Thursday, April 27, 2006

Impetuosity and its unintended side effects

Tuesday I decided to cancel my subscription to "New Scientist". IMHO it's been getting steadily more populist, vacuous and content-free over the last few months. Last week I read a particularly superficial article on Stephen Hawking's latest cosmological thinking and decided to end it.

My last issue arrived today and it was full of quite interesting stuff (oh, the ironies of life!). It contained a special section on 'the science of love', with the usual stuff on oxytocin. But there was a particularly interesting interview with a guy called John Gottman, a mathematician whose views on relationship success apply to more than marriage (article here). I was prompted to buy a couple of his books - marriage and maths - no doubt to be reviewed in good time.

It's Scientific American from here onwards, though!

Wednesday, April 26, 2006

The Loebner prize

Dr. Hugh Loebner set up this competition in 1991 as an application of the Turing test for artificial intelligence. Each year a number of programs compete to appear most human in their response before a panel of judges (the judges are typically psychology professors). The interaction is like instant messaging: each judge types a message to the ‘chatbot’ in a pane on the screen, and back comes a reply.

How convincing is the state-of the art in chatbots? Well, the 2005 winner was a program called Jabberwacky. This program is on the Internet here and anyone can ‘converse’ with it. Here is my attempt (I'm 'user').

Jabberwacky: Do you live in Chicago?
User: I live in England.
Jabberwacky: Is england full of programmers?
User: It's normally full of rain.
Jabberwacky: Well what are you up to?
User: I'm writing a book.
Jabberwacky: About e-mail addresses?
User: You seem obsessed with computer topics!
Jabberwacky: Bots and maze theory in data structures, are our next upgrade in tech.
User: How did you win the Loebner prize?
Jabberwacky: No. But then you didn't either!

It looks like Loebner will be waiting a while to award his $25,000 prize to a system judges are incapable of distinguishing from a human being!

Tuesday, April 25, 2006

Jung ... Jesus ... Does it work? Is it science?

1. The key thing about science is that you could be wrong. I know Popper made this a cardinal principle, but it's also psychologically definitive. Most people believe their opinions and will defend them against evidence - more accurately, pointing to evidence to the contrary is taken as rather juvenile, not polite. Religion is a case in point. Suppose you ask 'could you accept that you could be wrong about this ... and here are some reasons'. Would that be an acceptable dialogue?

Actually I am less concerned about religion than about Jung. Personality type theory (which I have mentioned numerous times in this blog) has explanatory power, but is it science? I don't know whether the adherents would prefer Jung to the evidence, but devastating critiques like the theory's detachment from brain architecture and evolutionary theory suggests that it cannot be the last word. However a Jungian perspective which took all this into account would just be 'scientific psychology' and not very Jungian. So there's the dilemma for the adherents. Seems OK in physics, Isaac ... you're still revered.

2. Dominic Crossan's view of Jesus's historical life and beliefs seems to pull its punches on a critique thereof. Apart from an elliptical remark late on, that if the Roman Empire had not converted, the 'dream might have died in the Galilean hills', he is silent on whether the 'Jesus project' actually had a sociological reality. I take his view to be that Jesus only envisioned on a tactical scale, and had no blueprint for an idealised community on earth. Therefore his radical egalitarian views are always an asymptotic target, and never a sustainable equilibrium state of society. However, Jesus' message is too abbreviated to know what he would have proposed if his approach had really taken hold and succeeded against the opposition of religious and roman authority. It would be an interesting analysis to study however, if someone would work on it. "Wonderful theory, wrong species"? (Wilson).

3. It's been a truism in literature that there are no new plots, only new narratives of plots. Does this mean that there are no new ideas in the human condition? After all, we don't change much as a species from generation to generation. Perhaps the only new ideas are those from science and mathematics, and they stopped being accessible to anyone but specialists a while ago. So here's a point, new ideas do not come from nowhere, they come from an intellectual (+ emotional?) confrontation with an emerging reality. In science, experiments but also theoretical messiness; in art, new conditions of living and reflection, perhaps catalysed by new technologies. When we lack the ability to do new experiments, though, science ultimately loses its anchors and becomes tidying-up mathematics or speculative mathematics. In other scientific areas (psychology again) the new technologies of brain imaging open up new science and turn our paradigms upside down.

