Saturday, July 16, 2016

NATO workshop on predetection of terrorists

The Millennium project has asked me to speak next week at the NATO workshop on predetection of terrorists and the like. This was one of the workshops described at:,
but it was postponed a couple of weeks from the initially planned date, to fit with other events within the world future society.


I spent some time thinking about the sensitivities. They will post the summary paragraph:

Neural networks, related technology and emerging uncertainties

Recent years have seen a vast explosion in interest in deep learning for applications like face and speech recognition, due to an NSF grant we funded in 2008 which proved that methods we had developed decades earlier really do outperform old paradigms when given a chance. More recent fundamental work can change the game in many ways, some for the better and some for the worse, tightly linked to issues like neuropsychology, like understanding insanity, and related technologies like quantum computing and crypto. 

They also asked for a bio, so I sent:

Dr. Paul J. Werbos, a Fellow of the IEEE and of the International Neural Network Society, received the Neural Network Pioneer Award and the Hebb Award from those societies, their highest award for fundamental substantive contributions.  His invention of a general form of backpropagation in the early 1970's provided the foundation not only for the first rebirth of neural networks in the 1980's but also for the more recent rebirth of deep learning. Before his retirement from NSF in 2015, he led three areas of research: adaptive and intelligent systems, emphasizing neural networks; quantum theory and modeling for electronic and photonic systems and devices; and electric power. He led the development of a special topic in 2009, for crossdisciplinary research in engineering and neuroscience (COPN), which funded the specific award to Ng and LeCun whose results led to the renaissance in deep learning in the US. 

(Gee, I didn't mention my three years coordinating the joint DHS-DNDO "ARI" research program addressing nuclear terrorism research. Whatever...)


I elaborated a bit by email:

I look forward to discussing these issues with you all, although there is no way that all the important themes can be handled in detail in the available time. Also, a lot of things I would want to say really should not be said in public. None are classified; I do not even have a security clearance, but there are other kinds of sensitivities. Also, I notice in the Hillary Clinton case that many are now prepared to put people in jail for saying things which were not classified when they heard them. I will feel a bit safer in email discussion, if
you all have the usual safeguards that gmail normally provides as an option (https always on, and double identity verification). I am also amused that one of the evaluations of Hillary Clinton said she would have been more secure if she used gmail instead of her own server. 

But... until I hear more about your servers... maybe I should limit myself to things I **MIGHT** talk about in this session.

Neural networks, and the next generation of deep learning, certainly have the potential to predict all kinds of things much better than anything anyone has seen as yet in the US. Deep learning as the computer science world has recently seen is just one part of one step on a progression which has already been mapped out at a mathematical level, with lots of allied applications. 
But certain caveats apply:

(1) The risk of unintended consequences is very severe, and has substantially limited the applications side of what I have done in the most recent years. There are many risks of centralization of power, well beyond what the Constitution envisions, some similar in flavor to a movie I finally watched after sustained pressure from my son: Captain America, the Winter Soldier. I have also learned there is more truth than I had expected in their extreme attitude of "Trust no one." I am glad that the 911 report is being declassified, and I hope people will understand at least the obvious implications. 

(2) The quality of results in a domain as tricky as terrorism still depends,
no matter how good the algorithms/architectures are, on the adequacy of the data available, and the richness of the conceptual basis. This leads to issues such as cryptography, cybersecurity and the dark web, for the first half of it. It leads to issues regarding what we know about brain and mind, for the second half. Huge changes are possible in both large areas. For what it's worth, I did some breakthrough work on the best available high resolution >100 channel data from prefrontal cortex this past month, making new connections, but it seems the journal will take more time than I expected before I can talk about it (more a legal issue than a security issue, but law and security do intersect). 

(3) The US has fallen so far behind in some of this that our decision-makers do not even know how many ways they might be blindsided. This, just in the past few years. I have to admit that Lamar Smith gets a lot of the credit. 

(4) One area which I would feel less worried about discussing is quantum computing and communications. The whole world now knows how
digital quantum computing, as pioneered by David Deutsch of Oxford (and misattributed to folks who deserve about as much credit as Greek philosophers who said "I think there are small things in this universe") 
allows a degree of security, with quantum key distribution, not found in 
classical communications, and conversely that digital quantum computing may allow us to break codes which had seemed unbreakable. But
few people know about the implications of continuous and analog quantum computing (AQC), as in a couple of papers I have posted

America's lead in a lot of these areas ended, by coincidence (I think) at about the time when Howard Brandt died. Howard Brandt, among other things, was the key technical adviser to the US security community of quantum information technology, and he also led annual workshops at SPIE/DSS bringing together the various disciplines which feed into that. In 2014, he invited me to talk about some new results related to that area, and also invited speakers form India and China well ahead of our folks in applying it to secure communications; I was supposed to see him again in the 2015 version of his workshop, to discuss substantial breakthroughs of major technological importance, but he died unexpectedly just a week or two before the meeting. In the past, the US held the lead, but as of now China and the EU are the only places with the requisite laboratory capabilities. Are the Chinese actively moving on things basically unknown now in the US (again with lots of credit to Lamar Smith)? Of course, I do not know, but I do know that the key guy in Sichuan has moved up to Tsinghua, at the very center of power in China.  

It turns out that time and triple entanglement are right at the center of some new capabilities which I had not intended to talk about, except to two or three people, until we were ready to demonstrate phase three in the lab. But thanks to Smith... well, maybe no one will believe me if they do not know the physics.

By the way, I have wondered whether collaborators of Aharonov in Israel might be willing and able to understand and follow through on what the US no longer can. 


All for now. 

No comments:

Post a Comment