This was the response I submitted today to:
https://www.federalregister.gov/documents/2023/04/28/2023-08995/request-for-information-rfi-on-developing-a-roadmap-for-the-directorate-for-technology-innovation
You too could submit your response to this open call for suggesions for he new roadmap called for in new legislation!
Dr. Paul J. Werbos
1. Context: Greater Needs Which TIP Could Meet To the Maximum
The CHIPS and Science Act, and the creation of TIP, offer us all a dramatic new opportunity to increase the probability that the US and humanity as a whole can survive the deep and serious new threats which have arisen in recent years.
From my experience as an NSF Program Director in 1988-2015 (reflected in comments below), working across many disciplines and seeing connections which had fallen between the cracks, I saw both huge unmet opportunities connecting the 10 technology focus areas of TIP, and new risks which my friends in the UN system call “existential”, threats to the very existence of the human species.
1.1. The Global Risks (Which Will Drive the Security Outcomes)
Since my retirement from NSF, I have been working more intensely with the Millennium Project (MP) (www.millennium-project.org), a
global network of leading futures research nodes, and also with the IEEE and the National Space Society (NSS), in order to help as much as I can with the seven grand challenges which I start from in a new book chapter (Consciousness_v5_Bangalore, attached):
=================
1.1. Changes in Technology Which Demand Deeper Consciousness
The website build-a-world.org posts a few of the extensive new materials already available in 2022 from IEEE, NSF and other deep scientific efforts on:
Four Great Fears (“Existential Threats”)
-- Internet/AGI/IOT if we fail to reach Sustainable Intelligent Internet (SII)
– Climate extinction, all of our species and many others
– Misuse (1) of nuclear tech (e.g. werbos.com/NATO_terrorism.pdf) and (2) biotech, e.g. “Borg”
l Two great Hopes GROWTH:
- Quantum leap in attaining human potential, including our connections to each other and our noosphere
- Economically sustainable settlement of space (e.g. National Space Society)
l SITUATIONAL AWARENESS FOR HUMANS IN COSMOS :
Use of AGI and new constellation sensing technology (from electromagnetics to new force fields such as axial Higgs) to observe the sky and better understand the place of our species in the larger cosmos
This paper cannot get deep into all these complex issues, but this list should be enough to wake us up to the fact that better understanding and expression of consciousness is a matter of life or death for all humans.
============================================
Working through the MP, many of us have helped support and upgrade the work of the UN Secretary General Guterres to develop new mechanisms to understand and meet these challenges (https://www.un.org/en/content/common-agenda report/assets/pdf/Common_Agenda_Report_English.pdf). We had substantial discussions with them and with the Chinese about creating new divisions to focus more effectively on the emerging internet/AGI/IOT complex of risks and on climate.
In our recent discussions, the official representatives of most nations supported his call for the creation of a new UN agency on AI, intended to oversee the entire spectrum of new issues, technologies, and RD&D needed to survive the emerging complex of threats.
Unfortunately, this new agency, on its own, may simply not be able to map, understand and advance the whole complex of new technologies needed to survive the minefield of new risks.
The World Academy of Arts and Sciences (WAAS) has offered to try to build up the all-important new network of RD&D and information, to feed into the new global policy decisions and global investments, but even the WAAS does not have the level of scientific and technical depth on its own even to know what the key issues are emerging from the most advanced research from the NSF and IEEE research communities and review capabilities.
1.2. Overview of the New Opportunities and Global Needs For TIP
These gaps and needs in the UN system provide a huge new opportunity and need for TIP to fill the gap, to build the new partnerships, stretching from universities and private laboratories (and other agencies and nations) all the way to the global system, strengthening the constructive role of the best we have in the US.
Ideally, TIP should create a high-level office, sibling to the International Office of NSF, charged with a complete integrated response to all seven of these challenges.
However, because this response must focus down to ten pages, because I have passed on suggestions regarding the climate threat by other channels, and because the nuclear and biowar technologies involve sensitive information, I will focus here on new directions and opportunities to make the world much more effective in addressing the internet/AGI/IOT threats.
2. TIP Focus Areas, New Opportunities and the Internet/AGI/IOT Connected Complex of Risks
2.1. AGI
“AGI”, artificial general intelligence, refers to a rapidly emerging new type of artificial intelligence or machine learning (focus area one on your list).
Today’s culture, ranging from mass media but even to “experts” widely cited in policy and media circles, simply does not have a trustworthy roadmap of what AGI already is, and will be in the future.
The reasons for this are two-fold: (1) a need for more connections between disciplines, which TIP and a new focus area within TIP is uniquely placed to create; and (2) gaps in public education, an urgent new piece of the STEM agenda.
