‘A Thousand Brains’

George Rebane

Jeff Hawkins, cofounder of Numenta, is the author of A Thousand Brains – A new theory of intelligence (2021).  To many, me included, Hawkins has been a national treasure since his first book On Intelligence (2004).  After inventing the Palm Pilot and becoming wealthy, Hawkins has devoted his life to brain research as it impacts our understanding of intelligence, consciousness, and our ability to create machines with these attributes.  In this enterprise he and his colleagues have devised a new understanding of how our brain’s architecture, specifically the neocortex, is able to perform all the wondrous functions which we all take for granted.

My son-in-law who is a senior AI researcher at Microsoft Research, sent me the book as part of our ongoing conversation that includes the art of the possible in intelligence studies and, of course, the Singularity.  For me Hawkins’ book was an immediate page-turner that jumped to the head of my list of books-in-progress.  The 257-page book is divided into three parts: A New Understanding of the Brain, Machine Intelligence, and Human Intelligence.

In the first part Hawkins builds on the work of famed neuroscientist Vernon Mountcastle, who revealed the structure of the neocortex (the new brain) as consisting of hundreds of thousands of cortical columns or cores, each of them with pretty much the same make-up of interconnected neurons which also have very select and important dendritic connections to other cores, often located on the other side of the brain.  The important concept here is that the cortical column is itself a tiny brain that is able to learn and retain ‘frames’ of perceived reality which it then assembles with other cores containing complementary frames to result in a voting process from which our experienced reality emerges.

These common components that make up the neocortex are ubiquitous in their functioning.  When those that get inputs from our visual subsystem, these become visual processors and generate our visual world; when connected to our auditory subsystem, they do auditory processing and create our world of sounds, and so on.  Hawkins’ exposition of how these cortical cores connect to our sensors, manipulanda (e.g. hands, feet), and parts of our ‘old brain’ is an absorbing read.  On every page, the reader’s brain is coming up with all sorts of alternative AI implementations that have now been introduced into his art of the possible.

And for me the stuff I learned in the first part makes the second part on machine intelligence very exciting.  The people at Numenta and other similar AI companies are all experimenting with synthesizing software and hardware structures that can mimic the functions of the cortical core, and then devising ingenious architectures that join thousands of these cores into new kinds of learning machines.  The commonly recognized holy grail among these enterprises is AGI – artificial generalized intelligence, of which today not even the most advanced forms of deep learning artificial neural nets are capable.  All of the ballyhooed AIs today are still one-tune bands, no matter how sophisticated tunes they are able to recognize and play.  None can efficiently construct their surrounding reality and transfer that knowledge (i.e. generalize) to operate in different environments or situations like any year-old human infant, or, for that matter, as can a lot of critters.

All of today’s advanced AI is focused on developing machines that will learn to be intelligent and (possibly) conscious.  No one believes anymore that intelligence will be programmed into some future super-computer by clever human computer scientists.  Just looking at the structure of a part of a cortical core under a powerful microscope disabuses such hubristic dreams.  As Hawkins describes one little piece of the ubiquitous (looks similar everywhere) neocortex –

Under one square millimeter of neocortex (about 2.5 cubic millimeters), there are roughly one hundred thousand neurons, five hundred million connections between neurons (called synapses) and several kilometers of axons and dendrites.  Imagine laying out several kilometers of wire along a road and then trying to squish it into two cubic millimeters, roughly the size of a grain of rice.  There are dozens of different type of neurons under each square millimeter.  Each type of neuron makes prototypical connections to other types of neurons.  Scientists often describe regions of the neocortex as performing a simple function, such as detecting features.  However, it only takes a handful of neurons to detect features.  The precise and extremely complex neural circuits seen everywhere in the neocortex tell us that every region is doing something far more complex than feature detection.

Hopefully, you begin to see the difficulty (impossibility?) of replicating such complex computing structures in parallel executing software.  But in this part of Hawkins’ dissertation on machine intelligence, the man does lapse into dreams of writing code when the discussion turns to possible dystopias in our pre-Singularity future – meaning that machines can become our malefactors even before reaching peer intelligence with us.  Hawkins, somewhat glibly, comforts us by assuring that we can avoid creating such reprobate robots through SMOP – simple matter of (timely) programming.  In short, writing control code that limits the scope of learning so as to prevent the machine getting bad ideas about how it should carry out its duties of operating our factories and controlling our environments.  There I part company with Hawkins, but as always, I stand ready to be corrected.

