Floridi and Williamson on AI; Bohm and Simondon on Thought

There's a very interesting interview here https://conditiohumana.io/floridi-interview/?utm_campaign=direct with Luciano Floridi about AI. Floridi's seminal work on the nature of information, although rather too cognitivist for my liking, is something that one cannot think without today. His contribution to the ethics of information, which sees information ethics as a variety of environmental ethics, is highly important. Floridi has been talking a lot about AI, and in the interview he proposes that the most interesting aspect of AI is that it is used as a mirror of nature: that we come to know ourselves through the technology. I agree. I would go further to say that the essence of information lies in intersubjective engagement (and by extension, consciousness), not in some abstract "stuff" that exists between us. The power of "information" as a topic is that it gives us all something to talk about, where everyone is uncertain about what it is they are trying to grapple with. It's all rather scholastic - and I quite like that.

AI is, I think, also like this. It is a shared disruption to our ways of thinking which gives us all something to talk about. When we see AI as a "tool", we get it wrong. That our institutions see AI as a "tool" says something about our institutions, with their rigid hierarchies, than it does about the technologies of AI.

Ben Williamson's post here, https://codeactsineducation.wordpress.com/2019/02/01/education-for-the-robot-economy/ articulates some of the institutional problems with AI. Here the institution in question is the OECD, but really it could be anyone. They are all struggling to maintain their position and status in a world which is being turned upside down by technology. And it is interesting that "education" becomes the sticking point - the point at which these large hierarchies focus on to say "this is what we have to do". As if they know! As if anyone knows! As if the cult of expertise has escaped the massive explosion of options that technology has given us. As if expertise itself isn't under threat from technology. Which it is.

I am having a personal reminder of this, because last week I self-published my book "Uncertain Education". Since I've been writing it, or thinking about it, for nearly 8 years, it was time to go public with a document that bore the scars of its gestation. I self-published with a combination of Overleaf for typesetting (in Latex) and Blurb for printing and distribution. Both work very well, and the printed result is indistinguishable from a normal printed book (even printed at Lightening Source, which also prints "ordinary" books for Amazon). The print-run thing is over. And with it, the artificial scarcity of the "final document". Everything can be changed very easily in an agile way.

So what of the expertise of the editor, the typesetter, the reviewer, etc? The cloud takes over. The expertise becomes distributed. Many eyes looking at this thing, alongside my own eyes which see a thing now in the environment which once only existed within my own private world, are a powerful driver for making small incremental improvements. What matters are the ideas, and they tend to survive awkward moments.

The cult of the expert is one of the reasons that education maintains its structures and practices and its hierarchies. It is because a teacher is seen as an expert to mark a piece of work that we have double-marking, exam boards, quality procedures, and so on. Individual teachers are not trusted, so there has to be a cumbersome mechanism to keep everything in check to ensure that the stamp of quality can be granted. So what if we do Adaptive Comparative Judgement on a cloud-scale for marking student work? What if we create distributed databases of judgements from peers and teachers all over the world about the quality of work? This is what technology affords. It's not AI as such. And yet, its fundamental mechanism is the essence of what Warren McCulloch realised his neural networks were: a heterarchy (see https://vordenker.de/ggphilosophy/mcculloch_heterarchy.pdf)

It is the cult of the expert which drives the OECD to make proclamations of scarcity about education. They declare knowledge to be scarce, and so maintain the fiction of the "knowledge economy" - what do they mean by "knowledge"? What do they mean by "economy"? They declare "coding skills" to be scarce - really? They declare the "right metrics" to be scarce, without any consideration as to what a correct data analysis might be. Worst, they essentially declare "being human" to be scarce. What nonsense! Why do they do this? Because they want to keep themselves in business.

Take the expert out of all of this and the system reorganises itself naturally and heterarchically. There is no scarcity of knowledge. There is no scarcity of metrics, because every metric is merely an alternative additional description of reality, not a commandment. There is no knowledge economy because what matters is not what is known, but the uncertainty that accompanies it. Coding itself is merely a technique for amplifying artificial descriptions of the world and creating objects and new options to act. It is not scarce either.

