Not much new to report after the second game of the Watson Jeopardy Challenge. I've added a few updates to yesterday's post on Watson and the result was as expected, though Watson struggled a lot more in this game than in the prior round, deciding not to answer many questions due to low confidence and making a few mistakes. In a few cases it was saved by not buzzing fast enough even though it had over 50% confidence, as it would have answered slightly wrong.

The computer scientist world is abuzz with the game show world over the showdown between IBM's "Watson" question-answering system and the best human players to play the game Jeopardy. The first game has been shown, with a crushing victory by Watson (in spite of a tie after the first half of the game.)

Tomorrow's outcome is not in doubt. IBM would not have declared itself ready for the contest without being confident it would win, and they wouldn't be putting all the advertising out about the contest if they had lost. What's interesting is how they did it and what else they will be able to do with it.

Dealing with a general question has long been one of the hard problems in AI research. Watson isn't quite there yet but it's managed a great deal with a combination of algorithmic parsing and understanding combined with machine learning based on prior Jeopardy games. That's a must because Jeopardy "answers" (clues) are often written in obfuscated styles, with puns and many idioms, exactly the sorts of things most natural language systems have had a very hard time with.

Watson's problem is almost all understanding the question. Looking up obscure facts is not nearly so hard if you have a copy of Wikipedia and other databases on hand, particularly one parsed with other state-of-the-art natural language systems, which is what I presume they have. In fact, one would predict that Watson would do the best on the hardest $2,000 questions because these are usually hard because they refer to obscure knowledge, not because it is harder to understand the question. I expect that an evaluation of its results may show that its performance on hard questions is not much worse than on easy ones. (The main thing that would make easy questions easier would be the large number of articles in its database confirming the answer, and presumably boosting its confidence in its answer.) However, my intuition may be wrong here, in that most of Watson's problems came on the high-value questions.

It's confidence is important. If it does not feel confident it doesn't buzz in. And it has a serious advantage at buzzing in, since you can't buzz in right away on this game, and if you're an encyclopedia like the two human champions and Watson, buzzing in is a large part of the game. In fact, a fairer game, which Watson might not do as well at, would involve randomly choosing which of the players who buzz in in the first few tenths of a second gets to answer the question, eliminating any reaction time advantage. Watson gets the questions as text, which is also a bit unfair, unless it is given them one word a time at human reading speed. It could do OCR on the screen but chances are it would read faster than the humans. It's confidence numbers and results are extremely impressive. One reason it doesn't buzz in is that even with 3,000 cores it takes 2-6 seconds to answer a question.

Indeed a totally fair contest would not have buzzing in time competition at all, and just allow all players who buzz in to answer an get or lose points based on their answer. (Answers would need to be in parallel.)

Watson's coders know by now that they probably should have coded it to receive wrong answers from other contestants. In one instance it repeated a wrong answer, and in another case it said "What is Leg?" after Jennings had incorrectly answered "What is missing an arm?" in a question about an Olympic athlete. The host declared that right, but the judges reversed that saying that it would be right if a human who was following up the wrong answer said it, but was a wrong answer without that context. This was edited out. Also edited out were 4 crashes by Watson that made the game take 4 hours instead of 30 minutes.

It did not happen in what aired so far, but in the trials, another error I saw Watson make was declining to answer a request to be more specific on an answer. Watson was programmed to give minimalist answers, which often the host will accept as correct, so why take a risk. If the host doesn't think you said enough he asks for a more specific answer. Watson sometimes said "I can be no more specific." From a pure gameplay standpoint, that's like saying, "I admit I am wrong." For points, one should say the best longer phrase containing the one-word answer, because it just might be right. Though it has a larger chance of looking really stupid -- see below for thoughts on that.

The shows also contain total love-fest pieces about IBM which make me amazed that IBM is not listed as a sponsor for the shows, other than perhaps in the name "The IBM Challenge." I am sure Jeopardy is getting great ratings (just having their two champs back would do that on its own but this will be even more) but I have to wonder if any other money is flowing.

