The AI side of the equation

I started my career studying Artificial Intelligence at MIT.  Back then the researchers thought that we would have computers that were smarter than humans in short order.  What I discovered after observing what the AI people had done was hardly anything close to “intelligence.”   Marvin Minsky called the bluff and wrote an article back in the day basically spelling out in a somewhat comical way that using the variable “learning” in a computer program didn’t mean the program learned anything.

What we have done with AI since then is to build smarter and smarter algorithms and if there is a learning variable in those programs it doesn’t mean those programs are learning anything either.    Some of these algorithms running against massive data stores like the contents of the internet and with virtually unlimited processing power can appear to produce answers to questions that look like they “get” the question we asked and give us an answer that is better than a human being can.   I am not saying Watson has no value but as a “learning” “sentient” intelligence, NO!

The problem I believe is that computer algorithms are designed to operate on predefined abstractions.   No matter how smart the programmers are that create these algorithms and abstractions they are limited by the abstractions the programmers put in.  For instance, if they teach it about mathematics it can be programmed to be better at solving differential equations than a human being.  However, it really has no idea what a differential equation is.  It will NEVER discover Set Theory or come up with a way for you to go on your vacation for less than $1,000.   Hard AI advocates have said that if we keep programming expertise into the algorithms eventually it will cross the line into something we call “intelligence.”   No.  I emphatically say NO.  The problem is that we haven’t learned how to represent real “learning.”

It is surprising what we have NOT accomplished.  Given $10s of billions in resources and unlimited computer power an algorithm can answer memory based questions like a brain made from 15 cents of common chemicals with no external memory storage other than what is sitting between the two ears.  The brain runs on a few calories of energy an hour, the computers on billions of calories/hour. Seriously after 35 years of research, the work of many diligent scientists and many other companies and research facilities worldwide the computer still has nothing approaching the human brain in real “intelligence.”  In my opinion our success with traditional computer science is F.

I want to make clear I am talking about the goal of making AI machines which have the ability to learn new concepts, to act creatively to solve problems, i.e. having the ability to create new abstractions from experience and to leverage those abstractions in making higher and higher cognitive reasoning.  The goal of much of AI today is to create programs that can perform in some specialty, i.e. vision or math or answering questions from known information then the field is making progress in producing reasonable and useful tools.   By constraining our work to less than “abstracting algorithms” we can avoid what Stephen Hawkings and Elon Musk and what even myself worried about many years ago which is computers that threaten humanity with replacing us.   In essence the path AI is on today is to build useful machines that leverage mans intelligence and skill.  Indeed these machines may replace lots of human labor but always at our command and control.

A further concern I want to emphasize, which should be obvious, is that the fact that computers have not so far been able to do “abstract learning” doesn’t mean that programs can’t behave in extraordinarily dangerous ways.   For instance, thinking that if we put computers in charge of our nuclear weapons we have no problems if the computers are programmed not to exhibit “abstract learning” is not true.  Such programs controlling nuclear weapons or cars or anything dangerous can be programmed with mistakes or with wrong decision logic which causes them to make fatal mistakes potentially hugely fatal mistakes so we shouldn’t also think that just because we haven’t created our nemesis that we are safe to use computers and assume their answers are right all the time.

Please check my comment at the bottom as the DeepMind acquisition from Google seems to be attempting to solve the real learning problem.   I am tentatively giving the AI researchers a D based on DeepMinds success.

Grade for AI Researchers : D

Progress from Biology side

Other things have bothered me a lot from the biology point of view.  When we look at the human brain researchers have NOT discovered where the hard drive for the brain is.  Where exactly and how exactly are memories stored in these neurons?  They can tell you parts of the brain that seem involved, some studies have identified it must be in dendrites but where, how?  No idea.  They’ve discovered different kinds of memory but the memory that I pull up smells, people’s faces, formulas, algorithms, no I don’t believe they can tell you how the brain actually stores the information or how it retrieves or matches against it.   How does the brain do such things as “recognize” patterns?   How does it create “abstractions?”  How does a brain have an “AHA” moment?  What is consciousness?  What are those pesky EEG patterns about anyways?  Why does meditation work?  I could go on but we have only nibbled at the edges of the brain operation never really having any sort of answer for things that should have been cracked by now.   We see the neurons firing at the edge going in, we see neurons firing in the brain, we see neurons sending signals to body parts.  What happens in-between is still 100% mystery.

