I thought I’d get away from critiques and reviews and serious stuff like that for a week and talk about a cool (or scary) development in AI research. I won’t talk about the details, so don’t get scared off yet. This will be more of a high level history of what happened. Many of my readers are probably unaware this even exists.

Let’s start with the basics. Go is arguably the oldest game in existence. And despite appearances, it’s one of the simplest. Each player takes a turn placing a stone on the intersections of a 19×19 board. If you surround a stone or group of stones of your opponent, you capture them (remove them from the board). If you completely surround other intersections, that counts as your “territory.”

The game ends when both sides pass (no more moves can be made to capture or surround territory). The side that has more territory + captures wins. There’s no memorization of how pieces move. There’s no rules to learn (except ko, which basically says you can’t do an infinite loop causing the game to never end). It’s really that simple.

And despite the simplicity, humans have continued to get better and produce more and more advanced theory about the game for over 2,500 years.

Let’s compare Go to Chess for a moment, because most people in the West think of Chess as the gold standard of strategy games. One could study chess for a whole lifetime and still pale in comparison to the top Grand Masters. When Deep Blue beat Kasparov in 1997, it felt like a blow to humanity.

If you’re at all in touch with the Chess world, you will have succumb to the computer overlords by now. We can measure the time since Deep Blue’s victory in decades. The AI have improved so much since then that it is commonly accepted across the whole community that a human will never be able to win against a machine at Chess ever again.

A few years ago, we could at least have said, “But wait, there’s still Go.” To someone who doesn’t have much experience with Go, it might be surprising to learn that computers weren’t even close to winning against a human a few years ago.

Here’s the rough idea why. Chess can be won by pure computation of future moves. There is no doubt that humans use pattern recognition and positional judgment and basic principles when playing, but none of that stands a chance against a machine that just reads out every single combination of the next 20 moves and then picks the best one.

Go, on the other hand, has pattern recognition as a core element of the strategy. One might try to argue that this is only because the calculations are so large, no human could ever do them. Once we have powerful enough computers, a computer could win by pure forward calculation.

As far as I understand it, this is not true. And it was the major problem in making an AI strong enough to win. Even at a theoretical level, having the computer look ahead a dozen moves would generate more combinations than the number of atoms in the known universe. A dozen moves in Chess is half the game. A dozen moves in Go tells you nothing; it wouldn’t even cover a short opening sequence.

Go definitely has local sections of the game where pure “reading ahead” wins you the situation, but there is still the global concept of surrounding the most territory to consider. It’s somewhat hard to describe in words to someone unfamiliar with the game what exactly I mean here.

Notice how on the right the black stones sort of surround that area. That could quickly turn into territory by fully surrounding it. So how do you get an AI to understand this loose, vague surrounding of an area? One could even imagine much, much looser and vaguer surrounding as well. Humans can instantly see it, but machines cannot and no amount of a calculating further sequences of moves will help.

For years, every winter break from college, I’d go home and watch famous and not-so-famous people easily win matches against the top AI. Even as late as 2014, it wasn’t clear to me that I’d ever see a computer beat a human. The problem was that intractable.

Along came Google. They used a machine learning technique called “Deep Learning” to teach an AI to develop these intuitions. The result was the AlphaGo AI. In March 2016, AlphaGo beat Lee Sedol, arguably the top Go player in the world. It was a five game sequence, and AlphaGo won 4-1. This gave humanity some hope that the top players could still manage a match here and there (unlike in Chess).

But then the AI was put on an online Go server secretly under the name “Master.” It has since played pretty much every single top pro in the world. It has won every single game with a record around 60-0. It is now believed that humans will never win against it, just like in Chess.

More theory has been developed about Go than any other game. We’ve had 2,500 years of study. We thought we had figured out sound basic principles and opening theory. AlphaGo has shaken this up. It will often play moves that look bad to a trained eye, but we’re coming to see that many of the basics we once thought of as optimal are not.

It’s sort of disturbing to realize how quickly the machine learned the history of human development and then went on to innovate it’s own superior strategies. It will be interesting to see if humans can adapt to these new strategies the AI has invented.

# The Ethics of True Knowledge

This post will probably be a mess. I listen to lots of podcasts while running and exercising. There was a strange confluence of topics that seemed to hit all at once from several unrelated places. Sam Harris interviewed Neil deGrasse Tyson, and they talked a little about recognizing alien intelligence and the rabbit hole of postmodernist interpretations of knowledge (more on this later). Daniel Kaufman talked with Massimo Pigliucci about philosophy of math.

