Life, the Universe, and Everything

An interesting consequence of the Second Law of Thermodynamics: Suppose that you have a hollow, uniform metal container. Heat it to around 700 C, until it glows a dull red. Any glass lens inside of the container would, obviously, bend some of the light emitted from the inside walls of the container, redirecting it so that it hits a different spot on the metal.

However, notice that if the lens refocuses more of the light onto a particular area, it will be absorbing more light than it emits, and so it will heat up. Other areas will, of course, cool down due to Conservation of Energy. This creates a temperature differential which can be exploited to extract work from the system, even though it was initially at equilibrium.

Thus, no possible lens can focus or concentrate the light emitted from the inside of a uniform container, without doing work itself. This should hold true regardless of what material the lens is made out of, the shape of the container, and numerous other parameters.

A post-scarcity economy is a common theme of science fiction; in the future, people reason, technology will have progressed to the point where the cost of producing material goods is either zero or trivial. Obviously, some resources will remain scarce in such a society, particularly resources tied to social status, and things such as historical artifacts that can never be replaced. But in such a society, people can easily acquire all the material ’stuff’ that suits their fancy.

It appears that we may now be approaching this point. There is a ???????? ????? ????????huge glut of used items washing around America, selling at garage sales and internet auction sites like eBay, often for pennies on the dollar. The major expense for a large category of goods, particularly ones which are easily made from raw materials, is now shipping and brand. Books- even books that cost $50 to buy new- are now rarely worth more than $5, once shipping is taken into account (try it and see).

Consumers, prodded by decades of advertising, are still hooked on the idea of buying things new, even for big-ticket, rapidly depreciating items like cars. It is, of course, illegal to sell used items as if they were new, so I predict that this situation will persist, at least in the short term. In the meantime, you can save a huge pile of money by buying things from the general glut at a tiny fraction of retail price.

Back in the days of my wild and reckless youth, quantum mechanics was presented as a classical, political dichotomy: superposition interpretation vs. many-worlds interpretation. Both of them seemed, at the time, to be nonsensical. The idea that conscious knowledge is related to physics experiments was obvious hokum. And if there were multiple worlds, then where did the split branches go? There was no room for septillions of split worlds in Minkowski space (or, later on, in 4D Riemannian space). So I went to work making up my own interpretation, using the word “states” instead of “configurations” (which I hadn’t heard of), and then just ignoring the world-branching process.

The reality, of course, is quite different, and I dearly wish that this was taught in lower-level physics. The layer of reality that we see- essentially Minkowski space- is not the lowest layer of Russian dolls. The lowest layer that we’ve identified is configuration space, which is basically the Cartesian product over every linearly independent (pure) quantum state of every particle. Configuration space is more fundamental than the Riemannian manifolds of GR; everything that is real, is real within configuration space. The laws of QM always operate on amplitude distributions over configuration space, not on 4D space. Three dimensions of space and one of time are just an interesting side effect of the underlying reality.

Every “world”, as people call it, is really a blob of amplitude within this configuration space. This blob, as you move forward along the time axis, will gradually throw off additional blobs in all directions. The splitting process, so far as we know, is continuous in roughly the same manner as classical heat diffusion; there are no discrete “jumps” where worlds suddenly split off. The fact that the blobs are blobs, rather than points, means that all matter is fuzzy (spread out in space) to a small degree. However, such fuzziness rapidly disappears in any given worldline, due to the exponential decay in amplitude in between blobs.

Proposals to “communicate” between different “worlds”, therefore, are fundamentally misguided and rely on a misunderstanding of the situation. If we had a good view on the situation, we would not see adjacent “parallel universes” floating in space, but the gradual diffusion of matter into a thin mist of probability as quantum splits added up. The real situation is even worse than that; we can’t see the mist of probabilities, except to a very limited extent, so we can never directly detect the quantum ghosts of different objects in all of their possible positions.