The best true global overview of this problem can be seen by carefully reviewing the diverse viewpoints represented at:
http://1dddas.org/activities/infosymbiotics-dddas2020-october-2-4-2020/dddas2020-video-presentations. The 2020 and 2022 conferences were both organized by Dr. Frederica Darema, who originated the interagency DDDAS effort and whom my wife Dr. Ludmila Dolmatova Werbos and I still assist. In both conferences, one group of highly respected experts would say “X is impossible,” while another presented its demonstrations of X, which in turn was just a first step on a well-plotted roadmap leading to what some people call “artificial super intelligence” (though few of those people know the mathematical principles which make that possible, doable, and risky, but risky NOT to do.)
Another overview of the diversity of mainstream experts can be seen in the background material collected by the MP to prepare for its proposal this past week, in response to an invitation from OpenAI.
Even today, the best strategic roadmap now in existence to make some of the important connections which need to be expanded may be found at: https://www.nsf.gov/pubs/2007/nsf07579/nsf07579.htm. That EFRI/COPN roadmap was the outcome of a unique deep crossdisciplinary dialogue across all the relevant program directors and research areas at NSF, pointing towards a new integrated understanding of mathematical general intelligence (MGI) including both AGI and a new approach to understanding general intelligence in the mammal brain.
The NSF EFRI BRAID topic was also an important positive step, but we now see a much larger world of unmet opportunities and gaps than we did even in 2009.
In a talk to the World Economic Forum, Sergey Brin described how he initiated the great Deep Learning Revolution, which will go far beyond the early narrow work on Generative AI -- work which is large compared to the past but tiny compared to what is coming, for good or ill. Brin describes how he “knew” that deep learning would not work, based on what all the mainstream AI experts told him, until he saw the outcomes of a grant funded by EPRI COPN, a grant to Ng and LeCun which I personally pushed through the system over several types of friction and inertia still in existence. For a link to Brin’s talk, and to my own plenary talk at NIPS 2017, see
http://drpauljohn.blogspot.com/2017/01/deep-learning-and-new-ai-you-aint-seen.html
2.2. Quantum AGI -- Connecting To Focus Areas 2 and 3
By the time of the 2017 conference, NIPS had grown into a major focus of the computer industry with neural networks, due, in my view, to the great efforts of Yann LeCun in computer science. But before that, the rapid expansion of neural networks and deep learning was first pioneered by the World Congress on Computational Intelligence (WCCI) series of conferences, a joint activity of IEEE and of the International Neural Network Society, dating back to 1988. Even today, WCCI conferences are the technical lead in many areas, especially any of those which involve engineering applications, such as the core missions of climate and energy, nuclear and biotech, and space and vehicle technology.
In 2022, IEEE granted me the Frank Rosenblatt Award, its lead technical field award for all of the technologies covered in the WCCI conferences. More important: I attach the abstract of my acceptance speech, which gives a one page roadmap of the past and future of AGI, with citations. That includes a link to the roadmap of 2014
https://arxiv.org/abs/1404.0554 which China (led by Tsinghua, NSFC and CAS) then followed, putting it ahead of the US in many crucial technologies still little known in the US mainstream computer science community. (For example, I have seen their new systems applying machine learning to interception missions in space beyond what any US groups have mastered as yet.)
But my 2022 abstract goes on to say:
=====================
This year, we have opened the door to a new revolution. Just as adaptive analog networks, neural networks, massively and provably open up capabilities beyond old sequential Turing machines, the quantum extension of RLADP offers power far beyond what the usual Quantum Turing Machines (invented by David Deutsch) can offer. It offers true Quantum AGI, which can multiply capabilities by orders of magnitude in and application domain which requires higher intelligence, such as observing the sky, management of complex power grids, and “quantum bromium” (hard cybersecurity). See "Quantum technology to expand soft computing." Systems and Soft Computing 4 : 200031. https://www.sciencedirect.com/science/article/pii/S2772941922000011 and links on internet issues at build-a-world.org.
======================================================
One of the public interest lawyers in the NSS community asked me to file a patent on QAGI as I defined it (VERY different from older attempts to combine quantum and AGI), complete with a discussion of how to do the hardware implementation on the preferred platforms (quantum dots, above all). That reminded me of the time when I worked for NSF, when the Office of General Counsel (Charles Brown) urged me to patent some of my work before then. (See https://patents.google.com/patent/US6581048B1/en). Therefore, I did go ahead and file that patent, now pending. Full implementation across the wide range of applications domains would be a large undertaking; therefore, I intend to use this patent as a way to accelerate the field, as I did with the previous work (without which we would not have had the deep learning revolution!).
