The capability of AI is likely to lead us to bad decision-making. But it needn't. What is remarkable in the statistical amalgamation of training data (particularly computer code) is the capacity to represent old thinking, which was probably never properly understood, in new ways which can be more readily understood. Some of this old thinking is deeply cybernetic, which also challenges present analytical techniques.
One of those analytical techniques which straddles the cybernetic approach and present analytical approaches is Shannon information theory. Gordon Pask's take on entropy was quite different from Shannon. He saw entropy as an enacted property by a physical machine, not as a calculation of a number of bits of information. The difference between high and low entropy was the amount of work that needed to be done in order for a system to establish stability following a perturbation. This is modellable as Shannon information, but it has a number of advantages over the Shannon equations. Primarily it places the emphasis on the whole system relationship which is not measurable as such, but which is enacted in the system behaviour.
This is a much more biological way of approaching the whole issue of surprise. Surprise isn't a metric. It is an enacted systemic relationship.
How can our present AI help with this? Partly because it makes it much easier to make "maverick machines" - those devices through which Pask explored his understanding (and in the process lost a lot of others!). Most importantly, those devices are analogue, not digital. But of course, present digital AI is quite good at creating (on-demand) analogue simulators.
This really speeds up the process of getting up to speed with Pask's brain. I wonder if, in fact, very few of the other cyberneticians really understood him. Even his students became somehow dogmatic about his ideas rather than really thinking in the same way that he did. This was probably the fault of the pedagogical relationship between him and them (his airy Victorian engineer persona probably didn't help - and set a bad example to the succeeding generation). What if he'd been able to playfully create analogue computers on-demand to explain what he was talking about? (indeed this process was directly modelled in his conversation theory - the shared modelling environment at the bottom of his diagram). But there's more to it than merely the concepts. The analogue machine had to be coupled with a brain that operates with it.
There's something in these analogue machines that is missing in other cybernetic approaches. For example, Spencer-Brown misses something with the binary division of the mark, even if self-reference produces some interesting results. Shannon misses it by focusing on numbers for operationalisation. Maturana and Varela miss it because their "embodied cognition" (perhaps more Varela), while also embracing self-reference, isn't really embodied at all but metaphorical. Beer perhaps is the closest. For him, the organisation is the enacting device. But it is still difficult to envisage without simply falling back to Beer's demarcation of system 1, 2, 3 etc. and the march of his disciples.
With Pask we get the organic recursive system which must work on itself to establish homeostasis. The degree of complexity relates directly to the amount of work expended. This is why Pask's musicolour is particularly powerful. The organic system is us - and I think that is a clue as to how this enacted entropy might be operationalised. The key to an organisation's health lies in the physiological mechanisms of each of its workers. To understand those mechanisms we need to look to both cybernetics and biology.
Putting the details of that aside (pending a new paper!) that means that occupational health is deeply connected to organisational viability. Pask's enacted approach to entropy could be very powerful as an organisational tool focusing on the adaptation of each individual. Could this approach identify risks of pathological organisation or corruption? Maybe it could.
No comments:
Post a Comment