Besides the hard drive problem the other thing that I find fascinating is we don’t know how the pattern recognition algorithms are being run.  This is hard to believe because this is all that a brain does to some extent is match patterns.  It’s sole operation is to look at patterns and continually match those against prior seen patterns.  99% of the computation the brain is performing should be patently obvious because this pattern matching should be taking a lot of obvious work.  In a computer you would be burning billions of watts of energy pattern matching historical databases constantly.  Disk heads would be flying about in a constant state of activity.  We don’t see that.

Instead it seems like the pattern recognition and memory storage are hidden from us and happen as a side effect of brain operation.   We only seem to see things at the edge, signals in and out, some parts of the brain flash that are being used but how?  No answer.  Not even a start.   Again, in the sense of how far have we come in understanding how intelligence works I have to give brain researchers a low grade.

Progress from the Biology side: D

Why so little progress?

Attacking the wrong problem:  Chicken problem

The big problem with all these attempts at AI from the beginning is what I’ve called the “chicken” problem.  I said back in school many years ago that our goal should not be to make a computer as smart as a human.  That was hopeless considering where we were.  It was better to set our sights on a possibly achievable goal.  Show me a computer program that can be as smart as a chicken.   Many people may think I am jesting and chickens are not that intelligent but in fact the opposite is true.   A chicken is an autonomous living thing which has to make decisions about life and death situations.  It has to recognize threats, recognize food, handle repair, predict the future based on past experiences but be able to correct its mental model if it is wrong to survive.  Chickens have survived for a million years whereas if we didn’t keep fixing them computers would be extinct in 3 years.    We have a tendency to always underestimate our partners on this planet and overestimate our knowledge but if you actually thought of trying to mimic the behavior of many of these “lower” creatures in even limited ways we would be writing an incredibly complex and long program and I don’t believe we even know how to start.   So, one reason we’ve made so little progress IMO is that we frequently have been attacking the problem wrongly by trying to make a computer appear to do hard things like solve differential equations we have not tackled the real hard problem of abstraction making and learning that even a chicken can do.

Not looking in the right place:  Pre-built structures and processing

The basic problem is “abstractions” and our inability to come up with a way to construct “abstractions” in a systematic way.   I have included a paper on some recent vision AI research.  Frequently in descriptions of the brain from people who have studied the brain they have discovered parts of the brain they think do this kind of abstraction or do that abstraction as if the algorithms like a computer program have been programmed into that structure of the neurons there.  They describe the brain structure as if DNA programmed this abstraction into the cortex.    I doubt it for several reasons.  First, the amount of specificity they are talking about frequently if added up across all the things they think are pre-programmed into the brain would require a massive program far beyond what the DNA could encode.  It is possible some of what they say is true and some brain regions are customized or dedicated to some kind of work but I believe the number of these special areas must be very limited because of what i call the combinatorial problem.  The fact that some parts of the brain appear to process input signals in some abstract way is pure coincidence.   I believe the brain adapts to the input and many of these regions form naturally as a result of the inputs not preceding the inputs.  Therefore another dead end in my opinion is trying to study the brain to learn the abstractions it makes or really to in any way presuppose what are the right set of abstractions.   After all the whole point of the learning algorithm is to learn abstractions so why short circuit the process.  Start at the bottom and see if we can get a program to do the basics.

Quantum Mechanics in evolution is pervasive

The number of examples of nature using quantum mechanics for basic functions is large and now goes back possibly a billion years or more in the tool chest of evolution.  Check the articles at the end for the latest information on where we have discovered quantum effects in animal and cell physiology.

Plants:

There is a lot of evidence emerging over the last 10 years that nature has leveraged quantum effects all along possibly from near the very beginning of life.  One of the most interesting is the discovery that photosynthesis operates using quantum tunneling.  When a photon hits a plant leaf a molecule attached to the chlorophyll molecules called the clorophore absorbs the photon.  The clorophore molecules are next to each other and they form a quantum coherent state that allows the electron that emerges from the photon hit to be ferried with zero energy cost essentially to the place it is needed to complete the breaking apart of CO2 molecules into Oxygen and Carbon for the plants and our benefit.  This effect is similar to superconducting.

Consider this:  Plants depend on this efficiency which could be 1,000 or 1,000,000 times more efficient than classical mechanics would allow to grow to the point they have.  If they didn’t use this they may never have produced a food supply sustainable for animals, or even been possible for themselves.   So, this is a rather astonishing discovery.  Life itself may depend on quantum effects and it is likely if it uses quantum mechanics for this that it uses it for other uses.

