Einstein's. We are inclined to treat the world as semantically transparent - as though it was possible to be quite explicit about the meanings and uses of the words we use to give our accounts of it. But we know that this heuristic leads us into confusion. When we try to be (exhaustively) explicit about the meanings of words we find that we need to be explicit about the meaning of our explicit account as well, and so run into an open question problem. Some minimal ambiguity must remain. So long as it is not 'methodologically' relevant, it may do no harm ...

The appearance and persistence of the heuristic of semantic transparency is not surprising - it's a productive metaphysical posture, for one thing, and many specific ambiguities turned out to be resolvable, for another.

Once again, however, we mustn't imagine that a specific grammatical structure reflects an independent metaphysical substrate. Semantic transparency is a product of our language use, not a grounding for it. While a world in which radical semantic divergences were irresolvable would not support language users, we cannot draw any conclusions about this beyond that we do not live in such a world.

In particular, we can't say now how we might resolve future threats of semantic divergence. Perhaps by reviewing what we mean by 'playing dice', for instance ...

Epistemological Empiricism - an Alien Experiment

I may have posted this before, but I came across a copy of it, and wanted to make sure:

No adequate answer to the question of how we validate the knowledge-acquisition role of sensory input has ever been found.  It may be a “mad” (Wittgenstein) question, but it is very hard – embarrassingly hard, for traditional sensory empiricism – to say why.  And it needs to say why, if it is to give the kind of account of knowledge acquisition suggested by the paradigm case.

In fact, however, it is quite easy to demonstrate that no account of this kind can be given.  Consider the following thought experiment:

The central characters are some newly discovered aliens.  These aliens actually look surprisingly human, except for some obvious differences that I will come to.  They walk on two legs, they speak languages that we can translate and they have many social, technical, and artistic accomplishments which, though superficially strange to us, are entirely recognisable to us for what they are.  They are our new friends in a friendless universe.

The difference between these aliens and ourselves is that they have no discernible sense organs of any kind, and their language is entirely devoid of elements that depend on or refer to sensory perception.  Instead of saying “I see the painting on the wall” they say, instead, “There is a painting on the wall”.  Instead of “I hear beautiful music playing” they say “There is beautiful music playing”.  If we ask them how they know about these things, they are entirely puzzled - surely they just “know” them.  How can there be a process preceding knowledge?

After some interaction with us, and a great deal of epistemological discussion between their philosophers and our philosophers, we are able to establish the following facts about their knowledge-acquisition processes:

(1) They are likely to know more about things that are physically close to them than about things that are far away, although there are many exceptions to this rule.

(2) They sometimes believe that they know things, and later turn out to be wrong.  Aliens that do this a lot receive community support from other Aliens.

(3)  Some of them have specialist knowledge - they know things that others do not.  Some who have specialist knowledge are acknowledged among their kind as having unusual talents.  Others have it because they have spent a lot of time thinking about things, talking to aliens that have different knowledge from theirs, and working out new relationships between different kinds of knowledge.

Perplexed by this limited understanding our philosophers write to their philosophers with the following helpful advice.  They say:

"It is not possible to know things directly - you have to find them out somehow.  There must be some process by which you acquire knowledge from the world.  Look at us - we have eyes to see the world with, ears to hear it with, we can taste it, smell it, and feel it.  This is how we know about the world.  If we were like you, and couldn't account for our knowledge, we would be extremely insecure - we would be afraid of getting things wrong.

"We advise you to discover the basis of your knowledge - there must be some hidden connection between the world and your understanding, some sensory apparatus, perhaps complex and mysterious, which you and we have not yet discovered.  Until you discover it, your knowledge must always be qualified and uncertain.  Put your best brains on it immediately!"

At the same time as our philosophers are licking the stamp on this communication and putting in the intergalactic post, their philosophers are sending us the following message:

"We know about your sensory apparatus - you have ears to hear with and eyes to see with.  You can taste smell and feel the world around you.  But you are entirely mistaken about the role these have in knowledge determination - surely you must understand that knowing comes before sensing, that your sensory apparatus would be useless to you if you didn't already know that it was reliable.  Without a secure knowing, your senses would provide you only with unreliable sensations - not much better than entertainment.