What are we left with? It's very similar to my process of publishing and gradually improving my book. It is moving away from the objects of knowledge - final statements, artefacts, etc - and moving towards expressing thought as a process. There's a lot of stuff in my book on David Bohm's ideas about dialogue. How right I think he was. Dialogue is about inspecting thought as process, because all the stuff around us is produced by thought. Organisations like the OECD (and our universities for that matter) have become pathological because they do not see themselves as the product of thought. But they are.

If Bohm is right, then so too is Gilbert Simondon. Thought is transduction - the process of making and maintaining categories. The objects that we have are the result of transductions being configured in a particular way. If we want a better world, we need to change our transduction processes. Simondon's genius is to see that the highest levels of human development are tied up with the realisation of the capacity to control the transductions which make us "us". Particularly, it is the capacity to make us "us" - the capacity for individuation - within a technological environment, which is at the heart of the educational and technological challenge of our time.

Artificial Intelligence in a Better World

There's an interesting article in the Guardian this week about the growth of AI and the surveillance society: https://www.theguardian.com/technology/2019/jan/20/shoshana-zuboff-age-of-surveillance-capitalism-google-facebook?fbclid=IwAR0Nmp3uScp5PNzblV2AkpnQtDlrNIEDYp54SdYa4iy9Ofjw66FgDCFceO8

Before reading it, I suggest first inspecting the hyperlink. It's to theguardian.com, but the file it seeks is  "shoshana-zuboff-age-of-surveillance-capitalism-google-facebook?fbclid=IwAR0Nmp3uScp5PNzblV2AkpnQtDlrNIEDYp54SdYa4iy9Ofjw66FgDCFceO8" which contains information about where the link came from and an identifier to my account. This information goes to the Guardian, who then exploit the data. Oh, the irony!!

But I don't want to distract from the contents of the article. Surveillance is clearly happening, and `Platform capitalism' (Google and Facebook are platforms) is clearly a thing (see Nick Snricek's book here: https://www.amazon.co.uk/Platform-Capitalism-Theory-Redux-Srnicek/dp/1509504877, or the Cambridge Platform Capitalism reading group: https://cpgjcam.net/reading-groups/platform-capitalism-reading-group/). But the tendency to reach the conclusion that technology is a bad thing should be avoided. The problem lies with the relationship between institutions which are organised as hierarchies trying to cope with mounting uncertainties in the world which have been exacerbated by the abundance of options that technology has given us.

In writing this blog, I am exploiting one of the options that technology has provided. I could instead have published a paper, written to the Guardian, sent it to one of the self-publishers, made a video about it, or simply expressed my theory in discussion with friends. I could have used Facebook, Twitter, or I could have chosen a different blogging platform. In fact, the choice is overwhelming. This amount of choice is what technology has done: it has given us an unimaginably large number of options for doing things that we could do before, or in other ways. How do I choose? That's uncertainty.

For me as a person, it's perhaps not so bad. I can resort to my habits as a way of managing my uncertainty, which often means ignoring some of the other available options that technology provides (I really should get my blog off blogger, for example, but that's a big job). But the sheer number of options that each of us now has is a real problem for institutions.

This is because the old ways of doing things like learning, printing, travelling, broadcasting, banking, performing, discussing, shopping or marketing all revolved around institutions. But suddenly (and it has been sudden) individuals can do these things in new ways in addition to those old-fashioned institutions. So institutions have had to change quickly to maintain their existing structures. Some, like shops and travel agents, are in real trouble - they were too slow to change. Why? Because their hierarchical structures meant that staff on the shop floor who could see what was happening and what needed to be done were not heard at the top soon enough, and the hierarchy was unable to effect radical change because its instruments of control were too rigid.

