Lambda the Ultimate

As for Google's pagerank patent... there is prior art to the basic concept. Plus, I don't think most people would consider Pagerank to really be an "artificial intelligence" thing, just a really clever way of ranking web pages, the only downside being that it's trivially gamable.

Could you please provide a better link, your example of prior art came a few years after and references the original PageRank paper.
And how is pagerank trivially gameable? Making a hundred links from a page is no better than making one.

Plus, I don't think most people would consider Pagerank to really be an "artificial intelligence" thing, just a really clever way of ranking web pages, the only downside being that it's trivially gamable.

The point is that this "really clever way of ranking web pages" succeeds in exhibiting behaviors that can appear intelligent, as in the example which Erik gave. Explaining the trick, as Paul did, doesn't make it any less successful. Are you so sure that your own intelligence is anything more than a collection of such "clever ways" of performing particular comprehensions on its input data?

But really, in writing about a "collection of tricks" I was referring to the common criticism that's been aimed at chatbots like Eliza and Alice, and Turing test competitions such as the one for the Loebner Prize, which tend to encourage a focus on such tricks (see e.g. Chatterbots, Tinymuds, And The Turing Test). Usually, "collection of tricks" is used as a criticism to indicate that something is not "real" AI. I was suggesting, not entirely seriously, that perhaps this thinking is backwards — perhaps we just haven't identified the right collection of tricks yet, but if you keep an eye on the patent office, sooner or later they'll all show up — let's call this the million patents approach to AI, by analogy to the million Shakespeare-typing monkeys.

In any case, using relationships between objects to analyze relevance and popularity is not simply a trick. The aspect which is a trick is that the content of the pages containing links isn't really "understood", which is part of what leads to gameability (trivial or otherwise). But in combination with other techniques, it's a perfectly valid means of analysis. And gameability is not necessarily an indication of lack of intelligence — after all, there's a wealth of evidence to show that humans are "trivially gameable" in a multitude of ways. In many cases, our gameability is in fact a direct result of identifiable tricks, processing shortcuts which we use in order to avoid having to think through every decision from scratch.

A nice simple example is herd behavior — following the actions of other people. You wrote that you "don't think most people would consider Pagerank to really be an artificial intelligence thing", revealing a desire to delegate the question of whether Pagerank is an AI feature to a majority vote, rather than developing your own arguments on the subject. Aside from the question of the logical validity of this approach, it is also gameable: if I wanted to convince you that Pagerank was an AI feature, all I would have to do is convince most people that it is. Luckily, most people don't know squat about AI, so that shouldn't be too difficult!

Usually, "collection of tricks" is used as a criticism to indicate that something is not "real" AI.

People used to talk about a "grandmother cell", a neuron that would fire when you saw your grandmother, and then laugh, "haha - obviously the brain doesn't work like that". Recent research shows that the brain *does* work like that: common concepts have small numbers of neurons devoted to them.

Of course, the grandmother cell theory is now called The Halle Berry neuron theory.

Recent research shows that the brain *does* work like that: common concepts have small numbers of neurons devoted to them.

My understanding is that it is slightly more subtle than this (but I'm certainly no expert). As I understand it, in the part of the visual system that is primarily concerned with object recognition, cells get more and more specialised as you move along the pathway away from the eyes, until you reach a point (V4? inferotemporal cortex?) where there are small groups or even single cells that react strongly only to very specific stimuli (particular faces etc). But that's not the same thing as saying that these individual cells are solely responsible for recognising your grandmother. Rather (again, from my naive understanding), some cells are more specific to grandmother-recognition than others; many cells are involved in the processing of all/most stimuli.

While your explanation is largely correct, i think the specific reason for this lies in Eric's previous post (use the cached version). He links to his paper with the anchortext 'typoholic'.
Associating anchortext with the page that is being linked to is a pretty well-known technique in search engine circles. I think you're giving google too much credit regarding nouns/adjectives, though only google really know for sure.

[on edit: i changed the 'previous post' link to a google search, because the LtU archive link i posted is doing odd things...]