It turns out this was a moon halo. The page says that the phenomenon is “familiar,” but I’ve never seen it in my life, and had I not looked up for no reason, I would have missed this one, too! By my hand measurement, it spanned 45° in diameter, which is a pretty big portion of the sky. Jupiter was also visible within the ring of the halo. Quite amazing.

As you can see it carries on just fine and can fool a casual observer. But the longer you carry on a conversation with it the more apparent that Cleverbot is frustrating to talk to, not so much that it isn’t human — after all, all of the responses are taken from human sources. If it weren’t so good at emulating a human from which you expected more, you wouldn’t be frustrated.

Cleverbot is frustrating in two ways:

• it isn’t interesting
• it doesn’t make sense

In other words, it lacks deepness, like a shallow human. Why?

Some say that Cleverbot isn’t an AI because it’s just humans talking to humans via the intermediary of a selection algorithm. One is tempted to be dismissive like this but it’s too convenient. It places too much weight on the role of the corpus and too little on that of learning and interaction. Cleverbot clearly captured some aspect of the human intellectual process that raised it to the level of a shallow conversation partner. What it neglected though will be enlightening to the question of what is deepness.

The most obvious way in which Cleverbot is shallow is in novelty. This is a result of mimicry of others. Mimicry is of course fundamentally human because we learn by mimicking. It doesn’t preclude novelty of some sort. In fact, although it isn’t clear at what level Cleverbot replicates from the corpus, whether in phrases or entire sentences, we are forced to accept that some novelty is injected into the selection process, as responses don’t come out the same every time. However it’s not enough to make Cleverbot consistently interesting, probably because Cleverbot doesn’t synthesize new ideas, merely passing on received wisdoms at the appropriate moments. It seems deep novelty requires more atomicity, and more computation.

The more subtle way in which Cleverbot is shallow is in contextual memory. Cleverbot, probably also for computational reasons, just makes remarks in response to the most recent context. As a result, it changes topics frequently, and doesn’t pursue a line of inquiry to its satisfactory conclusion because it doesn’t know what a line of inquiry is. It can be described even as evasive or flighty. While each response usually makes sense, Cleverbot doesn’t make sense over the span of a conversation, which requires more persistence.

It seems deepness is related to computational resources. And if we believe more intelligent machines — those that exhibit deepness and frustrate us less — to be more “human,” then we must expect a high level of computational competence as a human trait. Deep conversation is a measure of that. Disturbingly, the kind of conversation that Cleverbot can make is rather the more commonplace case among humans, yet it borders on meaningless drivel because it produces nothing new. Even more disturbingly, this is despite that Cleverbot can be quite clever sometimes, being stimulated to produce some of its store of wisecracks after elicitation by a motivated human counterpart who steered it to them. It shows that it doesn’t take much to carry on one half of a conversation, and deceptively so especially if one of the two is playing the role of an intelligent “human.”

Maybe being “human” isn’t that hard. Maybe the binary identification of machine/human should be considered a necessary quantification artifact, and the Turing test really measures a gradation of deepness instead (it is a probabilistic test after all, and of course subjective). If so, let’s run the Turing test on people in a double-blind way.

However, there is a way to figure this matter of structure from first principles (and perhaps generate a more unique style as a result), albeit with the tradeoff that you cannot be quick, you must be careful.

The first principle for aesthetics is that the character must stand … this is something my old man told me, actually, so I didn’t figure this out myself, but it is very true. If you hold up the piece of paper and look at the strokes as struts of a building, it must look like the character is architecturally sound, i.e. reasonably symmetric if need be, balanced in weight so will not tip over, is not poorly supported with too small a bottom and too big a top, etc. This isn’t too difficult if the character is mechanically drawn, but the trick is to do it even with asymmetric calligraphic strokes and multi-part characters with asymmetric radicals and caps.

The second principle for aesthetics is about spacing, and this is much like optimal typography and typesetting. The strokes should be spread out evenly so that where they appear parallel, they appear to have nearly identical spacing as other such spaces. Otherwise there will be ugly bunching and voids. This is very difficult because the strokes are written in order so there is a pre-commitment issue. Once you commit to a particular stroke, it also commits the spacing requirements for the rest of the character. So one slightly off stroke and you are screwed. This is more a problem for large writing, since bigger mistakes are possible.

Then is the issue of multiple character layout. This wouldn’t be so much of an issue if all characters were the same shape and complexity, but they are not. Some are extremely sparse, and some are very dense. Some are tall and some are fat. They all have to be laid out on paper to look like they take up the same space and also evenly spaced from each other. There is also the compromise of making inter-stroke space appear similar in multiple characters. So one needs to deal with some visual artifacts and vision tricks. As a result, the characters will not all be the same size and will not be spaced evenly, so this is a very tricky thing to get right. You can have perfectly written individual characters but still a terrible collection.

And finally here is a side point: people say Simplified characters are uglier than Traditional characters for calligraphy. In fact this cannot be true. What happens is Simplified characters are sparser and sparser characters writ large are the most difficult to get correctly (not to mention there are no classic master’s character books to trace in Simplified). They are ugly only because (or to the extent that) they are not written well. The bastion of poor practioners (like me) is in small dense characters that distract from scrutiny and generally look pretty good no matter how you write them.

In the simplest of examples, you pay some entry fee to play the game, $1 is put in a pot by a counterparty, then a coin is repeatedly flipped and the pot is doubled on every coin flip by the counterparty, until “tail” comes up. You receive the money in the pot. The expected gain of this game is infinite, regardless of the initial entry fee. So it would seem that one should always play the game, regardless of the amount demanded as entry fee. But, as the article points out, “few of us would pay even $25 to enter such a game.”

(This seems to be one of many variations of the paradox.) The explanations given in the link to resolve the paradox aren’t satisfactory. “One can’t buy what isn’t sold” can only be considered a joke, while “expected utility” is somewhat plausible, but doesn’t strike at the central issue, because it can be circumvented with an equally counter-intuitive paradox fitted to the chosen utility function. In contrast to the Gambler’s ruin paradox, I don’t think that an artificial finite bound on the money supply (in this case, of the counterparty) makes sense as an explanation, but what it reveals as the logarithmic growth of the expected gain against the money supply and the general consequence that imposing some kind of finiteness may explain the paradox, is instructive.

Of course, the only way to get any gain is to actually play the game. If you repeatedly play the game, your gain does eventually go to infinity. So why would you be reluctant to pay even $25 to enter? It must be because those large pay-offs are so infrequent that to make the initial money back would take too long. Suppose the entry fee is \(W\). Suppose you call it a day when you have a positive pay-off. For that to happen in round \(n\), it must be true that

where \(f_i\) is the number of flips before tail comes up in round \(i\).

Let’s call n-tuples \((f_1, f_2, \dots, f_n)\) that satisfy the above by the set \(S_n\). The probability of winning in round \(n\) is then

from which we can surely get the average number of rounds it will take to win the game \(\sum_{i=1}^\infty n p_n\). If this is incredibly large even for modest \(W\), which is likely the case, then that would explain the paradox, since a game that on average takes longer than a lifetime to win would be played by no one.