Birds

Recent studies have performed tests on various types of birds that have geolocating capability.  When the researchers turn on magnetic fields too small to move a single iron molecule the birds become unable to figure out anything.  When the field is turned off the birds return to being able to navigate.  Such a small magnetic field could only be operating through some kind of quantum effect.  We don’t know exactly how yet but clearly conventional physics is off the table.

Senses – Eyes, Smell, Hearing, Touch

Smell is interesting because this sense in particular is one we have been particularly poor at trying to create our analog.  We have no good idea how to do the pattern matching the noses of animals and humans do routinely.   Studies have pointed to quantum effects going on in human and other animal smell.   This is not surprising at all.      There is good evidence that all our sensory input is being transduced by quantum effects for us and for most animals.  It turns out our eyes use the same mechanism as plants for photosynthesis.  All of our visual receptors contain a similar molecule which allows them to ferry the single photon of energy to the nerves and create a magnified macro nerve signal from a single photon.  So, it seems all or most of senses of most animals are helped by evolutions reuse of quantum computers.

Immune system

I have not read this yet but I would lay big bucks that the immune system leverages quantum effects to recognize foreign viruses.   There is evidence that the same microtubule type structures seen in post-synaptic membranes that I will talk about in the next section are seen in T-cells and recent studies have implicated these in quantum effects.  How else to explain the unbelievable skill of our immune system?

The Brain

Would it be surprising that if nature has been using quantum technology for a billion years in our senses it would be leveraging quantum effects in our brains?  I think it would be ridiculous if it didn’t.  Many things about the brains function happens to correspond with some of the things quantum computers are good at, i.e. pattern recognition, randomicity, optimization.   To me it would be ridiculous to assume at this point that the brain doesn’t use quantum effects given we now have solid evidence evolution uses quantum mechanics.

In January of this year scientists discovered quantum effects in microtubules in dendrites.   Previous studies have shown very robustly that memory and cognition must take place within dendrites.   One study was very conclusive that if dendrites were not involved memory did not take place and another showed that dendrites were involved in remembering details about specific incidents so that the theory that memory is diffusely stored in multiple neurons is not strictly true.  Also other studies have shown that brains were unable to function cognitively with dendrites suppressed.

Even more recently scientists discovered how neuronic activity can cause CaMkII (a 6 legged dual sided structure) can activate (program /phosphorylate) 6 bit tubulin molecules on microtubules in post-synaptic dendrite junctions.   They showed additionally that this could lead to neuron firing.  Thus it seems that we finally have a mechanism that neuron sequences can be stored on microtubule tubulin molecules and that these “memories” can then trigger neuron firing.  Finally the hard drive may have been discovered if not the pattern matching machinery.

Penrose and Hameroff (*) propose a specific location for quantum activity to happen (post-synaptic junctions in dendrites and soma.)  The recent discoveries seem to support their theory.  There is still some doubt about this but P&H don’t say they have all the answers and admit there’s is a work in progress as far as precisely how it works.    The P&H theory is called “Orch OR” for orechestracted objective quantum reduction.   You can look up on the internet lots of stuff about this theory.

In P&H’s theory every 20ms or so the brain has a decoherence event.  This happens in the P&H version due to quantum gravity, however, it is not necessary to invoke that to cause decoherence.  There could be any number of physical processes that could cause this.  What’s important is that they have postulated for once a reasonable model of how the brain could actually work using quantum computing.  Given that we have seen nature doing high temperature quantum computing and we ourselves have now built quantum high temperature material in Correlated Oxide it seems eminently plausible the human brain is doing some quantum computing.

DNA and Evolution itself

In the book Physics in Mind(1) Werner describes the likely way that the eyes and other senses evolved.  He also describes the specifics of the quantum effects involved in many of the things described above and puts forward an eminently more likely theory for describing how DNA evolves and changes to create things that have been puzzles.

I have struggled with evolution for a couple reasons.  The sheer complexity of life has never seemed sufficiently explained combinatorially by DNA (what I call the combinatorial problem).  When the human genome project said there were only 30,000 genes it was apparent to me something was missing.  I said repeatedly that it was impossible to construct and operate a human body with 30,000 genes.  Some vast amount of “information” was missing.  We discovered several years later that control DNA patterns were located in the 3 billion junk DNA base pairs. We call this epigenetic information.  It represents an entire 2nd coding scheme and understanding it has been a major challenge of genetics in the last few years.    I don’t believe this is the entire answer either.   Werner gives some ideas to help solve this puzzle using Quantum demons.  I believe this book is worth reading.