In my last years at NSF, my core activities included AGI, electric power and a quantum technology activity https://arxiv.org/abs/0709.3310 which was discontinued after my retirement. As part of that, I was NSF liaison to QISCOG, the interagency group to coordinate quantum information sciences. From that work, I learned that the US faces a huge gap in educating people to understand the importance and use of master equations (which I learned about from the texts by Walls and Milburn and Carmichael, whose group has been a great partner to me in recent years, and from the ECCS Nanohub project). Simple education about the true foundations of quantum physics is a crucial [art of actually getting something useful done from this new crossdisciplinary thrust area.
2.3. Quantum AGI -- Focus Areas 6 and 8
The entire roadmap of AGI, from technologies already coming online to advanced QAGI, and to further generations possible beyond what is envisioned in my patent pending, is relevant to all ten focus areas, in concrete ways I hope we will discuss more in the future.
Focus areas 6 and 8 call out for special mention, however, because of their direct and urgent implications for the international big picture.
Security of communications has been growing rapidly as an issue, threatening areas from currency systems, financial systems, weapons control, to fake news and more. Traditional bromium approaches to cybersecurity were simply not enough to prevent very massive and embarrassing hacks in recent years (which should be well known to anyone whose personal data was stored in OPM computers). These risks are growing rapidly, and do constitute an existential risk to us all. (It may be compared to the growing mass of miscalculation and gaps in information which led to World War I, but this time the potential damage is much much greater.)
With this submission, I will attach a cybersecurity position statement which was passed unanimously by the IEEE Technical Policy Committee on Energy, where electric utilities had the greatest voice -- though other players entered the game as it went up higher in the system.
At higher levels, there has been an intense but quiet debate: do we rely on Mutual Assured Destruction or on a mix of deterrence and defense, to prevent the level of risks which Gingrich and Congressman Trent Franks warned about (https://www.amazon.com/Second-After-John-Matherson-Novel-ebook/dp/B002LATV16/)? Many of us believe that it is growing riskier and riskier to rely on US ability to shut down other countries, when we ourselves are more vulnerable and becoming more vulnerable by the week. Many believe that relying on deterrence is the main lesson of the Cold War, but they need to understand history better. We need to rely in greater depth and understanding on the foundation created by Von Neumann and Morgenstern, who many of us believe should get the real credit for our survival so far. Von Neumann and Morgenstern is also the foundation we built on to create “the new AGI,” including quantum RLADP.
If a hostile adversary (whose identity we cannot really establish) takes over a major currency or fleet of autonomous weapons overnight, the shock and the word urgent will be understood to be gross understatements. Since we have technology coming to dramatically reduce such risks, we really should proceed. Only by doing so in an open, scientific way can we avoid the kind of game play which led to World War I (and could actually kill us all, with great stunned shock as it happens).
3. Assorted Final Comments
Again, all seven of the challenges above merit much more science and connection, within and between each other. This submission is narrower, because the total range of issues hinted at here is already a challenge to our existing networks. But I do owe you a few final thoughts.
The NSF emphasis on STEM is also important to all these challenges -- most specifically to challenge 1 (a fear) and challenge 6 (the grand hope of accelerating full human potential). Collaborations between the engineering and neuroscience/psychology parts of NSF played a crucial role in leading up to the IERI initiative
https://www.nsf.gov/pubs/2004/nsf04553/nsf04553.htm. Deeper understanding of the brain and the mind could take us much higher. The requirements for deeper understanding and enhancement of the highest levels of human potential, unifying approaches from all over the world, have become more and more a matter of sheer survival. Artificial super intelligence WILL be built, one way or another, but we still have hope of raising human minds and souls enough to maintain at least a balance,
as is essential to survival not only of humans but of all our works. (I have chosen not to elaborate on my discussions with people like Teller and Schwinger within the scope of
https://www.nsf.gov/pubs/2013/nsf13554/nsf13554.htm, but this is an integrated strategy for survival.)
The main advantages of quantum RLADP over traditional optimal control, reinforcement learning and approximate dynamic programming (for which I received the Rosenblatt Award) is its ability to cope with “minefield or “needle in a haystack optimization problems. These are problems in which the value function (or error function to be minimized) is highly nonconvex. The patent also extends this to the “Quantum Annealing of Things,” putting actual physical systems like a big laser to be controlled or a vehicle to be protected from exploding into a state of quantum superposition. A great partner to seek would be https://www.desi.lbl.gov/policies/, relevant to the important basic science of challenge 7 but also to strengthening the power of the NIF fusion system to accommodate targets like the D-D pellets designed by Perkins, which could be a huge benefit to energy goals on earth and in space.