“And you cannot found your knowledge of the reliability of your senses on the evidence of your senses - this would be a non-sequitur of the most elementary kind.  Only on secure knowledge can you build other knowledge.

“This is obvious to the youngest child among us!"

On receipt of these pieces of advice, the philosophers on both sides fall into two different camps.  One camp says:

"Those people are defective.  They cannot have real knowledge of the world because they [humans: can't relate it to any reliable sensory input / aliens: think sensory input is more reliable than secure knowledge].  They may appear to be making sense, but we know their heads are filled with nonsense because [humans: their knowledge isn't supported by anything / aliens: they allow the most egregious question-begging to pass as reliable theory]."

The other camp says:

"Despite appearances, and entirely mysteriously, we know from the way they talk that these people seem to have acquired real knowledge of the world, despite the fact that [humans: they cannot base it on sensory input / aliens: they think sensory input is more reliable than definite knowledge]."

And, of course, it is the second camp that has it right.

Metaphysical Anxiety - the 'Inside' and the 'Outside'

'Traditional' epistemology has often concerned itself with the reliability of experience as a source of understanding, rather than directly with the reliability of our scientific theories. The link between these has been taken for granted by empiricists, a position that is congruent with a kind of 'common sense' picture of how our experiences and our conversations are related.

I have argued, through this blog, that this perspective is profoundly wrong. It is wrong for reasons which empiricists have been struggling with since the seventeenth century. In particular, it is wrong because it generates an open question paradox that cannot be resolved without doing serious harm to what we mean by 'experience', 'knowledge', 'truth' and 'meaning' - harm that quickly spreads to the intelligibility of any conversation in which these words are used.

It's wrongness, however, is less interesting than why - despite its wrongness - it has been so compelling. There seem to be two reasons for this - a positive one and a negative one.

The positive one is our visceral experience of the relationship between our internal ('phenomenological')  states and what we say. From a primitive expression of pain through to the articulation of an abstract theory, the 'internal' seems to drive the 'external'. In addition to this, we have a strong intuition that honest and informed theorising is linked to a definite set of internal circumstances and attitudes. Truth-telling and truth-living seem to be connected.

The negative one is that it seems hard - or has seemed hard - to otherwise account for the 'reliability' of our scientific theorising. If it is not grounded in 'experience' then in what is it grounded?

The occasional unreliability of experiences, and the occasional confusions that arise in our conversations, show that the problem has serious practical aspects.

David Hume spells out the dilemma very clearly - pointing out that while we cannot demonstrate that our sensory inputs - our 'impressions' - provide reliable knowledge, we nevertheless are compelled to rely on them. He also strictly insists on their being the only test of scientific knowledge - drawing a clear distinction between our knowledge of formal relationships between concepts - as found in mathematics and logic, for instance - and knowledge acquired via the senses (which includes scientific knowledge). The former can be demonstrated, and the latter we are compelled to rely on but cannot demonstrate. 

This distinction between 'analytic' statements and 'synthetic' statements was part of the bedrock of the Anglo-American analytic tradition in philosophy until well into the 20th century, when it was recognised as having two flaws:

The first, courtesy of Willard Quine, is that it is often contextual. A statement that can work as a principle of argument in one context (can seem to be analytic) can look more like a statement of fact (can seem to be synthetic) in another. Think, for instance, of conservation laws in physics - are these descriptive of the world, or are they principles that must be recognised in any physical argument?

The second flaw is that the distinction itself is neither an analytic nor a synthetic truth. It cannot be shown to be a consequence of formal reasoning, nor can it be shown to depend on experience. This has the consequence that either there must be some other ground for reliable knowledge, or that we cannot be sure that Hume was right. In fact, we can be quite certain that we can legitimately entertain the possibility that he was not, for reasons that are not analytic.

In this blog, I have argued that we do not need to 'validate' our experiential substrate, nor its relationship with empirical theorising, in order to validate the empirical theorising itself. We validate our theorising by grounding it on the possibility of theorising - upon those statements whose rejection cannot be part of a playable language game (Barry Stroud's 'privileged statements' - although he did not recognise their foundational importance). These cannot be restricted to 'rules of the game' (analytic truths), since we cannot make the analytic/synthetic distinction in a way that would support this. And since it is a fact about the world that we can theorise about it, so it would be a fact about the world if we could not, and so unintelligible.