I am a computer scientist.  I know how complicated it is to write the simplest program to operate things like robots or planes.   There are frequently millions of lines of code involved.  The code is extremely fragile.  Numerous bugs exist that cause problems.  As a result we go through extensive debugging processes.  These robots and planes do not have a small fraction of the code needed by a human body.   The human body has 100s of thousands of important chemicals, molecules and each of these needs to be regulated and put in action when needed at the right time.   When damage happens to the body it must put a repair process in place that pretty precisely fixes the problem employing many materials.  The body is beset by invaders who want to do it harm all the time.  These invaders must be detected, recognized and appropriate action taken.   Constant breakdown of the body occurs and must be repaired.   While the size of the DNA is impressive at 3 billion nucleotides it is impossible for me to understand how it is possible that in there is the coding to operate a human body.

When I mention this to biologists they seem unimpressed.  I don’t understand that reaction.   This reminds me of the fact that as early as the 1930s scientists were aware that galaxies were not possible.  The stars in the galaxies were flying 3 times faster than Newton and Einstein would allow.  Some bright cosmologists said back then that we would discover the universe was mostly made of invisible matter.  The issue was tabled for decades until recently.  When I went to school nobody mentioned this problem that the galaxies were disproving Newtons gravity law.  It was shuffled under the table.

The problem is really much worse than I described above.  If you look into each of these processes the human body does routinely you will see it is more complicated than you initially assumed.  For instance, looking at the ER (endoplasmic reticulum) in almost every cell of the body involves the construction of thousands of molecules and proteins, lipids, hormones.  The size of the ER and what it produces can vary dramatically during short time periods depending on demands and the chemicals needs, the stress on the body, the location and type of cell involved.   There are dozens of types of cell motor proteins for ferrying organelles and various proteins or other chemicals inside the cell as well as in and out of the nucleus and to other cells through microtubules and other gateways.   Some of the motor proteins can ferry individual molecules and some can ferry whole organelles like mitochondria.  Forget the human body, managing a single cell itself is like programming a city.  There are hundreds of cell types in the human body.

The complexity is mind boggling and imagining a program that could run a human body would require the programming output of all humanity for ages assuming we even knew how to start.   I get the idea that when I bring up issues such as these that the community is disregarding the problem.  There is no way that 3 billion nucleotides and 30,000 genes build, regulate, repair a human body.  I am not saying this based on some religious argument but based on solid foundation of understanding that the complexity of this is just too much.

I am certain the more we look into this we will find that the regulation and operation of DNA is controlled by pattern matching quantum gear in the cell nucleus somehow.   I would also bet that aspects of DNA replication and repair involve quantum effects.

Quantum computers

Quantum mechanics postulates that nature performs operations that figures the least energy solution to problems that are extremely difficult like the quantum tunneling we talked about for photosynthesis earlier.  It is easy to construct problems for a quantum system to solve in one step that would require millions of steps in a conventional computer.   Some have said that the quantum reality is leveraging a million universes to solve the problem.   Maybe so.  However, it does it the quantum underlying reality allows nature to perform tricks, tricks that turn out to be hard but also handy tricks if you are trying to build small living things that have to be very sensitive and learning.   How could the world be leveraging a million universes to solve problems?

Is quantum mechanics wrong?  Are we only deluding ourselves in thinking nature can possibly do such impressive calculations?  No.  It seems this is really how our world works.   One proof is we have built a quantum computer and it works.   Several years ago a company in Vancouver BC created the first quantum computer called the D-Wave.  This computer had a new version this year which has 1152 qubits.    While 1152 qubits is small number it still represents an enormous leap in computing power.   A 1152 qubits computer can store 2 to the 1152 number of 1152 bit patterns in quantum superposition and do pattern matching against these patterns in real time.   The current D-Wave computer costs 10 million dollars, requires a supercooled environment to run but is able to do feats compared to traditional computers.

This fact the D-wave works essentially proves to me that quantum mechanics is correct at least as far as that the universe is not deluding us that it can do these things we theorize.  Somehow nature is able to do things beyond our ability to understand.     The D-Wave is being used by google for accelerating pattern recognition.  I find myself worried about that but that’s a distraction.  The real point of this article is to explain how I think the human brain operates in some ways like a quantum computer from the point of view of common sense.

