[about 1400 words – i.e. a longer rambling post. if I could compress it into a tweet, I would have. of course, nothing – especially around this topic – says you have to post it or even read it]
Discussions of free will – and its likelihood and basis – been going on for a while. Even before Palm – even before papyrus. What’s different – here and now – are things like deeply insightful physical and functional brain neuroanatomy, and electronic neural networks whose scale – based on outcomes – has reached/exceeded some level of parity with some less adventurous aspects of human thought.
First, some old physics. Aristotlean physics.
You may remember the concept – a 2x2 matrix categorically explaining matter as hot or cold, wet or dry. In one sense, the guy was right. There had to be some sort of simpler categorical rules underlying things. He also understood how to get more “likes” than Democritus – who, in turn, understood this better than Leucippus.
It was Leucippus who conceived of – below a certain size – indivisible atoms. Aristotle viewed atoms the way the inquisitors viewed elliptical orbits. Bad for the brand. But – ironically – it was the whole of Greek philosophy and protoscience that set the stage for much of what we know about our world. And how to go about knowing even more.
[Aristotle: “The whole is more than the sum of its parts]
Now – fast forward about 2200 years to the time of Kekulé.
Like Aristotle, a pre-eminent heuristic scientist of his day. But – unlike Aristotle – able to avail himself of all sorts of scientific insight, which he parlayed into some absolutely brilliant, somewhat empirical chemistry.
Chemistry before quantum mechanics was sort of like evolutionary biology before Crick. No one entirely sure what was really going on, but the patterns were useful and repeatable enough to make lots of money. The periodic table, like the structure of DNA, had been analyzed enough to see the discrete patterns precisely.
As Kekulé would retell his purported dream a quarter century later, the structure of benzene had eluded him. Till one night, when he dreamt of a snake – presumably made of carbon atoms – seizing its tail, and divined the alternating single/double bonds of the benzene ring.
[almost halfway there]
Reading this fascinating forum, and with a lifetime of debauched education and career in mercenary physics and EE, it struck me that you all could use a good “Kekulé 2.0” moment. So let me humbly offer a wormhole into a multiverse where August’s great-great-great-great-granddaughter [henceforth, G6] has just won the Breakthrough prize for her insight into neurocognition.
A double-major in neuroscience and organizational psychology at Smitanvard [in this multiverse, Harvard, MIT, and Stanford had merged decades earlier], G6 had long pondered what could be the x-ray crystallography of her day and discipline. Patch clamp more like a cloud chamber than crystallography. After successfully founding a company making handheld crowd-control devices for HR vice-presidents, and serving as its HR vice president for a time, she had a familial epiphany.
About twenty minutes into listening to an intern interviewee’s answer to figure out how many pennies would fill the pentagon [correct answer: the question is dull – only thing worth filling the pentagon with is hundred-dollar bills], she dozed momentarily – and saw it.
A seething and writhing human hierarchy – not of VPs and middle-managers and massive masses, but of BoD’s chaired by and comprised of chairpersons and members, most of whom were members or chairs of other boards. But not the conventional lateral structure of interlocking boards, or even the holding-company structure of Warren and Charlie – or their younger selves, Sergey and Larry.
It was an atomic BoD [not to be confused with the nuclear BoD’s of some PE cos and hedge-funds]. These BoD’s – like the twisty little passages in the Adventure game of a half-century ago – could be interconnected in any manner. Any individual could be on – or chair – any BoD.
The way any BoD could consider any input or action – like the brain would consider any input, or consequent cognitive/robotic response – would be to put something on a BoD’s agenda. The way a board would take any action – after conferring, during which could have all sorts of re-entrant and renormalized conferring with BoD’s to which it was linked – would be to take a vote.
See, as informed and as expert as you all and the greater neuroscientific community are – I’m deeply and respectfully serious – it looks like there’re several fundamental physical and information-theoretic constructs that may apply, whose quantitative scale and qualitative structure you haven’t yet quite reached or anticipated.
- Is the static structure – like DNA – fully tractable, or does it need some sort of electronic resonance to simply be. Not even talking about learning. Some sort of zero-day cellular automata rules. Like solitons at higher scale, and gliders and glider guns at lower scale.
- Are the primitives – like FSMs or registers – completely separable, or is their dissolving into neighbors fundamental for function and efficiency. If this sounds too abstract, an example from complex SoC design. Humans need to think of a chip as having some sort of rectilinear floorplan, with rectangular – or barely more complex – regions allocated to different parts of the design. Or different design teams. Yet – when the chip is physically placed and routed, these blocks may dissolve into one another at the edges, where the gates tend to be less utilized. But this dissolution is a production artifact. The co-mingled blocks have no circuit-level awareness of one another, and may not interact till up to the system-bus level of integration.
- Our 3D/haptic and 2D/image based consciousnesses are incredibly and contextually configurable. Seated in a train, one can fixate on the signs and people on the incoming platform, signs and people within the car, or an immersive videogame on a smartphone. Another train-related example is to go ride a long tunneled escalator, and prompt your mind to think the tunnel is level, and the people are all standing at a 30 degree angle.
- In physical and biological stuff, much of the action is at phase transitions – with the holy grails being complex but reversible ones. They are energy efficient, and – like PCR – nature, given enough time and venture funding, happens onto some of the more profound ones. Incidentally, you all exist at the edge of three “metaphorical” phase states:
• HW <> SW – with FPGA’s being like frost forming on a window-pane
• Data<>instructions – and per JH thoughts/comments on sparse use of a large address space, double-linked-list constructs to spoof a CAM in conventional memory might be of some interest – especially with some HW access optimization
• Startups <> mature cos – where the startups continually seek new connections, while mature companies focus on pruning old ones, as the essence of their day-day endeavors
- Just as in AI, if the precision of calculation outruns the fundamental outcome error band – even with lots more information – I’m more likely to act based on 1% of the information from each of three statistically separable sources than waiting on 80% of one, though dogmatists of any sect would likely flay me for having such view(s)
- Most blasphemous. No one would think of trying to do chemistry with hot/cold/wet/dry as the eigenvectors. Or math, where every number had one of two values: 0 or 1. So, the notion of things that are foundationally T or F – aside from restricting things to a small subset of symbolic logic – completely ignores the sort of fundamental role boundary conditions or cutoff frequencies play in more numerical computation. Is this why intelligence – artificial or otherwise – throws up its hands beyond a certain point and just goes with the crowd.
Could go on for another several – but this is likely already wearisome, if not downright annoying.
Also, it’s not the crux of the message. You are close – and here’s why I know. As I’d watched the progressive visual complexity and nuance of videogames and multimedia entertainment, it wasn’t just a matter of more triangles/second. A long-term framework for building things up from a sparse variable 3D mesh (aka a primitive cerebellum) had to include things like hidden areas, color, followed by texture, and luster. The most profound qualitative breakthrough of Nvidia – accompanying the quantitative breakthrough of massive GPU parallelism – is the enablement of massively parallel ray-tracing.
See, while ray-tracing is an sub-cognitive graph-connected augmentation of a complex 3D physical model – Nvidia’s further acceleration of ray-based image-generation using AI is beginning to border on things like what [I think] Alpha Go does. Or that may be a distinction without a difference.
You are close. Best wishes for further – and ultimate, if there is such a thing – success. Godspeed.