Being an idiot savant

Watson doesn't really understand the Jeopardy clues, at least not as a human does. Like so many AI breakthroughs, this result comes from figuring out another way to attack the problem different from the method humans use. As a result, Watson sometimes puts out answers that are nonsense "idiot" answers from a human perspective. They cut back a lot on this by only having it answer when it has 50% confidence or higher, and in fact for most of its answers it has very impressive confidence numbers. But sometimes it gives such an answer. To the consternation of the Watson team, it did this on the Final Jeopardy clue, where it answered "Toronto" in the category "U.S. Cities."

Some years ago I made the proposal that airlines sell half of a middle seat at half price or less so that two coach passengers could assure they would have an empty middle next to them.

I shoot lots of large panoramas, and the arrival of various cheaper robotic mounts to shoot them, such as the Gigapan Epic Pro and the Merlin/Skywatcher (which I have) has resulted in a bit of a "mine's bigger than yours" contest to take the biggest photo. Some would argue that the stitched version of the Sloane Digital Sky survey, which has been rated at a trillion pixels, is the winner, but most of the competition has been on the ground.

A release from the National Federation for the Blind reports a blind person driving and avoiding obstacles on the Daytona speedway. They used a car from the TORC team at Virginia Tech, one of the competitors in the Darpa Grand Challenges. In effect, the blind driver replaced the "drive by wire" component of a robocar with a more intelligent and thinking human also able to feel acceleration and make some judgements.

Last year I wrote about an interesting but simple pedal powered monorail/PRT system called Shweeb which had won a prize/investment from Google. Recent announcements show they are not alone in this concept. Scott Olson, the original developer of the Rollerblade, has founded a company called Skyride Technologies to build their own version of a pedal powered suspended monorail.

As readers of this blog surely know, for several years I have been designing, writing and forecasting about the technology of self-driving "robocars" in the coming years. I'm pleased to announce that I have recently become a consultant to the robot car team working at Google.

Of course all that work will be done under NDA, and so until such time as Google makes more public announcements, I won't be writing about what they or I are doing. I am very impressed by the team and their accomplishments, and to learn more I will point you to my blog post about their announcement and the article I added to my web site shortly after that announcement. It also means I probably won't blog in any detail about certain areas of technology, in some cases not commenting on the work of other teams because of conflict of interest. However, as much as I enjoy writing and reporting on this technology, I would rather be building it.

My philosophical message about Robocars I have been saying for years, but it should be clear that I am simply consulting on the project, not setting its policies or acting as a spokesman.

My primary interest at Google is robocars, but many of you also know my long history in online civil rights and privacy, an area in which Google is often involved in both positive and negative ways. Indeed, while I was chairman of the EFF I felt there could be a conflict in working for a company which the EFF frequently has to either praise or criticise. I will be recusing myself from any EFF board decisions about Google, naturally.

Every day I get into my car and drive somewhere. My mobile phone has a lot of useful apps for travel, including maps with traffic and a lot more. And I am usually calling them up.

I believe that my phone should notice when I am driving off from somewhere, or about to, and automatically do some things for me. Of course, it could notice this if it ran the GPS all the time, but that's expensive from a power standpoint, so there are other ways to identify this:

Like me, you probably have a dozen "universal" remote controls gathered over the years. With each new device and remote you go through a process to try to figure out special codes to enter into the remote to train it to operate your other devices. And it's never very good, except perhaps in the expensive remotes with screens and macros.

The "burning" question for electric cars is how to compare them with gasoline. Last month I wrote about how wrong the EPA's 99mpg number for the Nissan Leaf was, and I gave the 37mpg number you get from the Dept. of Energy's methodology. More research shows the question is complex and messy.

So messy that the best solution is for electric cars to publish their efficiency in electric terms, which means a number like "watt-hours/mile." The EPA measured the Leaf as about 330 watt-hours/mile (or .33 kwh/mile if you prefer.) For those who really prefer an mpg type number, so that higher is better, you would do miles/kwh.

Then you would get local power companies to publish local "kwh to gallon of gasoline" figures for the particular mix of power plants in that area. This also is not very easy, but it removes the local variation. The DoE or EPA could also come up with a national average kwh/gallon number, and car vendors could use that if they wanted, but frankly that national number is poor enough that most would not want to use it in the above-average states like California. In addition, the number in other countries is much better than in the USA.

