This post announces a new type of quantum computing, which I call true Quantum Annealing (QuA), which offers potential capablties far beyond what is possible with the universal Quantum Turing Machine (QTM), the kind of quantum computing used in almost all quantum computing work in the US today. (Though other nations are less tightly managed that way.)
One possible benefit is a vast enhancement in our ability to detect backdoors in chips we import, like the backdoors which recently led to a huge leak of US information like the entire OPM database. But there are many other possible benefits, as we discussed recently...
This post begins with my initial statement, followed by a quick clarification of physics assumption, then by hardware and application details, finally with WHY this is importat to human survival, in my view.
================================
Initial post:
Serious mystics, from Deepak Chopra to the Order of Pythagoras, keep telling us we need to work hard on positive thinking. But the current flow of news makes it ever harder to envision how the human species might survive in the end. Even if we focus on just the leading two of the "existential threats" ( the most fatal type of climate change and of emerging problems linked to new internet technology,
(1) Climate extinction, the real risk and low cost ways to stop it
http://ww.werbos.com/climate_extinction_risk_and_solutions.htm
(2) Sustainable Intelligent Internet (SII), How to prevent a fast coming disaster which toadys’s regulat8nb strategies alone are only making worse ),
http://ww.werbos.com/How_to%20Build_Past_Emerging_Internet_Chaos.htm
things kappen every day now which make the POSSIBILITY of survival seem very far fetched, if we face up to all of what is happening. For example, just yesterday I heard what Biden's new recovery bill is supposed to do for climate\ and it isn't what I would call paying attention to the science. It smells more like a tribe of war heelers praying to science but making sure reality is kept far, far away... What good are good intentions if their net benefit is zero or worse? (The link above on climate points to some clues, but they need more.)
If they can't even learn and face up to a challenge so open and clear as this climate challenge, what hope is there for them on the internet side, which really requires very difficult real science? By the way, true Quantum Annealing (QuA) is a crucial part of exactly that science and technology, as I will explain.
Why would you even waste energy in playing a game where the probability of survival is near zero in the end? Ah, but what if you were guaranteed millions of replays, such that every time hou fail, you can go back and do it all again and get it right, and remember and exploit every PREVIOUS time you tried? What if you were playing a game where the natural, classical probability of survival is one in a hundred, BUT with QuA you get to play it millions of times if need be, until you get it right? If you believe or are capable of QuA, you are nearly guaranteed to survive, even in a game which appears hopeless under normal rule.
So here is my claim: we ARE playing a game of survival in a cosmos ruled by quantum physics. We CAN do quantum annealing in PSI. There is also a promising R&D pathway to being able to do it in engineering, in design of the next generation intelligent internet. Furthermore, that theoretical sounding thermodynamics I discussed in my last post is exactly what we can use to understand and do this.
Atter all, it comes from Von Neumann... as does the modern Turing machine, game theory and a lot of quantum theory.
I have been holding back a bit in explaining this because it is basically SO easy, once one understands the principles (as easy as E=mc"2), and it has such huge implications for the national security of big nations now in a risky and unstable balance already.
The link on Sustainable Intelligent Internet (SII) at the top of this post includes links to short, light overviews at the bottom on Quantum Information Science and Technology (QuIST), a term used in the past by the interagency Quantum information Science Coordinating Group (QISCOG), on which I was once the NSF representative before my retirement. One of those links gives an overview of five stages of QuIST technology. The first is David Deutsch's great universal quantum Turing machine, the brilliant but limited design which is the basis for almost everything you read these days in the US on quantum computing. The third is Quantum Annealing (QuA), which is very much an entry point to later and more powerful systems. (Claim: the brain of our noosphere utilizes this level of capability. Just as mammal brains have basic powers beyond that of a fish, the noosphere has qualitative capabilities beyond the mundane human brain. Working WITH the noosphere, to save it from us, is absolutely crucial to our own chances of survival.)
But: MY CONCEPT of true Quantum Annealing (QuA) turns out to be very different from the RELATED but weaker conceot in the present literature on Quantum Annealing.
More precisely: a Basic Quantum Annealing (BQuA) box would include: (1) an array of N "flux qubits", N binary processing elements, call them u(i) as i = t to N; (2) a simple standard "Ising type" Hamiltonian, H, such that the wave function psi( the vector u) evolves according to a term linear in u and a term of the g=form Hij*ui*uj, where the USER gets to specify the coupling constants Hij; (3) a way to initialize the u at a start time t0; (4) read out and coupling to the external world.