First off we are making progress in using quantum technology not only in computers but for a number of products.  One example I thought was exciting recently is taking our basic silicon semiconductor technology and replacing it with a material that like the photon tunneling effect that nature uses in plants.    In an article referenced below scientists are able to reduce the resistance by a factor of 10 to the 7 in a material called Correlated Oxide.  In their first experiment they were able to match the best semiconductor technology on the market in performance.   The Correlated Oxide has no need to operate at reduced temperatures at all.  In fact it works at a hundred degrees above room temperature perfectly well as at room temperature yet it demonstrates quantum effects.   So, it seems we have now seen at least one case where we can manufacture quantum mechanics operating devices that operate at room temperature like nature apparently discovered a billion years ago.

Quantum computers are programmed very differently than conventional computers.  A quantum computer has a number of basic operations that are not at all like traditional operators.  For instance, the “Fourier Operator”, the “Hadamard operation”, Grover’s Algorithm and Shor’s Algorithm.   These operations allow a quantum computer to solve difficult problems for instance that require searching for optimum solutions among a large number of possible answers.  The Hadamard operation is amazing because it allows us to build the first truly random number generator ever.

A quantum computer can store a vast array of data in superposition meaning that they all exist at the same time as probabilities. This is the nature of our world.  Get used to it.  🙂  In nature this is called the fuzzy probabilistic wave nature of things.  The “memory is stored as competing possibilities that all exist in superposition.  When the quantum system “decoheres” the result is the most probable state.   The trick in quantum computers is to turn problems into these kinds of things.  Turn problems into searching for an answer among a large number of options that is best according to some criteria into something like an electron traversing a series of barriers on the way to splitting a CO2 molecule.  Quantum mechanics stipulates the particles will always choose themost probable path which is the one that uses the least energy.  So, for example,  by turning our pattern matching problem into a quantum tunneling problem we solve the pattern match when nature decides how a particle will go from here to there.   Pretty nifty.

What is amazing is that in a  quantum computer we are actually making a particle to go from here to there essentially running a physics experiment for each quantum computer step.  In computers we think of the real world as “virtualized.”  The computer is performing calculations but there is no real world.   In a quantum computer we are running physics experiments every time we do an operation looking at what nature has figured out.  We don’t know how it does it so all we can do is watch and learn from what it does.  That is fascinating to me and does beg the question how is it possible it does it but that is a question for another time.  Sweep that one under the rug.

I want to put a little skepticism here.  Not all problems are tractable with quantum computers yet.  The D-Wave is a quantum computer which has limitations.  Scientists are working on more “complete” quantum computer that can solve more general problems theoretically.   My post is not to say D-Wave is the answer but it is evidence that quantum effects are real, that quantum computing works and that the brain could be leveraging these effects.

The TIBCO chair which I helped fund at MIT is held by Scott Aaronson has a blog and articles on this topic of what a quantum computer can do and can’t do.  I have referenced them below.   There are also some interesting sites on the types of problems that quantum computers can do.

If the brain uses quantum computing it probably does this on a scale literally billions of times greater than the D-wave and therefore the D-wave only hints at what the brain could be doing.

How does a Quantum Brain work?  What evidence can we see?

In my conception of how intelligence could possibly come about it requires a learning period when the brain has to process many inputs from the real world and produce enough abstractions to make sense of all the data coming in.  This process of producing abstractions is far from trivial because the input data is immense and the ways it could be generalized are virtually unlimited.  Assuming this model is close to what the brain has to do, how could the brain do this?

Imagine that all of the abstractions are in superposition in a brain quantum computer.  What we looking for from the quantum brain computer is the best matching abstraction that fits the input data.   This is something a quantum computer may be able to do.    When the decoherence happens the result is the abstraction that matches, the “AHA” moment.  The particle traverses the maze and gives us the answer to what is the best abstraction.

There is new evidence to support the microtubules as being part of this process(*).  A study showed that cutting microtubules between brain neurons caused people to lose memories.  Also, research has shown that anesthetics that remove our consciousness operate by stopping the functioning of microtubules in the brain.  Stopping the microtubules stops consciousness.

If the information is encoded on microtubules which seems more and more experimentally verified then what mechanism could possibly do the pattern recognition against these memories on these phospholated microtubles?  If these microtubules were being searched by some macro-non-quantum process we would see evidence.   What machine or process or chemical system could do this?  It seems we would have discovered some very active machine running around doing that already.  It’s a hard problem, there would be a lot of these searching robots.   If memories are stored this way then how could it be done if not by quantum mechanics?