The local mix varies a lot. Nationally it's about 50% coal, 20% gas, 20% nuclear and 10% hydro with a smattering of other renewables. In some places, like Utah, New Mexico and many midwestern areas, it is 90% or more coal (which is bad.) In California, there is almost no coal -- it's mostly natural gas, with some nuclear, particularly in the south, and some hydro. In the Pacific Northwest, there is a dominance by hydro and electricity has far fewer emissions. (In TX, IL and NY, you can choose greener electricity providers which seems an obvious choice for the electric-car buyer.)

Understanding the local mix is a start, but there is more complexity. Let's look at some of the different methods, staring with an executive summary for the 330 wh/mile Nissan Leaf and the national average grid:

• Theoretical perfect conversion (EPA method): 99 mpg-e(perfect)
• Heat energy formula (DoE national average): 37 mpg-e(heat)
• Cost of electricity vs. gasoline (untaxed): 75 mpg-e($)
• Pollution, notably PM2.5 particulates: Hard to calculate, could be very poor. Hydrocarbons and CO: very good.
• Greenhouse Gas emissions, g CO2 equivalent: 60 mpg-e(CO2)

Like just about everybody, I hate the way travel through airports has become. Airports get slower and bigger and more expensive, and for short-haul flights you can easily spend more time on the ground at airports than you do in the air. Security rules are a large part of the cause, but not all of it.

In this completely rewritten essay, I outline the design on a super-cheap airport with very few buildings, based on a fleet of proto-robocars. I call them proto models because these are cars we know how to build today, which navigate on prepared courses on pavement, in controlled situations and without civilian cars to worry about.

In this robocar airport, which I describe first in a narrative and then in detail, there are no terminal buildings or gates. Each plane just parks on the tarmac and robotic stairs and ramps move up and dock to all its doors. (Catering trucks, fuel trucks and luggage robots also arrive.) The passengers arrive in a perfect boarding order in robocars that dock at the ramps/steps to let them get on the plane through every entrance. Luggage is handled by different robots, and is checked and picked up not in carousels and check-in desks, but at curbs, parking lots, rental car centers and airport hotels.

The change is so dramatic that (even with security issues) people could arrive at airports for flights under 20 minutes before take-off, and get out even faster. Checked luggage would add time, but not much. I also believe you could build a high capacity airport for a tiny fraction of the cost of today's modern multi-billion dollar edifices. I believe the overall experience would also be more pleasant and more productive for all.

This essay is a long one, but I am interested in feedback. What will work here, and what won't? Would you love to fly through this airport or hate it? This is an airport designed not to give you a glorious building in which to wait but to get you through it without waiting most of the time.

The airport gets even better when real robocars, that can drive on the streets to the airport, come on the scene.

Give me your feedback on The Robocar Airport.

Key elements of the design include:

This year, I bought Microsoft Kinect cameras for the nephews and niece. At first they will mostly play energetic X-box games with them but my hope is they will start to play with the things coming from the Kinect hacking community -- the videos of the top hacks are quite interesting. At first, MS wanted to lock down the Kinect and threaten the open source developers who reverse engineered the protocol and released drivers. Now Microsoft has official open drivers.

I've written frequently about how driving fatalities are the leading cause of death for people from age 5 to 45, and one of the leading overall causes of death. I write this because we hope that safe robocars, with a much lower accident rate, can eliminate much of this death.

Passwords are in the news thanks to Gawker media, who had their database of userids, emails and passwords hacked and published on the web. A big part of the fault is Gawker's, who was saving user passwords (so it could email them) and thus was vulnerable. As I have written before, you should be very critical of any site that is able to email you your password if you forget it.

Some of the advice in the wake of this to users has been to not use the same password on multiple sites, and that's not at all practical in today's world. I have passwords for many hundreds of sites. Most of them are like gawker -- accounts I was forced to create just to leave a comment on a message board. I use the same password for these "junk accounts." It's just not a big issue if somebody is able to leave a comment on a blog with my name, since my name was never verified in the first place. A different password for each site just isn't something people can manage. There are password managers that try to solve this, creating different passwords for each site and remembering them, but these systems often have problems when roaming from computer to computer, or trying out new web browsers, or when sites change their login pages.