But not all hierarchies have died. Universities, governments, publishers, broadcasters survive well enough. This is not because they've adapted. They haven't really (have universities really changed their structures?). But the things that they do - pass laws, grant degrees, publish academic journals - are the result of declarations they make about the worth of what they do (and the lack of worth of what is not done through them) which gets upheld by other sections of society. So a university declares that only a degree certificate is proof that a person is able to do something, or should be admitted to a profession. These institutions have upheld their powers to declare scarcity. As more options have become available in society to do the things that institutions do, so the institutions have made ever-increasingly strong claims that their way is the only way. Increasingly institutions have used technology as a way of reinforcing their scarcity declaration (the paywall of journals, the VLE, AI, surveillance) These declarations of scarcity are effectively a means of defending the existing structures of institutions against the increasing onslaught of environmental uncertainty.

So what of AI or surveillance? The two are connected. Machine learning depends on data, and data is provided by users. So users actions are 'harvested' by AI. However, AI is no different from any other technology: it provides new options for doing things that we could do before. So while the options for doing things increase, uncertainty increases, and feeds a reaction by institutions, including corporations and governments. The solution to the uncertainty caused by AI and surveillance is more AI and surveillance: now in universities, governments (China particularly) and technology corporations.

This is a positive-feedback loop, and as such is inherently unstable. It is more unstable when we realise that the machine learning isn't that good or intelligent after all. Machine learning, unlike humans, is very bad at being retrained. Retrain a neural network then you risk everything that had been learnt before going to pot (I'm having direct experience of this at the moment in a project I'm doing). The simple fact is that nobody knows how it works. The real breakthrough in AI will come when we really do understand how it works. When that happens, the ravenous demand for data will become less intense: training can be targetted with manageable and specific datasets. Big data is, I suspect, merely a phase in our understanding of the heterarchy of neural networks.

The giant surveillance networks in China are feeding an uncertainty dynamic that will eventually implode. Google and Facebook are in the same loop. Amplified uncertainty eventually presents itself as politics.

This analysis is produced by looking at the whole system: people and technology. It is one of the fundamental lessons from cybernetics that whole systems have uncertainty. Any system generates questions which it cannot answer. So a whole system must have something outside it which mops up this uncertainty (cyberneticians call this 'unmanaged variety'). This thing outside is a 'metasystem'. The metasystem and the system work together to maintain the identity of the whole, by managing the uncertainty which is generated. Every whole has a "hole".

The question is where we put the technology. Runaway uncertainty is caused by putting the technology in the metasystem to amplify the uncertainty mop. AI and surveillance are the H-bombs of metasystemic  uncertainty management now. And they simply make the problem worse while initially seeming to do the job. It's very much like the Catholic church's commandeering of printing.

However, the technology might be used to organise society differently so that it can better manage the way it produces uncertainty. This is to use technology to create an environment for the open expression of uncertainty by individuals: the creation of a genuinely convivial society. I'm optimistic that what we learn from our surveillance technology and AI will lead us here... eventually.

Towards a holographic future

The key moment will be when we learn exactly how machine learning works. Neural networks are a bit like fractals or holograms, and this means that the relationship between a change to the network and the reality it represents is highly complex. Which parts of a neural network do we change to produce a determinate change in its behaviour (without unforeseen consequences)? What is fascinating is that consciousness and the universe may well work according to the same principles (see https://medium.com/intuitionmachine/the-holographic-principle-and-deep-learning-52c2d6da8d9). The fractal is the image of the future. The telescope and the microscope were the images of the enlightenment (according to Bas van Fraassen: https://en.wikipedia.org/wiki/Bas_van_Fraassen)

Through the holographic lens the world looks very different. When we understand how machine learning does what it does, and we can properly control it, then each of us will turn our digital machines to ourselves and our social institutions. We will turn it to our own learning and our learning conversations. We will turn it to art and aesthetic and emotional experience. What will we learn? We will learn about coherence and how to take decisions together for the good of the planet. The fractals of machine learning can create the context for conversation where many brains can think as one brain. We will have a different context for science, where scientific inquiry embraces quantum mechanics and its uncertainty. We will have global education, where the uncertainty of every world citizen is valued. And we will have a transformed notion of what it is to 'compute'. Our digital machines will tell us how nature computes in a very different way to silicon.