BUT THIS concept of BQuA includes BOTH the version of quantum annealing which is now common in the literature? (e.g. the vast literature available for the company D Wave,
A numerical implementation of" quantum annealing"
B Apolloni, D De Falco, N Cesa-Bianchi - 1988 - cds.cern.ch), AND a useful special case of what I call true Quantum Annealing, what I assumed in my review.
I was shocked to learn that THEIR variant of BQuA is not a true quantum computer at all, but that it is very straightforward how to modify that kind of design to produce a BQuA box which IS a true quantum computer, and permits a wide range of important capabilities. For example, one can hook it up so as to probe for backdoors in chips one buys from risky sources, like the hardware backdoors which recently led to the biggest, most harmful compromise of US government data systems like the OPM personnel database.
I was indeed stunned to learn that the DWave community and its followers have not yet figured out the simple things needed to make it work as I expected it to work.
In essence, the DQuA (what they do) starts from a user-chosen vector u, and orbits AROUND the space of possible values for the vector u, hoping to use quantum tunneling to get from regions of low energy H to regions where some states have lower energy. (Yes, they have lots of details.) They use "annealing" by assuming a COUPLING constant multiplying the term for interactions across bits, and dialing the coupling up or down. It sounds like an electron IN ONE state, tunneling at times to get to a better state.
But true quantum computers, like Deutsch's universal quantum Turing machine, isn't like that. A true quantum computer is based on QUANTUM SUPERPOSITION, like the ortinal Schrodinger cat. Deutsch, a major developer of the multiverse idea, described it as a kind of parallel computer, harnessing the power of parallel universes to perform different computations IN PARALLEL , and pick out what you choose in the end. How many parallel universes? Well, with 20 qubits (Q=20), it is slightly more than 2"20 universes in parallel, a million. BOTTOM LINE: there are lots of working Deutsch style quantum computers working out there now, some with many more than 20 qubits. thisis not science fiction. This is today's working technology. (You may ask what good it is, but that's another subject.)
A true QuA box should be able to manage Q real qubits, not just "flux qubits" which may or may not be real qubits, in parallel. Thus a classical simulated annealing or stochastic search box should be able to find good values of a vector u, to minimize some function H, but a true QuA box should be able to search through a million times as many possibilities if Q is 20.
How could such a thing be possible?
To begin with, the hardware to implement u should be the kind of hardware which is used to hold qubits in Deutsch style quantum computing. I can think of four good candidates. D-Wave uses something SIMILAR to one of the four -- the SQUID, first used in Maryland for quantum computing and widely used in China under Pan Jianwei. But D=Wave's is only similar; they set it up in a way which leads to far more noise than necessary, but that is not the main problem.
The main problem is that they don't seem to understand what ANNEALING is, what any real condensed matter physicist would know. (I have cited Chaikin and Lubensky many times here.)
One major ADVANTAGE of true QuA over quantum Turing machines is that decoherence is not a problem, in principle; losing information through noise is not a problem, when we don't want to perpetuate initial condition information anyway.
BUT: in true annealing, COUPLING TO THE ENVIRONMENT TO SHED HEAT is the essence of how the system state u goes to minimum energy. It is not about adjusting coupling coefficients, but about shedding energy to the environment. If one can do that, thermodynamics is what does the selection.
But this discussion leaves out one more crucial aspect of any optimization system. How does one define what is to be optimized, what the user WANTS the system to accomplish?
There are three possibilities. The easiest to implement is ENDOGENOUS SCORING, where the user's goal is simply encoded into the coupling coefficients Hij, and that's it. A quadratic optimization system. A fancy quadratic program?
Of course, that can be done ITERATIVELY, where an optimum is calculated, evaluated against more complete nonlinear standards, and the optimization problem revised, until the user is happy. D=Wave certainly has worked hard to develop that kind of interface
Second choice is INTERNAL COUPLED SCORING, where
a process inside the big computer system scores each possible u, and the overall score is fed back to the QuA box, perhaps to a set of power utility input bits. In truth, new papers are needed spelling out how the thermodynamics of this works.
Third is Quantum Annealing of Things, which is essentially the same except that the scoring system need not be a computer program but may be an external object, like a chip to be assessed. A macroscopic Schrodinger cat.
==================================================================
ASUMED PHYSICS:
addendum> What if you don't believe we really live in a multiverse?
In fact, I do not believe the usual psi dot = i H psi view of the deepest level of physics. But it works well enough for THIS kind of QED
type technology. But WE are certainly macroscopic Schrodinger cats, like then photon statesdepicted in
https://www.researchgate.net/publication/327164414_
=========================================================
SOME DETAILS:
Thanks, Shiva!