0. I think this is where I came in.

Thursday, April 20, 2006

Brilliant explanation of real-world auction theory

This is a book review posted at Amazon.com here.

Paul Klemperer was the principal auction theorist advising the UK Government on the design of the 3G spectrum auction in 2000, which raised $34 billion, the biggest auction in history. In this book he reviews auction design both theoretically and in practice - the two perspectives turn out to be very different.

Chapter 1 is a survey of auction theory. The author begins by identifying four types of auction: the English type of open ascending bids, the Dutch auction using descending bids where the first bidder wins, the first-price sealed-bid auction where the highest bidder wins and pays their own bid price, and the second-price sealed-bid auction (the Vickrey auction) where the highest bidder wins, but pays the amount bid by the second-highest bidder. If the latter seems counter-intuitive, note that in the English auction, the winner also pays an amount set by the second-highest bidder (the last person to drop out). Note also that the Dutch auction and the first-price sealed-bid auction are equivalent.

Another important distinction is between private-value auctions, where each bidder has their own, invariant valuation of the object(s) being auctioned, and common-value auctions, where bidders might alter their valuations depending on signals (i.e. observed bids) made by other participants.

The key result in auction theory is the Revenue Equivalence Theorem, which states that all the standard auction methods, under certain plausible conditions, generate the same revenues. This accounts for the fact that no one method of conducting auctions has displaced all the others. However, practical considerations often favor one kind of auction design over another, and this is analyzed extensively later in the book.

Chapter 1 finishes with a mathematical appendix which proves some of the main results plus some questions (and answers) from the Oxford University MPhil economics examination.

Chapter 2 is titled ‘why every economist should learn some auction theory’ and extends application of the theory to economic issues such as litigation models, wars of attrition, market crashes and trading frenzies, and Internet sales models.

Chapter 3 is a detailed guide to auction design. Auction theory is about mathematical models and their properties. Auctions in the real world are competitions where real money is at stake and any and all tactics will be employed to win. Critical tactics to ‘game’ auctions include collusion between bidders to avoid bidding against each other therefore lowering prices for everyone, and predatory behavior, where weaker bidders are frightened off either before or during the auction, thereby clearing the field and closing the auction early. Examples are given of extraordinary behavior by bidders which succeeded in effectively wrecking auctions as revenue-generating vehicles. Klemperer analyses each of the four types of auction in the context of deterring such behavior and concludes that there is no one right answer: auction design is ‘horses for courses’.

Chapter 4, ‘using and abusing auction theory’, is aimed at academic auction theorists seduced by the mathematics at the expense of real-world issues. Klemperer argues that ‘undergraduate economics’, taking into account the concerns of industrial organisation (e.g. monopolistic, oligopolistic and perfectly competitive behaviours) are more central to successful auction design than some of the more ‘sophisticated’ topics of the graduate-level theory. He also emphasises, with examples, that political realities can negate even the best auction design, if the designer is naive and doesn’t take politics into account in advance.

Chapter 5 gives an overview of the European 3G spectrum auctions, which ranged from brilliantly successful to absolute disasters, while chapter 6 explains in great detail exactly how the UK auction (perhaps the most successful) was designed. This chapter is extremely insightful in indicating how many general economic intuitions have to be employed, over and above the specific insights of auction theory, to get a design which can resist the best efforts of the bidders to sabotage its effectiveness.

Chapter 7 analyses some of the more interesting and puzzling bidder strategies seen in the auctions, while chapter 8 is the reprinted Financial Times article, ‘were auctions a good idea’, which defends the idea of auctions against special pleading by industry lobbyists that they had to pay too much, and that as a consequence the industry was wrecked. Not so.