It is the third part on the future of human intelligence, and how that may relate to extra-terrestrial civilizations, that things really get onto a philosophical plane where we all have a chance to be ‘experts’.  Hawkins reminds us that we live in a simulation, a simulation manufactured by the active neurons of our brain.  We have done many experiments which demonstrate that the world we experience is an incomplete approximation of what is actually out there.  Yet, so far, that has been sufficient for us to develop, thrive, and survive to arrive at a crossroads for our species and planet.  (Hawkins is a true believer in the coming cataclysm of manmade climate change.)

But before considering ETs and all that, Hawkins first delves into the whole question of machine-augmented human immortality.  There we encounter the now classical paradigms of uploading our brains, and merging with the machine.  Uploading can come in two major flavors – destructive and duplicative.  And the question of what is the ‘I’ immediately rears its head, and is considered by the author who does recognize that in the duplicative version, there will emerge two identical beings – the original bio-I and the new machine-I.  However, both beings will begin growing apart immediately because they will no longer share the same ‘world line’ going forward.

A lot of authors at this point have the bio-I check out that the machine-I is indeed a true copy in every possible way (perhaps even including in its physical embodiment).  And then without further ado the bio-I allows itself to be destroyed, sanguine in the belief that he has achieved immortality in the machine-I.  But is that really the case?  There were at one time two sentient and sapient beings, and now there will be one.  Where did the I-ness of the dear departed go?  The machine version already had its ‘I’, and there is no concept of two Is co-habiting a single ‘body’, being, or brain, especially since no one has described how the bio-I will get into the machine-I.  And then the question that immediately begs is ‘what is that bio-I that is so sensitive to its insipient mortality and so desirous of going on with life?’

Hawkins acknowledges that desire in the case of the two Is in which the bio-I still does not want to have his life ended; but then Hawkins drops the thread and goes on to discuss other cases.  The other uploading alternative is where the bio-I flips a switch, everything goes blank for a ‘second’ during which the last memories are uploaded, and then wakes up in the machine-I, exclaiming ‘Hallelujah, I’m alive and now immortal!’  But is that replicate I the original bio-I that died when the switch was flipped?

In any case, I am of the school – perhaps a school of one since I have yet to run into a classmate – that is not satisfied with such a short-shrift treatment of the I-ness that we all so intimately experience to the end of our days.  Which brings us to the ultimate self-enquiry question – ‘What am I?’, and somehow the explanation of a big bunch of well-oiled and smoothly functioning neurons doesn’t quite provide a satisfying answer.

So finally, we come to ‘Merging Your Brain with a Computer’.  A la Hawkins, “An alternative to uploading your brain is to merge it with a computer.  In this scenario, electrodes are placed in you brain that are then connected to a computer.  Now your brain can directly receive information form the computer, and the computer can directly receive information from your brain.”  In this scenario, the computer is, of course, connected to the internet or a similar future repository of all knowledge with unimaginable numbers and kinds of sensory inputs, and yet to be determined means of controlling mass and energy.  And we can even conceive of an exalted form of such a future being who can embrace the additional control/disposition of space and time.  But please notice, dear reader, that nowhere in this scenario have I posited the exit or termination of the bio-I.  The ‘computer’ is still a mere functional extension of the original I.

Hawkins believes that “the ‘merging of your brain with a computer’ proposal will likely be achieved for limited purposes, but not to the point of fully uniting brain and machine.  And a brain merged with a computer still has a biological brain and body that will decay and die.”

In my humble synthesis of the future, I believe that the bio-I can indeed merge with the machine through a process that in math and science is termed ‘analytical continuation’.  In such a perceivably continuous process the functions of the bio-I’s brain and being are incrementally transferred to the machine while, at no time, are there two Is and during which the bio-I experiences no discontinuity in the perceived reality of his world or his consciousness.  When such an analytical continuation process is completed, then there will still exist the subjectively perceived original I now embodied in a new machine form.  The old bio-body is then truly an empty shell (cadaver) that can be discarded with no loss of sentience or sapience.

If such a transfer can some day be effected, then will the answer to 'What am I?' remain unchanged?

Comments

[THOMAS O'TOOLE]

Dr. Rebane, Let me see if I have this correct; I can think therefore I am, making a machine envious if it could think! Close?

[The Estonian Fox]

Mr. Descartes would have had it correct even if he were dyslexic - 'Sum, ergo cogito' stands up just as well as 'Cogito, ergo sum'.

I'd be overjoyed if our politicians showed a teeny bit of awareness of either saying.

[Bob Hobert]

Amazing stuff. Most of us do not deserve immortality. Who decides?

[George Rebane]

BobH 756am - Probably the same as always. When the time comes, then Golden Rule #2 (Them that's got the gold makes the rules) kicks in, and the decision follows.

ThomasO 1105pm - When we get to the stage where machines can feel 'envy', then the game may well be over for mankind.