The long term solution is not passwords at all, it's digital signature (though that has all the problems listed above) and it's not to even have logins at all, but instead use authenticated actions so we are neither creating accounts to do simple actions nor using a federated identity monopoly (like Facebook Connect). This is better than OpenID too.

I decided to gather together all my thoughts on how robocars will affect urban design. There are many things that might happen, though nobody knows enough urban planning to figure out just what will happen. However, I felt it worthwhile to outline the forces that might be at work so that urban geographers can speculate on what they will mean. It is hard to make firm predictions.

There are many fields that people expect robotics to change in the consumer space. I write regularly about transportation, and many feel that robots to assist the elderly will be the other big field. The first successful consumer robot (outside of entertainment) was the Roomba, a house cleaning robot. So I've often wondered about how far we are from a robot that can tidy up the house. People got excited with a PR2 robot was programmed to fold towels.

This is a hard problem because it seems such a robot needs to do general object recognition and manipulation, something we're pretty far from doing. Special purpose household chore robots, like the Roomba, might appear first. (A gutter cleaner is already on the market.)

Recently I was pondering what we might do with a robot that is able to pick up objects gently, but isn't that good at recognizing them. Such a robot might not identify the objects, but it could photograph them, and put them in bins. The members of the household could then go to their computers and see a visual catalog of all the things that have been put away, and an indicator of where it was put. This would make it easy to find objects.

The catalog could trivially be sorted by when the items were put away, which might well make it easy to browse for something put away recently. But the fact that we can't do general object recognition does not mean we can't do a lot of useful things with photographs and sensor readings (including precise weight and other factors) beyond that. One could certainly search by colour, by general size and shape, and by weight and other characteristics like rigidity. The item could be photographed in a 360 view by being spun on a table or in the grasping arm, or which a rotating camera. It could also be laser-scanned or 3D photographed with new cheap 3D camera techniques.

When looking for a specific object, one could find it by drawing a sketch of the object -- software is already able to find photos that are similar to a sketch. But more is possible. Typing in the name of what you're looking for could bring up the results of a web image search on that string, and you could find a photo of a similar object, and then ask the object search engine to find photos of objects that are similar. While ideally the object was photographed from all angles, there are already many comparison algorithms that survive scaling and rotation to match up objects.

The result would be a fairly workable search engine for the objects of your life that were picked up by the robot. I suspect that you could quickly find your item and learn just exactly where it was.

Certain types of objects could be recognized by the robot, such as books, papers and magazines. For those, bar-codes could be read, or printing could be scanned with OCR. Books might be shelved at random in the library but be easily found. Papers might be hard to manipulate but could at least be stacked, possibly with small divider sheets inserted between them with numbers on them, so that you could look for the top page of any collected group of papers and be told, "it's under divider 20 in the stack of papers."

The folks at the SARTRE road train project have issued an update one year into their 3 year project. This is an EU-initiated project to build convoy technology, where a professional lead driver in a truck or bus is followed by a convoy of closely packed cars which automatically follow based on radio communications (and other signals) with the lead. They have released a new video on their progress from Volvo.

Two bits of robocar news from last week. I had been following the progress of the Stanford/VW team that was building a robotic Audi TT to race to the top of Pikes Peak. They accomplished their run in September, but only now made the public announcement of it. You can find photos and videos with the press release or watch a video on youtube.

Nissan is touting that the EPA gave the new Leaf a mileage rating of 99mpg "gasoline equivalent". What is not said in some stories (though Nissan admits it in the press release) is that this is based on the EPA rating a gallon of gasoline as equivalent to 33.7 kwh, and the EPA judging that the car only goes 73 miles on its 24kwh battery.

You're driving down the road. You see another car on the road with you that has a problem. The lights are off and it's dusk. There is something loose that may break off. There's something left on the roof or the trunk is not closed -- any number of things. How do you tell the driver that they need to stop and check? I've tried sometimes and they mostly think you are some sort of crazy, driving to close to them, waving at them, honking or shouting. Perhaps after a few people do it they figure it out.