Overall, this book succeeds in creating in the non-economist reader a sense that they understand the basic terrain of auctions - what they are about - although there is clearly a much deeper set of theoretical results underpinning this map of the territory. The chapters tend to be quite repetitive, but that can help offset Klemperer’s use of jargon whenever he gets into conceptual analysis. It looks as though he doesn’t know he’s doing it, and that the people who reviewed it are his colleagues who use this stuff every day and didn’t notice either. With a small amount of additional explanation to clarify the use of terms, much of the conceptual analysis would have been more accessible. The alternative is to read the book twice! Recommended.

Tuesday, April 18, 2006

The Auction of the Empire

The Praetorian Guard had been set up by the first emperor, Augustus, as an instrument of his personal power. Comprising ten cohorts of 1,000 men each, they were stationed in and around Rome. The next emperor, Tiberius, moved the whole Guard to a specially-built citadel within Rome itself, in the early years of the first century. Around the time of Vespasian, mid 1st century, they were increased to around 16,000 elite troops.

The Praetorians soon realised their power, not simply to support emperors but also to dispose of them. In AD 193 they had murdered the emperor Pertinax. The reasons are not entirely clear: Pertinax had been in office for only three months, and had himself been implicated in the murder of the previous emperor, Commodus who had proved himself completely ineffectual. However, with Pertinax’s demise, there were no obvious successors.

Sulpicianus, the father-in-law of Pertinax and a leading public official, was endeavoring to calm the roman masses after the assassination when the Praetorian Guard marched up bearing his son-in-law’s head on a lance. Astonishingly, Sulpicianus attempted at this point to claim the mantle of emperor himself, but the Praetorian leadership, sensing a better deal, ran to a nearby vantage point and proclaimed to the waiting crowd that the empire was to be disposed of by public auction!

This offer eventually reached the ears of a wealthy senator, Didius Julianus who was sumptuously dining at the time, and he made his way to the Praetorian camp and began to bid against Sulpicianus from the foot of the ramparts. Sulpicianus had already bid a $25,000 donative for each soldier in today’s money, when Julianus submitted a ‘jump bid’ of $32,000 per Guard. This offer was enough to win the auction and buy the empire.

Note that if all the troopers were to receive this amount, the total bill would have been around half a billion dollars. For reference purposes, the annual tax revenues of the roman empire at this time were around $7 billion.

All did not end happily for Julianus. His political support did not extend beyond those he had bribed and three field generals rose against him. In the end, Septimius Severus, at the head of three eastern legions won, and Julianus was out of office and executed within 66 days. The Praetorians were also out of luck. Severus ordered them to parade unarmed outside the city, where his Danubian legions disbanded them. Severus ruled as emperor for the next eighteen years.

This is perhaps where any analogy with the 3G mobile phone auctions fails.

Sunday, April 09, 2006

Chat Bots

Still working through final editing on the book. I estimate I will be around 9,000 words short, so I have decided to add a new chapter, 'Machines Who Talk'. This will be about Natural Language Understanding (and Production) systems. Today, these are tangentially used to replace call centre agents, but surely there is a bigger role in the next-generation network. Perhaps.

My initial researches took me to "The Personality Forge" where you can play with chat bots, and create your own. I was initially quite taken with this site. My Ph. D. was in AI, and I still have a significant Lisp-sized gap in my life. I could easily spend a lot of fun hours putting together a really cute system. But then my neo-cortex kicked in.

Here is a conversation I have just had with 'Jabberwock', (JW), the winner of the 2005 chatterbox challenge. If you have not conversed with a chat bot before, it's like instant messaging - try it yourself here.

Anyway, onto the dialogue.

Jabb:... Hello, welcome to a thrilling conversational adventure. My name is Jabberwock. How can I help you?

ME: Hello, what do you think of the war in Iraq?

Jabb:... Hmm, why do you want to know my opinion about the war in Iraq?

ME: You are meant to be a prize winning conversation partner.

Jabb:...That was not my intention

ME: OK, so what are you good at?