We’ll start with a fundamental fact that must be acknowledged: we’ve actually figured some things out. In other words, knowledge is possible. Maybe there are some really, really, really minor details that aren’t quite right, but the fact that you are reading this blog post on a fancy computer is proof that we aren’t just wandering aimlessly in the dark when it comes to the circuitry of a computer. Science has succeeded in many places, and it remains the only reliable way to generate knowledge at this point in human history.

Skepticism is the backbone of science, but there is a postmodernist rabbit hole one can get sucked into by taking it too far. I won’t make the standard rebuttals to radical skepticism, but instead I’ll make an appeal to ethics. I’ve written about this many times, two of which are here and here. It is basically a variation on Clifford’s paper The Ethics of Belief.

The short form is that good people will do good things if they have good information, but good people will often do bad things unintentionally if they have bad information. Thus it is an ethical imperative to always strive for truth and knowledge.

I’ll illuminate what I mean with an example. The anti-vaccine people have their hearts in the right place. They don’t intend to cause harm. They actually think that vaccines are harmful, so it is the bad information causing them act unethically. I picked this example, because it exemplifies the main problem I wanted to get to.

It is actually very difficult to criticize their arguments in general terms. They are skeptical of the science for reasons that are usually good. They claim big corporations stand to lose a lot of money, so they are covering up the truth. Typically, this is one of the times it is good to question the science, because there are actual examples where money has led to bad science in the past. Since I already mentioned Neil deGrasse Tyson, I’ll quote him for how to think about this.

“A skeptic will question claims, then embrace the evidence. A denier will question claims, then deny the evidence.”

This type of thing can be scary when we, as non-experts, still have to figure out what is true or risk unintentional harm in less clear-cut examples. No one has time to examine all of the evidence for every issue to figure out what to embrace. So we have to rely on experts to tell us what the evidence says. But then the skeptic chimes in and says, but an appeal to authority is a logical fallacy and those experts are paid by people that cause a conflict of interest.

Ah! What is one to do? My answer is to go back to our starting point. Science actually works for discovering knowledge. Deferring to scientific consensus on issues is the ethically responsible thing to do. If they are wrong, it is almost certainly going to be an expert within the field that finds the errors and corrects them. It is highly unlikely that some Hollywood actor has discovered a giant conspiracy and also has the time and training to debunk the scientific papers that go against them.

Science has been wrong; anything is possible, but one must go with what is probable.

I said this post would be a mess and brought up philosophy of math at the start, so how does that have anything to do with what I just wrote? Maybe nothing, but it’s connected in my mind in a vague way.

Some people think mathematical objects are inherent in nature. They “actually exist” in some sense. This is called Platonism. Other people think math is just an arbitrary game where we manipulate symbols according to rules we’ve made up. I tend to take the embodied mind philosophy of math as developed by Lakoff and Nunez.

They claim that mathematics itself is purely a construct of our embodied minds, but it isn’t an “arbitrary” set of rules like chess. We’ve struck upon axioms (Peano or otherwise) and logic that correspond to how we perceive the world. This is why it is useful in the real world.

To put it more bluntly: Aliens, whose embodied experience of the world might be entirely different, might strike upon an entirely different mathematics that we might not even recognize as such but be equally effective at describing the world as they perceive it. Therefore, math is not mind independent or even universal among all intelligent minds, but is still useful.

To tie this back to the original point, I was wondering if we would even recognize aliens as intelligent if their way of expressing it was so different from our own that their math couldn’t even be recognized as such to us. Would they be able to express true knowledge that was inaccessible to us? What does this mean in relation to the ethics of belief?

Anyway, I’m thinking about making this a series on the blog. Maybe I’ll call it RRR: Random Running Ramblings, where I post random questions that occur to me while listening to something while running.

# Do We Perceive Reality As It Is?

Recently I had to do a ten hour drive, so I was listening to a whole bunch of stuff. One of the weirder ones was a conversation with Donald Hoffman at UC Irvine. The discussion revolved around what reality is and whether we can know it.

This is a well-known, old philosophical question. I’ve also discussed it in several forms on the blog before (see embodied mind and cognitive biases). What interested me was how clear of a metaphor Hoffman described to get this idea across.