Right now this seems like fantasy. We have surveillance, nasty governments, crazy policies, inequality, etc. But we are in the middle of a scientific revolution. The last time we had the Thirty Years War, the English Civil War and Cromwell. We also have astonishing tools which we don't yet fully understand. Our duty is to understand them better and to create an environment for conversation in the future which the universities once were. 

Spooky Christmas!

Here are some improvisations I did this Christmas. I'm finding myself increasingly drawn to the creativity that the Seaboard offers me than to what I can do with the piano. This feels like a real change for 2019...

Communicative Musicality and Entropy

An activity which I've done a few times now is to invite people to simulate conversation in music using my Roli Seaboard. The result sounds a bit like the Clangers, but people enjoy it (and I get something much more expressive out of them than I would if I tried to get them to vocalise "talk"). What's interesting is that it's obvious when its done well - the musical conversation has a kind of coherence about it - which raises a fundamental question about the coherence of any shared "musicking" between people, and the prosody of language.

This coherence is, I think, a fractal structure, and another aspect of this analysis is to consider how the fractal might emerge. Any musical communication involves an "emergent alphabet" of utterances, which invites the suggestion that a fractal must express the emergence of this alphabet. The emergent alphabet issue also presents a problem when trying to apply analytical techniques like information theory to music: information theory relies on the fact that the alphabet is known at the outset, so it knows what to count. If the alphabet is emergent, it doesn't know what to count until (possibly) the end. However, it is a possibility that an alphabet at time t1 has a similar structure in terms of entropy as a larger alphabet at t2. This is where the fractal likely resides.

The critical question is at what point is it necessary to expand the alphabet? This is closely related to the question as to "at what point is a new concept introduced into a learning conversation?" My answer to this draws on the relations between entropies of basic elements.

Consider that a "basic" alphabet in sound contains four elements: pitch, rhythm, intervals and volume. They can be A, B, C, D. Over time, the entropies for each of these values can be calculated, and at different times the entropy for each  will either increase or decrease. For example, a note which is sung typically has an "attack" - which is an increase in volume, and hence an increase in the entropy of volume. It may then have a "sustain" period where the volume is constant: that gives a entropy closer to zero. Finally the note is released, which results in a rapid decrease in volume to nothing, which is also an increase in entropy. Every dimension is like this, having a period of increase and decrease, so for each of A, B, C, D there is a corresponding A*, B*, C*, D* for its inverse. This means that AA*, BB* CC*, etc are all effectively zero. In drawing the attack and decay of a note and encoding  an increase in entropy as 1, and a decrease as 0, we might see:

A   A*
1    0
0    1
0    1
1    0

Sometimes it may seem that the entropy of A oscillates very quickly with the entropy of A* (e.g. vibrato), or even that it is difficult over a period of time to determine whether on average there is an increase or a decrease: both seem to be simultaneously present. If we draw this then we might see:

A   A*
1    0
1    0
0    1
1    1
1    0
0    1
1    0
1    0

taken over a longer period of time, we would basically see:

A   A*
1    1
1    1
1    1

This means that the AA* pair is complete and in total is zero. The question is whether this is the trigger for the production of a new element in the alphabet. I think it is. Intuitively, what is described is the point at which a gesture or idea is thoroughly familiar to the point of being boring, and this requires something new. It is a way of describing the satiety of the alphabet.

Now what happens in communicative musicality? When two people are in musical conversation, there is in each person a different idea of what the alphabet might be. So person x might articulate an alphabet which is A,A*,B,B* and person y might articulate an alphabet which is A, A*, C, C*. The conversation articulates a combined alphabet: A, A*, B, B*, C, C*, AC, AC*. At what point does this alphabet become sated, where each element is 1?

In a conversation that "doesn't work", what will happen is that the communication breaks down. This means that the utterance of one person is not met by a corresponding utterance by the other. Equally, the other person might simply keep on repeating the same behaviour (the same alphabet) irrespective of the attempts of the other person to elicit a different response. Both these situations result in restrictions to the growth of the alphabet.