I have an article on the latest “software” attempts at simulating the human brain.  This technology is called CNN (convolutional neural networks) and is the best AI we have so far.  Variations on this technique have produced impressive results.  Please read this series of articles on CNN if you are interested in this topic.  Here.

If you read my series on CNN the last article discusses some of the problems CNN have not tried to solve that a brain needs to solve.   One of those problems is  associating memories with abstractions.   The brain when it finds a pattern match also finds the memories that go with that match.  CNN have no way of currently associating pattern matching with memories that could be called up.  The abstraction is general.  It’s like you saw the word “tart” but no memories of pretty things immediately pops into mind.   If you remember a smell it probably brings memories of someone or some incident but a CNN doesn’t have that ability currently.  These problems that I describe for CNN are hard problems and unsolved.  However the brain neurons work I don’t believe CNN accurately mimic them.   I have no doubt we have found part of the solution with CNN but we are far from understanding how to turn that into something more than a better face recognition algorithm for instance.

The quantum explanation solves two really hard things that I don’t think ever had a reasonable explanation for before.  The hard drive are these tubulin molecules and the pattern recognition is hidden because it happens in a quantum fog that can’t be observed by us until decoherence in which case we have the answer and neurons fire.   In the quantum world things happen in secret it seems.  During the period things are in coherence we cannot observe what is happening by definition.  When we try to observe we break the quantum fog so we wouldn’t see the pattern matching machinery, the AHA machinery because it is happening in the in-between moments we can’t observe.

This is an interesting fact that we don’t perceive the quantum fog because most of the world is in a quantum fog almost all the time.  Our “consciousness” is only able to perceive the moments when we observe things which are moments NOT in quantum fog.  If this is interesting I have a set of blogs on what our underlying reality is according to Roger Penrose, probably the smartest man ever to have lived.

The P&H theory also explains other things.  For instance, nobody has explained what brain waves are.  Well, it seems that the decoherence events are the brain waves.  The cadence that P&H calculate matches roughly what we see in brain waves amazingly.  Also, meditation has the effect of slowing brain waves.  This means extending the time between decoherence events.  In quantum computers this means the quantum computer can have more stuff in quantum coherent state and find better answers.  Therefore this theory explains why meditation works and why increasing brain wavelength would increase intelligence.  Making the brain less cacophony means that more parts of the brain are potentially in coherence meaning better reasoning.

Imagine that what is encoded in microtubules and tubulin is not just memories but also abstractions that are the result of pattern matches from before.  Each successful pattern match from before results in an abstraction being built.  The matching of additional patterns would be used to increase the probability of that pattern but there is a problem.

The way to construct abstractions is infinite practically.   The weightings on these abstractions are hard to stabilize.  Sheer luck may cause some abstractions to appear to be good for a while.  A computer algorithm will get locked into local minima and assume it has a good abstraction too soon.  What is needed is to continuously look at a large large set of abstractions and weights and pick the weightings that work best.  This is the optimization problem multiplied by a billion.   I want the quantum computer called the brain to look at all the possible weightings for abstractions and pick the best weights and abstractions considering all the past experiences.

Quantum mechanics does 2 things beautifully.  It produces the answer to very complex optimization problem but it is also random.  This random component is critical.

Quantum Mechanics keeps mixing up the stew with true randomness

When I’m trying to solve a problem I frequently will come up with a solution that is not right.  If a computer is used it will frequently repeat the same solution over and over.   I need the system to occasionally pick other answers because otherwise I may get stuck and never learn.

However, if I just pick a random answer every time it will take forever and again very frequently I will get the wrong answer. What I need is something that picks the best answer I have usually but occasionally lets the system think outside the box and pick a different answer, but a different answer that has some probability of being right, not absolutely random or serially random.  I need a special ordered randomness that is weighted by past experience.  Guess what.  That is precisely what quantum mechanics does.

This kind of problem solving would be hard for a single celled creature to do.  Yet nature probably needed randomness injected into life to expose the problems and explore the answers.  Quantum mechanics is undoubtedly essential part for how life started, evolved and does things.   I don’t believe we will find answers to many of life’s deeper mysteries like the brain, consciousness and how DNA really works without ultimately understanding how these systems utilize quantum mechanics.

This is circumstantial evidence of quantum mechanics being involved.  It couldn’t be used in a court of law to convict anyone and no physicist is going to fall over and say problem solved.  I understand but the problem with trying to explain the brain operation any other way than utilizing quantum mechanics is explaining the lack of any “machinery” to do any of the things I have described above.