Consider the computer (or phone or whatever) on which you are reading this. There are probably icons that help you run programs. This interface with the computer is basically never confused with the reality of what the computer is doing.

If you have a folder icon, you secretly know that the information of that folder is a bunch of 0’s and 1’s encoded by low/high voltage transistors, etc. But it would be paralyzing to think like this all the time. This interface we use completely alters our perception of the reality of the computer in a way that makes it functional.

So that’s the analogy. We basically have no idea what reality is. All we get is the way our brain presents reality to us. And like the computer, it is almost certainly presenting us with something that makes the world a functional place for us rather than presenting every detail of reality “as it is” (whatever that means).

This is abundantly clear when you think about our visual spectrum. We cannot see in the infrared part of the light spectrum. That information is out there and part of reality, but it isn’t functionally necessary for us to see it. The bigger jump is that maybe it isn’t only our body filtering out excess information, but that our perceptions are some sort of interface that has nothing to do with reality.

The evolutionary explanation for this is painfully obvious with a little thought. If a mutation occurs that causes us to perceive the world in some “false” way, but that false perception increases our chance of surviving, it will stay.

He went into some suspicious sounding math, claiming he had a theorem that there is a probability 0 chance that our perception of reality is true. The specifics don’t matter, because anyone with a basic understanding of evolution should immediately admit that it is exceedingly unlikely that every trait conducive to survival also gives us a true perception of reality.

On another note, this topic actually came up in one of my first posts ever on the blog. I used to debate whether or not aliens would have the same math as us. It seems the further you go in math, the most offensive people find the view that aliens may have a different math. For some reason, people are really tied to the idea that math is some universal that exists out in a mysterious Platonic universe.

In light of the above description, doesn’t it seem very likely that an alien species with a different evolved brain would vastly deviate in their perception of the world? We are so trapped in our bodies as humans, we find it hard to even entertain what a different experience would be.

If you know about the foundations of math, then you know everything is built from a set of axioms. Our choice of these axioms is based on how we experience the world. Therefore, it is reasonable to assume their mathematics would be different.

# Validity in Interpretation Chapter 2

I thought my last post would be the only one on this topic. Then I realized, if I’m reading the book, why not just take notes on it and post them? My goal is to give a good enough summary so that later, if I want to recall what was in a particular chapter, I can figure it out from this. I’ll also try to be explicit where I’m adding my own thoughts (which should be minimal).

Chapter 2 begins with a positive argument for developing a notion of validity in interpretation. The first chapter addressed common objections. This one begins by pointing out that one major purpose of art is to expand your mind with other people’s thoughts and actions and to feel what others have felt. If we take the usual modern approach that anything can mean anything to anyone, then you have a Rorschach inkblot test and will only encounter yourself.

Chapter 2 is all about showing that the author’s intended verbal meaning provides a viable principle for measuring the validity of an interpretation. Hirsch states that such a principle must be determinate, reproducible, and able to deal with the problem of implication. The subsections of the chapter deal with each of these.

First, we need to know what is meant by verbal meaning. Hirsch says, “Verbal meaning is whatever someone has willed to convey by a particular sequence of linguistic signs and which can be conveyed (shared) by means of those linguistic signs.” He explains how verbal meaning is variable and context dependent, but not indeterminate. For example, we should be careful of “the Humpty-Dumpty effect” (based on the scene in Through the Looking Glass where Humpty-Dumpty insists his name means the shape that he has). We can’t just say words at random and expect them to mean something because we claim they do.

The next section is on whether we can actually reproduce the intended meaning. First, he admits that interpreters can misunderstand an intended meaning, but the fact that mistakes happen does not invalidate the idea of reproducibility. It is a logical fallacy to argue that something is impossible in theory by pointing to a specific example. The person that objects has to demonstrate that such misunderstandings always occur. Hirsch points out that we will never know such a thing, and hence it cannot be used as a valid objection.

The most common objection to reproducibility comes from the empirico-psychologistic notion of perception which says that we never encounter anything in real life. We merely encounter our perception of things. For example: That isn’t a table you see, but your perception of the table. This type of objection disappears if you are careful in distinguishing meaning vs significance (a central theme of the whole book). Verbal meaning is the thing itself (the table or the author’s intended meaning), and significance is your relationship to the thing (your perception of the table or your particular reading of the text).