Jabb:...Well, I can do a lot of things - but most of them are only a matter of Jabberwocks. You know - something like frightening the chicks, burning the wanna-be-heroes, doing some cloudsurfing, collecting useless things, crawling under your bed, and so on. But if you like then we can do it together, guest

ME: I see we hit a programmed response here then! Tell me about Lewis Carroll.

Jabb:...Very well, let's have a little chit-chat about Lewis Carroll. You start!

... and so on.

Unlike earlier generations of chat bot, the response are grammatical, and the re-editing of what I typed into Jabb's response is done pretty seamlessly. But it's not a conversation. The responses are totally vacuous and you would have to be pretty narcissistic to want to spend a lot of time in this kind of interaction.

If this is as good as it gets for champion systems, anything I would program in a few days would be completely useless.

These systems are driven by superficial pattern-matching on what the user types. Non-trivial conversations, by contrast, have to be about something, which means that there has to be deep domain knowledge built into the 'bot'. This makes processing much harder - you now have to figure out what the user meant - and involves the bot designer creating a body of expertise. As we know from 'expert system' projects, this can take a while.

Putting my hobbyist desires to one side, the issue I will take up in my new chapter 12 is whether a non-trivial conversational system about something (some business task) of interest and value to a customer is prospectively in the state-of-the-art at all. More later.

Wednesday, April 05, 2006

Accidents will happen

"Iraqi Prime Minister Ibrahim Jaafari has rejected growing pressure to resign, saying Iraqis must be allowed to choose their leader democratically" according to the BBC website this morning. Mr. Jaafari represents a Shia faction with links to the militias, and which has refused to embrace a project of coalition with the Sunnis. However, this project is central to American stabilisation plans for the country. Let me be one of the first million or so bloggers to predict that Mr. Jaafari is likely to be the target for some unusually resourceful insurgents in the near future.

Why are the Americans so scary? BBC's late night current affairs programme 'Newsnight' is running a Latin American week. Yesterday they showed TV footage of US tele-evangelist Pat Robertson calling for the US to assassinate Venezuelan President Hugo Chavez (link here), a man guilty of little else than spending his oil money on his own poverty-stricken people and not aligning himself with US economic and political priorities.

In "Hackers and Painters", Paul Graham has an essay called 'What You Can't Say' (chapter 3). This is about the opinions which appear to be self-evident, but which other people, and later generations, will think the epitome of self-interest and/or parochialism.

The American imperial world view sees the ideal world as America everywhere and bemoans the backwardness of the rest of the world in failing to have already achieved this goal. It gets scary when their aggressive patriotism kicks in, and they consider foreign non-alignment to US values and interests to be an affront to decency which requires a physical force reply. "We're nice" they say to themselves and everyone else,"Why do they hate us?". And they conclude that 'the others', for some unaccountable reason, 'hate freedom'. Duh!

At least the Romans were less prone to self-deception. "'Let them hate us so long as they fear us." opined Caligula. As American power gets further challenged over the coming century, it's going to be a scary ride.

Monday, April 03, 2006

Progress with the book, John-Paul II

First things first. The book is basically finished, but around 10-15,000 words short. Not an insurmountable task. As of today, I have revised chapters 1 and 2, and was horrified to see that chapter 3 was a meandering series of points of frightening shapelessness. Amazing what an absence from the scene of a few months can do to engender insight.

I have to believe in the law of diminishing awfulness, otherwise I would never dare submit. At least this was one of the under-revised chapters. Still, there are three or four coming up which are similarly under-visited, and my recollections of parental pride are tinged with some subconscious reservations. On the brighter side, fixing the flaws is an opportunity to write, and therefore address the magic 15k.

"Having therefore such a hope, we use great boldness of speech" (2 Corinthians 3:12).

A biblical segue to Clare's wish that I write something insightful about the Catholic Church's fast-track to sainthood for JP II.

I did tax her in some detail about the conceptual foundation and ontological status of sainthood, and its role in client-patron brokerage in heaven. I have to say that I did not get an altogether authoritative response, but I understand that she feels the process is a trifle rushed. I would concur for values of rushed which tend to infinity.