But when it does work, there is adaptation in the utterances of both parties, which eventually results in an expanded alphabet that is shared between the people.

"Sunday evening": My Roli Seaboard is fantastic...

A little over 10 years ago, I did my first post on this blog called "Sunday afternoon", with a video of me playing the piano.

So here's "Sunday Evening" 10 years later, but it's not a piano, but a Roli Seaboard. What an amazing musical instrument this is. I am growing into it more all the time. I have never had this experience with any digital technology before...

The Digital Computer and the Implicate Order

All living things "compute" (literally, "com-putare"... they "contemplate with"). The human-digital computer system (the whole system of humans and machines), is a system where the computations in people are constrained by the logic circuits in a machine. Since we are allergic to the uncertainty that is produced within our human "computing system", the constraints of the digital computer are welcomed - they provide ways of attenuating our uncertainty and giving us "answers".

But if we look at living things as computers, and seek to contemplate with them, we would, I think, look at the world in a very different way. It is not the attenuation of uncertainty that we should seek from our contemplation. It is instead guidance on how to act to maintain the coherence of life.

This "acting to maintain coherence" is essentially a process of understanding when and how to generate redundancies in our human system. The digital computer can be used to generate redundancies, but a lot of the time it is used to attenuate reality and to generate "information", which is the opposite of redundancy.

What I mean by coherence is, at its most basic level, a hologram, or a fractal. It is a fundamental process which encapsulates totality. When things fall apart, the fractal loses its internal coherence. When this happens, it is necessary to generate new redundancies, and sometimes new variety. But we need to know what to do and how to do it.

The fundamental question we should ask ourselves is how we might apprehend this hologram. It is close to what David Bohm called the "implicate order". Essentially it is unknowable, but some features can be perceived - particularly in the growth of living things, and especially music.

Music is a kind of computation. Music specifically shows the ways in which redundancies are required to be generated to maintain coherent life: a new accompaniment, a new melody, a modulation, are all ways in which music computes the nature of perception. Each new moment is not an accident. It is an expression of the whole, or an intervention to reveal the whole.

Digital computers are powerful enough to give a glimpse into the computations of nature. It is the latter computations which are our best guide for making collective decisions.

From Topology to Holograms

There's a missing link in my thinking. On the one hand I am seeing holistic approaches to organisation as in Beer's work (in fact most cybernetic approaches are holistic) in a topological way. On the other hand, I am concerned with the fractal encoding of nature in things like holograms, where time and space are enfolded (this comes from Bohm).

The topology side reveals forms like the Mobius strip, trefoil knot, hexaflexagon, etc, where re-entry (which is something I'd not properly understood until quite recently) is a feature of a whole form. The connection with eroticism which I mentioned in my last post is something that puts a new dimension on this - a connection to lived experience, and perhaps psychological dynamics such as the double-bind. More than anything, I find this a productive way of thinking about profound questions such as "what drives the bee towards the flower, or the sperm towards the egg?"

Topologies enfold time in a peculiar way. We have to pass over them to understand them. They are structured space (synchronic), but the time of diachronic exploration is implicit in them.

But a topology is not an encoding - it is a manifest space. However, a topology can be encoded as a hologram.

Whilst a geometric form like a trefoil knot plays with space, music plays with time. The way in which music's playing with time might be encoded is the critical issue. I think this works according to the same principle as an object's encoding of space. Actually, in the case of a hologram, space and time are implicated in both, because a hologram is formed through the interference patterns of light, which implicates frequency, and in turn, time.

Music's interference pattern involves the interactions of redundancies or constraints. It's not just music - it's any diachronic process which involves this... learning and conversation are exactly the same. But where is the connection between the hologram's encoding of space and music's encoding of time?

A holographic encoding combines not only all dimensions - time and space are two of them (but we should also consider mass and charge since these also participate in the interference in light) - but is also able to represent the relation between these different dimensions in different ways. Aesthetic experience relies on this multiplicity of decodings of a hologram: we can be moved in similar ways by Beethoven, Shakespeare and Picasso. They all express something fundamental about the universe.