Our lack of progress and lack of anything to point to as answers to these basic important functions seems to put the weight of evidence to a quantum approach.

Another important thing in my mind is that we not only see humans doing “learning.”   Cats and dogs see the world, understand the world enough to operate, to prevent themselves from getting killed, to find food.  Not only cats and dogs but insects, even microbes themselves sometimes show random behavior and learned behavior.   We know single celled creatures by definition don’t have a nervous system of dendrites and interconnected nerves.  So, how do small creatures with only a few brain cells or even single cells have some learning ability?  Something is going on here beyond what we have discovered.   Yes, it may all be explainable by macro classical chemistry and biology but I am not sure why we would assume that is the case.

Some point out that the first layers of sight processing occur not in the brain but in the eyes themselves.  The neurons in the eyes learn to recognize certain invariances and certain patterns before it gets sent to the brain cortex.  There is no problem with this.  It makes sense to perform the first few levels of abstraction out in the eye neurons.  They are neurons and if they can be quantum computers too then they could compute certain first order abstractions so that the brain deals with higher order abstractions.   This does not contradict a quantum process going on either at lower levels or higher levels or both.

Some will say that quantum effects are impossible because the brain is too warm and wet.   This has been answered by Penrose but also is contradicted by the fact we know quantum things are happening in plant leaves and other places already.  Yes, we don’t know how nature does it exactly but it does do quantum tunneling at high temperature today is proven I believe.

The microtubles that Penrose and Hameroff talk about are special microtubules in the brain.  These are more static microtubules that form after the brain comes to a more completed state usually by about the age of 2 or 3 years.  This is fine because it corresponds to a troubling question some have had finding an answer to, namely, why do we not remember hardly anything prior to a certain age?  The explanation for why we have trouble remembering things from early childhood by P&H is that until the abstraction matrix is built and the brain stabilizes enough for the memory storage mechanism to be stable it is not possible for you to remember anything significant.

The cortex I have read is a remarkably uniform structure.  That would make sense and be numerically reasonable in terms of how much DNA would be needed to build a brain.  So, how come when researchers look at the brain they see these common areas where brain functions occur?  There is the possibility that there is a common way a quantum system would find the abstractions and networking to process human speech or vision in all humans.  In other words, the brain is uniform but the process of learning naturally ends up creating the structures we observe in similar ways in most or all brains simply as a matter of necessity in the way the cortex and brain structures itself to those inputs.   That would eliminate the need to program through detailed DNA encoding all the structures we find in the brain.   I’m not suggesting all brain structures are built this way but it could explain some.

The recent success of CNN may lead some to believe that a purely classical way of accomplishing this is possible.  I would point to the blog article I wrote to understand how CNN has enormous challenges that make it unsuitable as a general learning mechanism. Here.

It has been noted the brain has tremendous plasticity and that people who are given alternate electrical mechanical based sense organs as a result of loss of eyes or other senses can regain from the new different input a sense of vision for instance.   If the brain is just a uniform matrix that learns abstractions on top of any input device then that makes sense.   It also means that we could construct almost any new input device for the brain and it could make sense of it, that it would build the necessary structures to process the input and make sense of it.   That explains how animals learn how to utilize smells or directional signals that we are incapable of seeing or feeling.  It implies if we hooked up those senses to our brain we could probably learn to process them and have enhanced senses.  Scary and cool.

P&H have proposed a concrete theory of brain function which has been experimentally verified recently in some of its assertions.  Their theory contends that microtubules in the post-synaptic junctions of neurons and soma are highly regular fixed, water free environments that allow quantum coherence to exist for periods as long as 20ms.   P&Hs theory leverages Penrose’s theories of quantum gravity and discrete space time that he has developed called loop quantum gravity and the mathematics behind it called spin networks.  Penrose is not a lightweight!  He worked on seminal parts of Einsteins General relativity, published with Hawkins on black holes and has produced an incredible array of intellectual achievements.   I would say this man has been seriously underestimated by his peers and the world at large.   Amazingly he is still writing and doing research giving us new insights into the world at 84.

Conclusion

If the current theories of a number of scientists (and myself) are true then every human being has billions possibly up to a trillion quantum computers operating in the brain or to a smaller number of larger quantum computer a billion times bigger than for instance the D-Wave.  Quantum computers power is exponential in the number of qubits so a quantum computer with billions or a trillion qubits is essentially infinitely more powerful than all the computers in the world today combined.   That’s one brain.  Even a small brain would be incredibly powerful compared to conventional computers.  The fact is animals for instance routinely do recognition and complex problem solving that computers are well beyond our CNN networks or conventional computing.