Since you have a rea; interest in these issues, I should add a few details/
I NOW see the importance of the three stage approach to developing and building real QuA technology (which complements PSI efforts). But even stage 1 requires considering WHICH ACTUAL hardware platforms to use. I would claim that a rational program would not choose just one, but should juggle what can be learned from FOUR parallel coordinated efforts learning form each other.
These would be:
(1) true QuA based on quantum optics platforms (also used in some QTM work)
(2) SQUID platform, which D-Wave uses, but more like what TM uses
(3) Massive quantum dot arrays, like Samsung TVs with billions of optical processing elements in a square meter or less
(4) High frequency electronic systems, like what is used in GHZ electronics for cell phones
Developing better component models is VERY crucial to better design capability. Yeshua is closer to some of the experiments needed to get better component models than I am, but the link to Werbos and Fleury at ResesrchGate also has crucial information.
how could you and I be macroscopic Schrodinger cats if the cosmos MIGHT only be 4D underneath? It seems our states are like the photon states in that paper. But for now, the Deutsch physics are good enough, in most respects. The thermodynamics pf the interface are a trcicky issue, to be resolved by empirical work in the end.
=========================================
WHY THIS IS IMPORTANT TO HUMAN SURVIVAL:
hanks, John, for focusing more attention on this important question: HOW and WHY would the development of QuA technology reduce the probability of human extinction? I tried to cover MANY important points in that long and complex point, but this one is the bottom line.
In fact, I spent weeks pondering that question beforeI decided to post these details. WOULD the open, transparent development of all three stages of QuA technology really help with human survival, or hurt, on balance? There are actually many pros and cons; new technologies often cause more harm than good.
For me, the decisive reason is as follows:
AT PRESENT, the development of the global internet appears to be moving into directions which really do threaten the extinction of the human species, even more than the worst forms of climate change do, even sooner, in my view, based on lots and lots of information about Artificial General Intelligence (AGI), Internet of Things (IOT), and movement of the world economy to internet based control systems. IT IS POSSIBLE to design internet platforms in a way which prevents such risks (see http://ww.werbos.com/How_to%20Build_Past_Emerging_Internet_Chaos.htm ),
but the sheer inertia of a different approach seems overwhelming, for many reasons.
In a way, the key problem herte is the relatin between work/human-life and economic production. The Millennium Project (www.themp.org) guided and integrated dozens of sudies of the future of work, across dozens of research nodes all over the world, and came to some pretty strong conclusions: due to "automation", new computer systems destroying jobs at a rate far greater than people expected in the past, the most "optimistic" scenario led to only 70% unemployment worldwide. Some would say: "What's so scary about that? If we just lose 2/3 of the world population, or create a new gig economy or government support for the unemployed, why would that change things?"
Several reasons why. For one thing, the experts in these studies simply did not know the full range of advanced "AGI" technologies already in the pipeline, already being deployed in advanced organizations a few of us have more direct access to. See the link above. There is a whole new spectrum of technologies, and there is NO JOB ON EARTH for which a human is absolutely necessary, if the most advanced technologies in process are used.
In fact, I had a key role in the initial development of the most relevant family of technologies, "RLADP", Reinforcement Learning and Approximate Dynamic Programming. (Again, the link says more. For even more information and evidence and the connections to brain, see https://drive.google.com/drive/folders/0BzYEn42Vg7DKbFBhbm45ZFpubWM?usp=sharing.)
For the future of the internet, this is the key point: the global integrating RLADP math and designs which we already have in hand are designs which can manage the ENTIRE Internet of Things, basically the entire world, with NO HUMANS NEEDED. Humans have a lot to offer, in many ways, on many levels, BUT THE PEOPLE ON THE GROUND FLOOR ACTUALLY BUILDING THESE SYSTEMS, LED BY CHINA BUT GROWING ELSEWHERE TOO, ARE MOVING FAST AND ARE UNDER LITTLE PRESSURE TO FULLY UTILIZE AND SUPPORT THE HUMANS. Lots of OTHER people have been funded and empowered to do fuzzy papers on empowerment and regulations and human centered internet and what properties the new apps are morally required to have, BUT THOSE FOLKS DONT WRITE THE CODE, SPECIFY REAL CODE, OR CHANGE THE HORRIBLE OUTCOME IN PROCESS NOW. It is moving very fast.