The root of empirical theorising is not some link to internal experience, but those statements which must be true for empirical theorising to be possible. And, since 'empirical theorising is not possible' is, essentially, an empirical theory (about the knowability of the world), some of these statements will have the character of universally necessary truths.

To those who might, with Barry Stroud, point to the possibility of necessary truths outside the scope of what can be articulated in language, I have to ask, how can these be identified? How can we consider them, accomodate them in our theorising, without articulating them? What are they?

But Stroud's problem points back to a deeper issue that I want to address: What do we do with the phenomenological 'space' - the internal world that we inhabit, and from which we hurl our words at one another? What can we do with the possibility that this internal world may become unhitched from the public world, and that we might become unintelligible to others simply as a consequence of honestly recognising our own experiences and feelings?

This is exactly the experience, I think, that many people suffering from severe (and perhaps not so severe ...) mental illnesses have. Perhaps we hope that a true grasp of rational theorising might preserve us from this, and justify our conflation of  'madness' and 'irrationality'.

To try to get a proper - though possibly not completely reassuring - grasp of what is going on here, it's worth making a few observations.

The first is that relying on our internal processes to keep ourselves publicly intelligible must work pretty well on the whole. An empirical theory to this effect might not even be hard to test - arguably we are engaged in just such an investigation as we move through the world and experiment with communication strategies. No empirical theory can count as an argument for the reliability of empirical theorising in general, of course (pace Quine), but this is not what we are looking for here - we can ground empirical theorising elsewhere. As Hume suggests, the qualified metaphysical status of our 'impressions and ideas' doesn't make them generally misleading.

And as Donald Davidson (more or less) points out, they can't be generally misleading or we wouldn't have a language that worked well enough for us to speculate to that effect.

The second is that there is a relationship between the language acquisition process and the apparent 'boot-strapping' of our capacity for truth-telling from the fact that we can talk to one another. In the early pages of the Investigations, Wittgenstein draws attention to the question-begging involved in accounting for language acquisition on the basis of some proto-linguistic communication device (such as pointing at things). He also explores at some length the experience of learning new meaning rules - how they suddenly just 'make sense', so that 'we know to carry on'. There seems to be no underlying 'language of thought' or 'micro-program' that we use to calculate what we should say - and even if there seemed to be, it wouldn't help us here.

(Quine, of course, also remarks on the essential ambiguity of proto-linguistic devices such as pointing: What, exactly, is the teacher 'pointing' at?).

Wittgenstein suggests a way of thinking about this that anyone who has engaged with a small child learning to speak would find very recognisable: he points out that there are things that can be shown, but not said. Children learn to speak by being spoken to and listened to - not (obviously) by taking instruction.

Just as we do not need our internal states to ground our knowledge claims, neither do we need an account of this process that demonstrates the reliability of the child's language acquisition. The child's capacity to speak is discovered and demonstrated within the conversations we might have with it about this. We will either find it intelligible, or we will not - just as we do with others we are assessing as potential interlocutors.

And just as with other potential interlocutors, there is no 'matter of fact' about their general intelligibility. If we don't find someone intelligible, we can never know whether this is because they 'really' don't make sense, or because we haven't tried hard enough (haven't found the right way to talk to them).

The experience of the child - as with anyone exploring a new communication strategy - will be of trying to 'get it right'; of trying to find out 'how to go on'. What it is not is an experience of trying to master the explicit dictionary and grammar of a demonstrably intelligible language, in order to express a coherent internal reality. (The task is enormously complicated, of course, by the way the child's experiences of its experiments feeds back into that internal reality.)

So: while we are convinced that these internal processes must validate as well as inspire, we must (with Hume) be satisfied with just inspire. But the failure of one validation model does not imply that there is no validation model. We experiment with how to talk to one another, and so experiment with validation models at the same level as we experiment with stories about the world.

Our language does not depend upon our having a shared phenomenological space, nor even upon our phenomenological spaces being somehow commensurate beyond what is required for us to speak to one another. It is only our ability to speak to one another that convinces us that there is a shared space, and it is only within our conversations that we can give meaning to that conviction and to its content.

