Since the time of Aristotle, many people believe that the real purpose of life is just to find more happiness for themselves and others. But what IS happiness then, and how does it work? Having studied brains and minds, I would argue that this simple English word "happiness" is one of those very tricky words which should be handled with great caution, like a loaded gun, like another important word "consciousness." Since consciousness is as important as happiness, please forgive if I break off,and give an example of how people get confused when they take words good granted too much.
A cartoon in my mind: A famous wise guru sits under a tree, meditating on "what is consciousness?" He looks into the distance and thinks... "It is really about coherence... about illumination... Eureka, I have it! It must be millions of little lasers in the brain, yes, coherence in illumination". In fact, one of the more famous theories of consciousness these days is exactly that. There are many, many examples of bad puns misleading the highest levels of policy thinking and science; I pick this one now only because it is clear and related.
Happiness.. a tricky word but one we still all use. Many people do very serious research into aspects of happiness, of what makes people happy, and also the economics of happiness.
This morning I woke with clear understanding that the past 4 weeks, recovering from major surgery, were basically more happy than the preceding year and a half (start of retirement February 2015), despite much higher levels of pain and constraint. Is that weird or what? What can be learned from that, and what does it say about the general human condition? But if I am careful.. I'd say that the periodic of greater happiness goes back two months, to early March, start of our Mediterranean cruise/tour, which I have also posted about.
Before the cruise (BC), .. I have found it overwhelming to try to really face up to two major challenges: (1) what can I do to reduce the overwhelming probability that the entire human species goods extinct within just a few thousand years (possibly much sooner if some folks have their way) and (2) what can I do to maximize the inner spiritual growth which remains important to me regardless of whether extinction occurs? These two questions are the real bottom line, in my system of beliefs. All the economic prosperity and military victory in the world are meaningless, dust in the wind, if everyone just dies after that. And yet, these are very difficult questions. For simplicity, let me name these two issues "species survival" and "spiritual growth". (For the record, I still discuss spiritual growth with Yeshua, and remember " What profiteth a man if he takes ownership of the whole earth but suffers the loss of his own soul?". Some basic logic here.)
In retirement, with a (probably) adequate pension, I could afford to just ignore these core questions. Like many others, I could just relax, play golf, read amusing stuff.. whatever. But I still remember the rictus of a fake smile on the face of a Buddhist nun on chicken-dragon mountain in Korea, close to achieving her ghoial of total detachment and nonexistence, about to dissolve like dry paper dissolving into powder.. the smile of success and the dawning horror that this isn't what she really wanted. More analytically, I remember the logical foundations in my paper published in Russia and posted at www.Werbos.com/Mind_in_Time.
In a word: responsibility. I have been overwhelmed to the breaking point by this sense of responsibility or duty, , trying to keep on top of many many things. But: recovery from surgery was different. My sense of responsibility does not tell me to push my car or my body past the breaking point when the predictable outcome would be loss of the vehicle. So for just one month I told myself: "table the larger responsibility except for easy low-stress things. This month, the watchword is not responsibility but healing." So I have felt mandated just to heal. Mandatory listening to the body, which more and more implied a lot of simple relaxation ( in addition to medical problem solving, deciding when to take long walks and so on).
Relaxing -- three main activities, as I sit in the approved position in the big soft leather couch in the living room with legs propped up on the coffee table: (1) "the watch", cycling through CNN, France24, RT and now CNBC, listening on many levels but trying not to intrude in an inappropriate way; (2) lots of puzzles of the Sudoku family; (3) lately, more and more playing with brain data, just having fun with it.
Brain data: nice entertaining games have a finite scope and a finite span, a toy one can play with without the angst of worrying about inscrutable stuff. In my case, I had been invited to submit an extended abstract to a journal on systems neuroscience because of my chapter in a recent book edited by Freeman and Kozma; the abstract promised new analysis of actual time-series of data measured on the brain, to test a few predictions of my theory of how brains work (published in part in the journal Neural Networks). The abstract was accepted, so now I get to explore the actual data for a few months.
This game has been marvelously entertaining, but at times I still wonder: is it a bit irresponsible to just fall into the joy of discovery, and risk helping evil people misuse the resulting technology? Well, I have not written the actual paper yet, and the larger context is hard to sort out anyway, so for now I just explore and learn.
And it is amazing what basic things one can still learn. I certainly read many things about brain data and new technologies to study the brain before this year, at NSF, which gave me more access to that information than one can get anywhere else. But going to the front lines myself.. moved me up a quantum level in understanding the basics here.
Studying brain data reminds me a lot of the 1980s when I studied time series data on energy. I worked at EIA/DOE, which produced lots of energy data and analyses, including the Annual Energy Outlook (AEO), the oifficial energy forecasts sent to Congress and the public every year. These forecasts came out of big computer forecasting models. By the time i left EIA, I was lead analyst for long-term futures, and I had myself built the models responsible for more than half the energy serctoir, both supply and demand. Most of these were a type of model called "econometric."
The basic idea in econometric modeling is to come up with equations which predict future data as a function of past data and bif "exogenous assumptions" like the choice of tax rates. The idea is to study the energy sector, come up with several possible theories of what equations to use, and then do lots of studies testing theories against real time-series data. I was especially proud when my model of energy use in industry had only 1% error in predictions years in advance, the best anyone ever achieved there. (See my papers in Energy: The international journal, 1990.)
I still remember a meeting where my bemused boss said: "Paul has this interesting odd idea that energy modeling can be done like science.. testing theories against actual data, connecting theory with reality.." In actuality, time series data was NOT the only primary information out there.. but connecting to the reality of time-series data was an essential part of the game. Many things I learned about energy data apply to brain data. (Also, the brain itself analyzes the time series data of its experience.)