One of the minor bits of news which hit me was the new set of rules for Hong Kong, enforced by the folks organizing cybercontrol in China in general. There is a new rule: rules and systems to encourage transparency in the running of corporations are now forbidden, prevented by the State. The "Communist" party which has a new rule that the controllers of corporations are absolute powers, such that workers and citizens and investors are allowed no information. Karl Schoeder's novel Stealing Worlds is quite realistic in portraying future corporations which can be TOTALLY cyber run.
So then who DOES need humans (aside from humans, and do we count/)?
In fact, I have met people in positions of power in the cyber world who really DO expect and desire that no more "meat creatures" like humans should continue to waste energy by living on this planet. They have powerful allies in China and EU, to be sure. In fact, my direct personal awareness of those groups and their power was part of why I hesitated before making that post.
BUT: here is a problem they are not aware of. They are using a very basic level of RLADP intelligence, no higher underneath than the intelligence of a fish, or maybe even a cockroach.
GIVEN enough power, even a vast network of cockroaches CAN take over a planet, but that does not give them the true foresight (let alone attunement to our noosphere) to handle a wide variety of fatal challenges. I am reminded of the trilobytes, who proliferated ever so widely on a certain path... and then because of the weakness of that path, all died. If we start to depend (much faster than we think, much more than we think, as the new Chinese plans show) on the kind of RLADP or even lower systemsmoving every so fast, its weakness plus our dependency threatens our very existence. And in truth, that is the set of paths to extinction which worry me the most right now.
I am tempted to say more. There are other important challenges and important requirements for survival, and other benefits of true QuA. (I mentioned how QuATh, stage 3, quantum annealing of things, can detect backdoors in imported chips. But it could also vastly amplify the information throughput of the Breakthrough Initiative SETI project. And... other apps.)
Unfortunately, I worry that the balance of emotions and of power in China right now may prevent the kind of international collaboration, like the SII proposal for the new UN Office of existential Risks, which would have some hope of making this happen in a smooth way.
If, as some in China and Russia propose, there is a more complete decoupling of their economy/system and the West, with human-centered SII developed in the west while China and Russia follow the single actor RLADP path to human annihilation, internet decoupling might in fact give us a chance to survive even as the internal contradictions of their own new system destroys them. I hope it does not go that way. But total extinction of all of us would be worth.
QuA is one of several key elements to make the SII path.. more possible. If the noosphere is smart enough, it might help.
=========================================================================
Stan Klein of Berkeley finally asked:
--------The issues you are raising are super, super important. Is there some organization or group that is working on that problem? There surely are many others who are worried about that, but for some reason I'm not aware of them.
Yes, this is the bottom line. true QuA as defned here is a research area, with short term and long-term risky aspects. NO ONE has followed THIS approach, prior to its being posted here on the web! But of course D Wave has lots of important connections, as do the many links at www.werbos,com.
Above all, THIS MONTH the best hope has been an effort led by Jerry Glenn of www.millennium-project.org to support the creation of a new Office of Existential Threats in UN HQ in NY, under the Security Council. Through the Millennium Project and other media, I have proposed some crucial details of ow to do this: (1) DO include the division aimed at the precautionary principle, to assess and strategize how to prevent extinction of the human species by the worst forms of climate change; (2) ALSO include a sister division to study and strategize Sustainable Intelligent Internet, which requires many components, including a Quantum Information Science and Technology (QuIST) component, to include QuA as one of many topics;
(3) to prepare (2), support Jerry's call for people to contact their nation's representatives in the General Assembly to support an initial feasibility study, to combine and link not only the usual kind of policy studies MP has supported in the past, but also to connect to groups like the new MST intelligent systems research consortium, IEEE journals interested in focused new special issues, and industry-based discussion groups, to make SII possible not only in law but in technology.
=========================
My worries about current trends in China are based in part on papers I have read by Yan XueTong of
Tsinghua, which represent a new ideological synthesis similar in spirit to that of Zhu Xi of about 1000AD, a version of Confucianism well-tuned to unify feudal overlords and military, getting rid of the deeper qi and spirit connections from Meng Tzu. The "millionaire's club" and the military support a new program, which is the exact opposite of what Mao rightly fought for when confronted with something very stifling and dangerous and narrow. There are natural myopic economic forces which support that knd of coalition and filters, but natural forces also led the trilobytes to extinction. Xi himself has at times seen further, but does his understandable disrespect for the blind spots of other people strengthen his own blind spots to a fatal degree? I hope not, but there is now great grounds for worry.
No comments:
Post a Comment