Naming and Metaphysics

And, of course, there is this:

We can only determine that 'A' and 'B' are not two names for the same thing if we can account for the way we distinguish between the 'things' they might be names for. This doesn't reach 'outside the language'. It's just like the game of 'Pangolin' - Is there a question we can ask about something that must have a different answer for A than for B?

If there is no such question, we have no grounds for denying identity. To claim that there may be a 'real' distinction that we cannot discern is no more than to claim that we may have to review our conclusion in the light of new circumstances. If the possibility of review is unintelligible, then the identity is necessary.

Meaning, Use, and Method

Wittgenstein's conflation of meaning and use, and Davidson's (and others') conflation of meaning with 'truth conditions' could be mashed together into an approach that yields some interesting results.

The truth-condition account of the meaning of an assertion could be generalised in terms of 'instructions for use'. We could say 'The right time to say "It is raining in Nebraska" is when it is raining in Nebraska'.

This kind of account would allow us to go beyond assertions - so long as we can either show or explain the use of an expression, we could say that we can introduce someone to its meaning.

Now imagine two circumstances that we want to distinguish - circumstance A and circumstance B. We can only succeed practically if we can show that there is a methodological (in the broadest sense) wedge that we can drive between them - if we can show someone a way of discovering, for instance, that it is A, and not B, that we have an example of on a particular occasion. If there is no methodological wedge, and we still insist that a distinction can be made, we have nothing to point to. (A distinction based on naming itself would be question-begging.)

(Once, when I was visiting a university in Beijing, a local member of staff showed me a sculpture that they described as 'the pigeon of peace'. I struggled to think of how I might explain their 'mistake' to them. In the end, I didn't try - but various 'use' type explanations have occurred to me since ...)

The point of all this is that if I cannot communicate use distinctions to you, then I cannot communicate meaning distinctions either.

This has some important consequences - for instance, for whether we can distinguish between:

"The world is the way we say it is", and

"The world allows us to talk about it in the way that we do."

Semantic Experiment and Artificial Intelligence

We use 'linguistic tokens' in two distinguishable ways:

(1) In accordance with pre-existing grammatical and semantic rules.

(2) Experimentally - by exploring new possible rules and uses.

It is possible to imagine building a definite model of (1) for some particular state of our language use, but it is far from clear that modeling (2) is possible.

It is, in other words, possible to imagine programming (1), but not (2).

Human intelligence does both, and while there might be some sense in which a sophisticated self-learning algorithm might also do both, our grounds for thinking it might do this in a 'human-like' way are very weak. We don't have the appropriate kind of understanding of human intelligence to build an algorithm that can be guaranteed to behave like a human being, that would never diverge in some monstrous way.

Nature, after all, struggles to avoid monstrosity: and nature works more slowly, and has fewer combinatorial options available to it, than our AI experimenters do.

Turing avoided this difficulty by putting 'human intelligence' on both sides of the equation - we can't specify it, but, as human beings, we know it when we see it.

Until we don't ...

Grounds of Meaning

Imagine that I said to you "The only sentence with any meaning is the one you are reading now, and it means exactly what it says."

This has to be false, a kind of nonsense. Apart from anything else, it excludes the possibility of any statement explaining the meaning of the 'only meaningful statement' having any meaning.

Why does this incoherence disappear when we are dealing with a whole linguistic practice?

For instance, "Only statements which are part of practices we recognise as language use have any meaning, and they mean exactly what they say" (or some suitably qualified equivalent) must be literally true.

There are some differences between these statements, though:

The first refers to a closed group - it has only one member. The second is not closed - what counts as a 'statement which is part of our linguistic practice' is not determinate. This can, in fact, be discussed and adjudicated upon within the linguistic practice. (It can only be discussed and adjudicated upon within the practice.)

What is involved in a statement 'meaning exactly what it says' is not unambiguous. (Remember the 'Brexit means Brexit' sleight of hand). It's like the joke about a balloonist who lands in a field in the fog, and asks a passer-by where they are, getting the reply: "You're in a field".

We can only have an unrestricted capacity to make meta-linguistic adjudications (about meaning, truth, etc.) in an 'open' language - this is exactly why formal languages are always so disappointing. Not to say useless for any sort of serious enquiry ...

Crucial enquiries nearly always have a semantic element - incorporate some (tacit or explicit) investigation of meanings. Whether or not some specific statement 'means exactly what it says' is always open to interrogation.