We’re currently collecting information on animal, non-primate intelligence, so if anyone has any useful references, please send me an email. Thanks for your help!

Everyone has their own pet strategy for making money on Wall Street, but so far, I haven’t seen any systematic analysis of which ones work. People, as often as not, just guess at random and hope for the best. To help eliminate this gaping void in our knowledge, I propose a competition in trading strategies. Proposed rules:

- Any publicly traded asset is fair game. This includes stocks, bonds, commodities, derivatives, etc.

- You cannot borrow more money to invest; you’re stuck with what you have at the start. Leverage is cheating, and it’s also dangerous (margin calls were a primary cause of the Great Depression).

- The winning strategy is the one that makes the most money from a starting asset base, over a set time period, using historical data.

- There should be different categories for different starting asset bases ($100K, $1M, $10M, etc.) and different time scales (2 years, 5 years, etc.)

- Strategies may not use hindsight. For this reason, naming specific companies or assets (eg., GE stock or IBM bonds) is not allowed. Absolute times (eg., buy tech stock in 1996 and sell in 2000) are also not allowed.

- Strategies must be sane enough to be describable in a dozen or so pages of text, which must be understandable by anyone reasonably well versed in economics. This does not include citations, references to prior work, justification, the research process used, etc., just the details of how to make money.

I will personally fund such a competition with $250, so long as the rules are reasonably similar. Any other takers, or rules I should include?

EDIT: I am aware that many traders put a lot of emphasis on “instinct” or having a “feel for the market”. Large-scale markets developed too recently to allow for the evolution of specialized adaptations, so our gut feelings are more than likely to be wildly inaccurate.

“The public’s out there throwing darts at a board, kid, I don’t throw darts at a board; I bet on sure things.” - Gordon Gekko

Hedge funds, by and large, are an obscenely profitable industry; worldwide, hedge funds now manage around US $3 trillion in capital. A hedge fund can easily make as much money in a single quarter as a traditional investment would in an entire year. For now, hedge funds are unavailable to those of us with less than US $1 million in liquid assets, due to SEC regulations; sooner or later, someone is going to figure out how to allow the middle class to invest in these funds, but that’s another topic.

Hedge funds make most of their money, in general, through finding imbalances in the market and then exploiting those imbalances. Finding market imbalances requires some work, but it’s not impossible or even exceptionally difficult. If no market imbalances existed, every possible publicly observable variable would have no correlation with the future price of an investment, which is obviously absurd; conversely, every correlation is a potentially exploitable imbalance.

The main reason why most investors don’t beat the market is the failure to look for such imbalances, not the failure to find them. Most people, and even most brokers, pick investments by using a hodgepodge of inconsistent heuristics which are never written down and never tested. Human instinct can be shown to be horrendously wrong on a large number of objective tests, so this is not surprising for the field of cognitive psychology, but few investors have seriously studied the underlying concepts.

Once an imbalance is identified, it can be exploited through the other primary instrument of funds: massive leverage. The public is now at least somewhat familiar with the concept of buying on margin, as many middle-class Americans own stock, and margin calls were identified as a primary cause of the Depression. Leverage in general is less well known, but it amounts to the same thing: borrowing many dollars in loans for each dollar in capital, investing it, and then siphoning off the difference between investment return and interest rate. Leverage can increase returns, but it also increases risk, as you can wind up losing more money than you started with. Hedge funds try to decrease this risk by diversifying into many different investments, but it isn’t a perfect system, and many do go bust.

Suppose that you want to rank the members of a set X = {x1, x2, x3, x4} relative to each other. You can, say, assign x1 = 40, x2 = 30, x3 = 20, and x4 = 10 by some measurement metric Y. Or, you can assign x1 = 80, x2 = 60, x3 = 40, and x4 = 20. Or you can assign x1 = 4, x2 = 3….