This puts the challenge for AI to come up with a competitor using conventional classical technology out of reach.  It also means many of lifes lower creatures have quantum computing capabilities that explains the complex behavior and capabilities many of them have that are beyond our understanding, even the low life chicken.

I would not be surprised if we found that quantum mechanics is being used by nature in all kinds of aspects of life that we have heretofore presumed were classical chemistry.  As an example I would be surprised if we didn’t find that the mitochondria in some way utilize quantum mechanics to perform miracles of energy production and transport in the body, that all of our senses use quantum effects to enhance their function, that our immune system uses quantum effects to match DNA patterns of threat viruses, that multicellular creatures use quantum effects for DNA expression, DNA repair, replication and regulation.  In short that life critically depends on quantum mechanics at every facet.    I say this because many of these things are well beyond our ability to really explain how they could possibly work using conventional molecular chemistry.

A child starts out knowing virtually nothing but over a period of years this child picks up abstractions.  Some abstractions are useful, some are not.  The chlds brain is processing millions of inputs from senses every second and it must digest and find patterns to abstract from this jumble of data.

Vision recognition computer scientists are always struggling with what abstractions to program into their models of vision to try to match a humans ability to do recognition.  The human does it over a long period in an unbelievably wide range of scenarios.  We call these invariances.

The human recognition system is good at a whole lot of invariances.  You can take a face, put it under low light, lower the contrast, deform it in numerous ways, put masks on it from beards to obscuring it, caricature it, you can fuzz it up, look at it from different angles, go in close or look at the face from afar.  There can be objects in front of the face, different expressions and we are able to recognize.  When there is a difference our brains are able to detect the face and the exception.

Our best algorithms can’t even approach this fecundity regardless of the compute power we throw at them.   We keep trying to understand how the brain does what it does as if we could build an intelligent system and shortcut the process of learning that the brain goes through.  The problem with this is that they have not solved the problem which is really the problem that is the crucial one to be solved:  Namely, learning.   All you do by trying to shortcut the learning process by learning the abstractions the brain comes up with naturally is to delay the time when we have to face the fact we don’t know how learning works.

I believe that the advent of quantum computers and the recent understanding of how to leverage quantum systems even at room temperature by nature and now by us in Correlated Oxide means we are on the verge of a (excuse the pun) Quantum leap in AI.

Our ability to understand how quantum systems are employed by nature gives us a much deeper understanding of the complexity of nature and the ways nature has solved problems beautifully.  It will give us tools to leverage natures jump start on this technology to see what we can do with it.

I hope that this also gives us a renewed appreciation of the amazing creatures we live close to and who were the guinea pigs literally to help us get to where we are.   They are more than a food source.  So far as we know they are our only real relatives possibly in the universe we will ever know.

We are potentially also on the verge of a biological quantum leap in both understanding and capability.    Lastly this gives us an understanding of how special our abilities are requiring billions of quantum computers we are a long way from having computers be a competitor to us.   I believe people doing hard AI don’t understand the path they are on is fruitless and pointless with the methods in use so far.

Article to read more:

Books:

Physics in Mind, Werner Loewenstein

Penrose Books

Quantum Mechanics in nature:

Photosynthesis efficiency tied to Quantum Mechanics

birds-might-actually-be-using-quantum-mechanics-to-find-their-way-through-the-skies

Discovery of Quantum effects in the brain

Microtubules in the brain discovered to be quantum correlated

Smell may be quantum sense

Quantum Mechanics and Quantum Computers:

CNN Convoluted Neural Networks, the best Software AI and some suggest could be dangerous intelligence

How does CNN work?

Why CNN isn’t capable of being an AGI (General intelligence)

Quantum Computer fundamentals

Quantum Microscope

for-electronics-beyond-silicon-new-contender-emerges

MIT TIBCO Chair Scott Aaronson on the limits of Quantum Computers

 and Is the D-Wave Real?

Brain Quantum Mechanics

quantum consciousnes

Orchestrated_objective_reduction

The visual Recognition problem

Scientists claim brain memory code cracked

Other Interesting articles:

uci-team-is-first-to-capture-motion-of-single-molecule-in-real-time/

infrared-new-renewable-energy-source

Quantum world record smashed

Philosophy

elon-musk-we-are-summoning-the-demon-with-artificial-intelligence