There are risks and opportunities both far larger than most people (including ersatz experts who do not know the underlying technology) begin to imagine. Those risks include threats of extinction of the entire human species far more imminent than death by climate change.
Here is a key slide from my talk on that "AGI" challenge, a talk requested by the world President of IEEE: In brief: the old choices on the left, which a new Director of Engineering at NSF asked us to help implement (in 2014), WOULD have resulted in extinction of the human species, in my view. I met the key people, saw the details. One of the pathways to extinction is the pathway depicted in terminator II, closer than some people want you to know. See https://www.youtube.com/watch?v=rE38LG2DVAc, from Stuart Russell, whose text was cited for years as the number one text in AI, but of course these videos upset certain people in the military closer to implementing that technology than even Russell knew. (How did I know? It is a neural network technology. Russell's texts understated what neural networks could do, but my PhD thesis was the first step in creating that technology. See scholar.google.com, thesis with twice as many citations as Walter freeman's best, but many of them in Chinese. Hes, China is ahead.)
The RIGHT column in this slide describes an alternative pathway, consistent with human survival. The right column depicts what I now call "SII", Sustainable Intelligent Internet. Quantum information technology (QuIST, Quantum Information Science and Technology) is ONE of the five key elements required for SII.
To what degree is the mundane human brain a quantum computer? We have discussed that already in huge detail on this list. In the end, I see several POSSIBILITIES, and an open need for science to FIND OUT which is true, as per: https://www.nsf.gov/pubs/2007/nsf07579/nsf07579.htm . So far as I know, nothing in recent years has changed the bottom line of what we know. Also, the role of quantum theory in human consciousness is not so central as some have consiedered; see my paper for the Festschrift for Henry Stapp, linked to at werbos.com/religions.htm.
But when we BUILD intelligent systems ("AGI") or try to understand noospheres, we are free to consider the whole range of designs allowed by QuIST, not restricted by neuroscience. That is a technology which will massively change the internet of the future, one way or another. In my view, the US and EU are massively behind both China and what is possible in QuIST. WHY are they so far behind? That is a very interesting question, well within the scope of the Cosmos and History (CH) topic, but in this post I will try stay within the AGI/Q scope.
The key point is that there are five basic TYPES of general quantum computing:
(1) The universal quantum Turing machine, brilliantly expounded by David Deutsch of Oxford. I think of this as aleph zero quantum computing, because it is as infinite and unlimited as a Turing machine, in manipulated infinite bit strings, exactly like the INTEGERS which are infinite in number. Infinite as in aleph-zero. Cantor's mathematics of infinity is a crucial foundation for understanding what is going on. Aleph-zero quantum computing is the kind of quantum computing which uses "qubits," the quantum version of the bits that an ordinary Turing machine relies on. Binary variables. When enthusiasts proclaim that "all advanced real life is digital," they assume that binary digits cover everything.
(2) CONTINUOUS VARIABLE quantum computing. Until I went to the 2015 SPIE.DCS conference on "Quantum Information and Computation IX" (S1127 in the annual SPIE conference on technology for defense and security held in Baltimore) I was not aware that ANYONE in the US was working on the more powerful generalization of QuIST which is possible when using continuous variables (real numbers) instead of just binary digits in quantum computing. Since there are aleph-one real numbers, I think of this as aleph-one quantum computing.
I was invited to speak that that conference by Howard Brandt, one of the VERY PIVOTAL breakthrough thought leaders little known to the public or to Congress the people. I knew Howard from QISCOG, the quantum information sciences coordination group, which I was then a member of (as I was one of the people at NSF funding QuIST research). That group coordinated work in QuIST all across US government agencies, of which NSA and DARPA were most prominent. Howard Brandt was the real brains behind all that, and I met him at a different, larger, more operational government QuIST conference in Baltimore.
At SPIE 2015, Brandt intended to create a kind of explosion on the scene, by showing us the power of the new continuous variable paradigm. ONLY LATER when I was writing my paper for the special issue of Quantum Information Processing which emerged from that session did I learn how Seth Lloyd, a professor at MIT deeply respected as a theorist and educator in the QISCOG, community, had already published significant work in continuous variable quantum computing. the four speakers which Brandt had invited in order to create a new wave were me, two theorists doing basic research (one Chinese/Australian and one Indian/American linked to Subhash Kak), and a US contractor who seemed to be saying "and yes folks, Howard is bringing it HERE, and we can implement it." But Brandt himself died unexpectedly just two weeks before the conference, and was unable to perform the great show he had planned for the world. The researchers showed clearly that they could make and break codes far beyond what aleph-zero quantum computers could handle, but we in the US never got the benefit of that. Again and again, I see US people counting qubits like counting sheep, oblivious to the giant ravens flying over their heads.
(3) The D-wave TYPE of quantum computing, https://en.wikipedia.org/wiki/D-Wave_Systems, which It think of as aleph-zeroN, as Vancouver technology, or as quantum Boltzmann. They still use qubits, but INSTEAD OF RUNNING programs (the way old Turing computers do) they use a different ARCHITECTURE for computing, more suitable for LEARNING (hence neural N). Old fashioned Turing thinkers, educated in mainstream computer science, tend to wrinkle their faces in disgust at this silly irrelevant type of architecture, designed to perform optimization tasks rather than general purpose Turing programs. But in the neural network field, we know that all types of brain-like intelligence can be MAPPED into such optimization tasks, allowing unlimited power for those who understand enough to use it.
(4) ANALOG Quantum Computing (AQC), one of the two main themes of my paper with Dolmatova in Quantum Information Processing https://link.springer.com/article/10.1007/s11128-015-1146-2 . In that paper, we cited previous work by Siegelmann and Sontag (and others) showing that analogy neural network types designs ALREADY achieve a computational capability beyond what Turing machines are capable of, by combining real numbers and network architectures. Proper use of such architectures (the main foundation of mundane brain capabilities) ALREADY gives post Turing capability, even without quantum mechanics. And so half of our paper discussed how neural network type architectures can be COMBINED with continuous quantum computing, to get another while level of capability, which I think of as aleph-two type quantum computing. (D-wave is of course a subset of this.) An AQC level computer should be able to break ANY code generated by ANY lower level of intelligence (e.g., crack any bitcoin type of currency or even the new digital rembini) if there is enough data available, and the data requirements are far less than with lower level intellligent systems.
(5) Crosstemporal AQC. The other half of our QIP paper, though we VERY carefully left out many details because this could become very scary, especially in a world where people can barely avoid killing themselves with autonomous drones (the baby predecessors to what you could see in Terminator II). I have the impression that China, New Zealand and Australia have the greatest chance of actually getting that far, but the emerging new internet shows great risks of getting at least half of us killed in the US in the coming decade through inability to rationally deploy even pre-quantum security technology represented on the slide. Still, those risks are a matter for CH analysis, the challenge of staying alive before we are ready to grow out of the baby stage. Last year I even saw possibilities for South Korea or Japan to get into the AQC game, but after the new virus that seems less likely. Crosstemporal AQC, like crosstemporal PSI, entails a mix of risks and opportunities, not advisable for groups who do not know what they are doing (like most nation state governments on earth). Best of luck. we really need it.
No comments:
Post a Comment