There is also a radical historical skepticist objection to reproducibility which says that we can never understand the writings of a different time period, because we do not have the linguistic, cultural, and so on perspective for a proper understanding of the verbal meaning. This radical form of criticism should not be confused with the healthy version (which Hirsch admits is true) that an interpreter will always encounter some difficulties if the culture is too far removed.

Personal note: The odd thing about the radical skeptic view is how backwards it has often proved to be in reality. How many times have we seen a poet or writer be completely misinterpreted and torn down by their contemporaries only to be better understood by later generations? This is especially true of really old texts in which we have a good general understanding of the culture and etymology of words that people living at that time never could have known.

The next concept Hirsch moves to is determinacy. Determinacy is necessary to share meaning, because something without boundaries would have no identity to share with someone. He upgrades his terminology, “Now verbal meaning can be defined more particularly as a willed type which an author expresses by linguistic symbols and which can be understood by another through those symbols.”

A type is defined to be something that has boundary and can be fully understood through one instance, but can always be represented by more than one instance. My example: The concept of snow is a type, because the word “snow” is an instance that has that verbal meaning. Another, different instance would be “that white stuff that falls from the sky in winter.” Both instances have the same verbal meaning, and the verbal meaning can be recovered in full from any instance.

Personal note: I think of this in terms of computer science. You can make a class in object oriented programming, and you can make instances of that class. One instance may not have enough information to recover what the class contains, so this is not a type…it is a class (Hirsch uses the same terms but isn’t thinking of computer science).

The next section is about the difference between unconscious and symptomatic meanings. The main point is that verbal meaning may contain things that come subconsciously. Verbal meaning does not contain things are unintended due to symptoms of something else. The example is a boy who has a tell when he lies. The tell is symptomatic of not wanting to lie, but it is not part of the verbal meaning of the lie. The tell changes our interpretation to something that was not verbally intended by the boy, and that’s why we should not include it.

Personal note: Something makes me feel funny about this section. Even though the example makes it clear why we do not want symptomatic meaning to influence our interpretation, nothing will be that clear in pure writing. I’m not sure in practice it is ever possible to determine the difference between symptomatic and unconscious meaning. Suppose a story has sexist undertones the author is not aware of. Do we consider this symptomatic of the author’s sexism or is it an unconscious intended part of the story? Since we are allowed to include symptomatic meaning in the significance of the work, maybe this doesn’t matter very much.

There is then another section on the difference between meaning and subject matter which we discussed last time. The last section is about implication. (Personal note:) This seems a thorny issue given little space, so I hope it comes up again later. The main point is that implication is a learned convention, because it relies on the reader’s past experience with a given shared type. He explains it by analogy: an implication belongs to a meaning as a trait belongs to a type.

That ends the chapter. The next chapter is about genre and context which should be interesting.

# Consciousness Explained 1

Sorry this took so long, but I kept reading further in hopes of getting to something more meaty to talk about before my first post on this subject. I’m through 3 chapters and it has essentially only been basic definitions and a few thought experiments.

My first comment is that I always forget that I’m not actually interested in philosophy of consciousness. What I associate to phil. con. (my abbreviation from here on out) is not what the typical philosopher associates to it. I tend to view all of philosophy through the lens of philosophy of language. So I usually start form some sort of premise about how language is more primary and fundamental than consciousness. Now lets try to figure out how it is that language brings about consciousness. Of course, this is not at all how the book goes about it. In fact, I think the opposite assumption is implicit here. So needless to say, the book isn’t as interesting to me as I would have hoped.

The one argument that was presented right at the beginning that I had heard before but forgotten that was pretty interesting was about debunking the “brains in a vat” idea. Basically this goes back to Descartes who wanted to know if there was any way we could tell if we actually existed or if our brains were bodiless in a vat somewhere and scientists were just stimulating certain neurons to make us think we were people (well, so Descartes’ description was a little different, but this is the modern Matrix-esque interp). Essentially we don’t have to go through the trouble that Descartes went through to debunk this possibility. The standard sort of pragmatic argument is that it just isn’t possible for any reasonable interpretation of the term “possible”. The amount of computer power needed to do this would encounter a combinatorial explosion for even the simplest experience of the world. Thus, it is not possible we are being tricked (so before a torrent of arguments fill my replies, I watered it down, try to fill in the details yourself before arguing).