A lot of very basic things need to be done, to connect theory with reality, in understanding intelligence and consciousness in the brain. (No I am not saying that intelligence and consciousness can only exist in wet brains; see Mind_in_Time for the basics. A certain level of intelligence exists even in the mundane brain of the smallest mouse, and we can benefit from understanding even that "meager" level of intelligence, which can handle IT tasks still beyond today's technology.)
The gap between theory and real empirical time-series data is a lot more serious in neuroscience than what I saw in energy modeling thirty years ago. Which is the real science, following the real scientific method?
What do we need to do to create more true science here?
What do we need to do to create more true science here?
For example -- intelligence in brains emerges in large networks of neurons, Biological Neural Networks (BNN). There is a large and important field of computational neuroscience which develops mathematical models of such large networks. (There are also bigger models of small numbers of neurons, beyond the scope of this blog post.) Those models mainly come in two families: (1) smooth models based on ordinary differential equations (ODE), in continuous time, without clocks (asynchronous); (2) spiking neuron models, which are also asynchronous, but which assume that the inputs and outputs of neurons are discrete pulses, "on" or "off." Many people think they can develop useful technology by simply implementing today's models in computer hardware, directly, without really understanding the engineering challenge of how to learn the IT tasks which real brains learn to perform. Billion dollar investments have been made in that direction, mobilizing political iron triangles and the usual misleading DC (and EU) PR, getting in the way of actually learning how to do what they promise.
But has anyone ever tried to compare the predictions of such mathematical models with actual time-series data of what neurons input and output? This question is incredibly basic, and I find it amazing how little attention it has received so far. But I need to be careful to summarize the story accurately.
There is another large and important part of neuroscience, systems neuroscience, which has worked hard to use real empirical evidence to understand how brains work and how intelligence emerges in brains, at a whole-systems level, in living awake animals (including humans). Part of my message here is that we really need to build stronger connections between systems neuroscience and computational neuroscience, in order to develop useful mathematical models capable of more serious learning and intelligence. (One of my fears is that idioits might deploy this technology in a way which dumps us into Terminator or Winter Soldier outcomes. That is a deadly serious concern, entangled with very complex and irrational political movements all over the earth today.)
I am very grateful to have had a chance to work closely with Walter Freeman and Karl Pribram, two of the great giants of systems neuroscience (sadly both recently dead), and to have listened to a number of others. Yet I still remember in 1964, the first day of Harvard's only undergraduate course in neuroscience at the time, when the teacher Charlie Gross (a former Pribram student) said: "I would like you all to think hard about how much we would understand radios if we studied them the same way we now study brains. We pull out this capacitor, and the radio emits a whining sound, so we call that the whine center, and then just throw away the radio. We pull out this inductor from the next radio, and the radio sparks, so we call it the spark center, throw it out, and so on. Couldn't we do better?"
How can you find out how such a complex system really learns new and complex things, with such limited data to work from? For myself, I tried to absorb what systems neuroscience really learned from a lot of high level experiments, but focus more on the more fundamental mathematical challenge of trying to understand that very small class of neural network designs actually capable of learning to predict their environment and achieve what they seek in that environment (similar to happiness). (See www.werbos.com/Mind.htm.)
But none of that really tracked the actual time series of what neurons input and output, down at the level where basic computations are performed.
Now -- I am not going to say that no one has ever tried to make this vital connection. To the contrary. I was really delighted this past week to read the super important paper by Fujisawa and Buzsaki (and ..) in Nature 2008, which looks like the very best effort ever made yet to bridge the worlds of time series theories of learning and actual time series recorded from biological neural networks (BNN). The paper is a real eye-opener. Also an eye opener is the underlying database information on pfc-2, a database available to the entire research community posted at crcns.org. (This database includes what fujisawa and buzsaki used in their paper, and a lot more. The crcns website is one of the most important organizational achievements ever in this field.) It is the best of the best, an excellent starting point -- and a sober indicator of work still needed.
If you read the paper itself, three important points stand out. First, they were able to do useful "spike sorting," worthwhile estimation of exactly when the neurons close enough to their probe (>100 of them) emitted spikes. Second, through simple cross correlation, they could reconstruct a whole network of neurons, in which one neuron would excite or inhibit another with a time lag of about 2 milliseconds. (That's close to what Pribram told me to expect with standard synapses.) Third -- they tried to show how the learning in these experiments would CHANGE those synapse strengths, but could noit disentangle when correlations changed due to learning and when it was just due to changing conditions and nonlinearity (which were clearly present).
Having spent years leading nsf review panels and observing policy in many areas, I have seen a lack of real strategic thinking in most areas. Here I can imagine what many would say: "Oh, this project was a failure. In real time-series data, it has yet to be proven that learning exists in the cerebral cortex at all. Since it is unproven, it would be wrong to do or perform any more research on this general topic." Likewise, it is speculative to assume humans will exist at all for living, so should we simply rule them out? If mice thought that way, they could never even cross a field to get food, food fear of foxes. Life is a game of probabilities, not only for mice but for us, and rational strategic thinking reflects that. But also -- it is not really speculative to say that learning occurs in the cerebral cortex.
Also.. it is fascinating how the "spikes" in this database are so very different from those in the simple binary pulse models. Their attributes, in complex fet and spk files, look more like the "volleys" I have assumed in my own very different neutral network designs. But.. best I not say much until I do more work, and think about the larger context. Of course, after I read the 2008 paper, I followed up by reading more of the more recent buzsaki papers, which Walter Freeman recommended very highly before his death.
No comments:
Post a Comment