No matter which one you choose, as long as the metrics are all multiples of each other, the elements of the set are still in the same position relative to each other (which is all we care about, by assumption). In math-speak, the measurement metric Y is invariant under a positive linear map, as the important properties don’t change when you throw in arbitrary constants. It’s generally inconvenient to use huge constants, so people usually renormalize Y to some neat number, such as 1 or 100.

Humans, however, aren’t born with this kind of math built in, and so you can pull tricks by failing to renormalize. Very few people will notice if your percentages add up to 105%. Without quantitative analysis, it’s even worse, as you can use qualifiers (”important”, “big”, “profitable”, “useful”, etc.) to stop people from renormalizing without setting off alarm bells. This is a very, very old trick; the best way to counter it, generally, is to quantify the metric and then make sure it’s renormalized during every step. Some cases where this comes in handy:

- Probability. For reasons of mathematical sanity, probabilities are always renormalized to 1, although there are some cases where you can get away with other numbers (eg, Bayes’ Theorem still gives the same result if you multiply the prior by 100). Nevertheless, quacks worldwide still fail to renormalize by claiming that they predicted every result with high accuracy.

- Utility. Utility functions are invariant under positive linear maps, and they can generally be renormalized to whatever you like (finite numbers are usually necessary, as explained here). Making your life wonderful by assigning a high utility to everything is the same mistake as an amateur economist failing to account for opportunity costs.

- Priority. If you generalize priority to quantity of resources allocated, rather than using a simple preference ranking, it should be invariant. I am still stunned by how many managers insist that every task is extremely important; this corresponds to a crisis of Bayesian affirmation.

- Grades of all varieties. Failure to renormalize is well known as grade inflation. Note that renormalization is not bell curve grading; it corresponds to, say, the fungibility of x/4.0 GPAs and x/100 averages.

- Competitions of all varieties. Professional sports are generally immune to renormalization failures, as there can obviously be only ten teams in the top ten. However, this phenomenon is rampant on school sports teams, thanks to the “self esteem” culture.

The problem of recruiting FAI researchers has been discussed in greater detail elsewhere, but currently, we don’t even have a clear idea of how easy/difficult recruiting will be. Therefore, I suggest that SIAI or a team of volunteers draw up a dozen or so questions, with known, well-defined answers, such that being able to solve the questions is an obvious pre-requisite to doing useful FAI work. Such a survey can be distributed extremely easily, at little-to-no cost, using already-existing networks such as SL4.

The point of this should not be to judge the hireability of any one specific individual, which also depends on ethics and ability to work in teams, among many other factors. It should, however, be useful in determining how widely distributed background knowledge is; hopefully, it should also tell us which fields are not well-covered by existing literature and which may need further exposition.

Rationality is the process of deriving true beliefs from observations. We know of a number of useful heuristics which can help us derive true beliefs, but how are these rules generated? Humanity didn’t start off knowing that postulating spirits as an explanation was, in general, a bad way of finding things out. The two main general optimization techniques- trial-and-error and natural selection- both require an enormous number of trials (O(2^n) and O(n), respectively). It took us thousands of years, and an enormous collaborative effort, to stumble upon something as low-complexity as the scientific method. Trying to derive anything more complicated by just guessing and checking is probably hopeless.

Eliezer’s OB posts have touched upon this problem before, but so far as I am aware, there are no published meta-rationality heuristics of any significant complexity. To name a simple example of such a heuristic, studying lots of cognitive science will give you more knowledge as to how the brain can fail, which leads to the discovery of ways to correct such failures. As another example, doing a stack trace of the brain whenever a wrong belief is corrected will, if successful, identify the specific pattern of thought that caused the wrong belief. I’m sure there are more to be found, somewhere in idea space.

As of several days ago, Google has placed a malware alert on this blog. There is no malware of any description on Accelerating Future, but hidden spam links have been inserted into my HTML from an unknown source. I will remove these as quickly as possible; my apologies for the inconvenience.