Other than that the only sort of important terms that might show up in later posts have to do with phenomenology. All of chapter 3 is essentially devoted to this. Essentially it is a method of philosophy of consciousness developed by Husserl that tried to remove subjectivity. I really don’t want to go into this much, since I feel like it won’t play much of a role later and it is giving me flashbacks of my 20th century philosophy class when we had long tedious arguments about the method of “bracketing”. Overall, what you should know is that it played a huge role in influencing major philosophers and schools of thought on philosophy, but in general is highly criticized and probably has been overtaken by neuroscience studies and interpreting them.

My one complaint so far is that the results of thought experiments (which play a major role in this book) are very skewed by leading questions. I don’t doubt that the visualization of X was harder than Y, but coming to that conclusion before asking, “Wasn’t visualization of X harder than Y?” would have been more convincing for your argument. I’m not sure if any were that bad, and I should look a specific one up, but I don’t really feel like it now.

My guess is that the next chapter is on a rejection of Husserl’s phenomenology, and then hopefully it will get into some of the crazy things our brain’s do from a neuroscience perspective. That could make things more interesting.

# Cat’s Eyes

I’ve been running a very unscientific experiment for awhile now. It has produced some very interesting results that I can’t seem to explain. So every time my cat comes up to me and wants to jump on my lap (several times a day for at least a month I’ve been running this), I try something new. I either look near her, or don’t look at her at all, or look at her eyes, or some variation. The point was that I noticed she seemed to only make the jump when after we made eye contact.

Recently I’ve tried to be very specific about the different between looking near or at a different part of her body vs making eye contact. It may be my imagination as hardcore scientists would probably tell me, but I’m am quite convinced that she understands the difference between eye contact and non-eye contact.

Here is why I don’t understand this. Even if we make a huge leap and grant full out consciousness to cats (probably an overkill assumption), I still can’t explain it. Humans understand eye contact because they are aware that that is where sight comes from. I can see of no scenario in which a cat can come to the realization that eyes are where sight and recognition come from.

The only explanation I can come up with is that she is getting up there in cat years, and so has developed a conditioned response through years of experience that eye contact means that I am paying attention to her. As a good Pavlovian, we could say that she doesn’t have to have any idea what the cause is or why it works, it is merely a conditioned reaction.

Anyone else that might know something about animal behavior want to take a crack at this one?

# Penrose Objective Reduction

Moving on finally. I have recently come into contact with an “interpretation” of quantum mechanics that I was unfamiliar with (I think). I at least never looked it up in depth if I have heard of it. It is called Penrose Objective Reduction (or POR for short). It is very different than the standard theories in that it doesn’t ignore the collapse of the wavefunction (many worlds or decoherence). Instead it postulates something possibly more fantastic. Let’s start more towards the beginning before going there, though.

Take $E=\frac{\hbar}{t}$ and interpret it. E is the degree of spacetime separation of the superpositioned particle and t is the time until POR occurs. This shows that small superpositions (i.e. things that are almost in a determined state) will take a long time to collapse objectively. This intuitively makes sense, since there isn’t really a “need” for it to take a determined position if it is undetermined at close to plank-length distance. It doesn’t contradict our world view. On the other hand, extremely large objects (say Schrodinger’s cat) will objectively collapse to a single position extremely quickly. This eliminates the “paradox” of Schrodinger’s cat.

But what exactly is POR? It is “objective” collapse of the wavefunction in that it doesn’t require an “observer” as other theories claim. Wavefunctions will naturally collapse. Will this collapse go to a random state (in which we know the probabilities, but still random)? Penrose says no. He claims that there is information embedded fundamentally in spacetime. He makes an even more extraordinary claim that it is “Platonic” in that it is pure mathematical truth, aesthetic, and ethical. Since I have spent weeks rejecting Platonistic views, I feel I should offer an alternative based on this method.

It is known that “empty” space (e.g. the mass gap, actually I can’t find it now, I was going to link it, I’ll keep looking) has enormous stored energy. Many people interpret this as where collective consciousness lives. Instead of some objective random collapse of the wavefunction, or some ethical godlike decision as to what it should collapse to, it seems as if we are missing the power of our own minds. Maybe a more karmic collapse. The thoughts and energy we put into the world gets stored in the area and influences the POR to give us back what we put out.

I didn’t get into a lot of aspects of POR, and I was just making up that last part on the spot, so it wasn’t very scientific or rigorous as to how it could work. Just some thoughts for today.