Tom Barbalet perusing his book shelf. I think I also read some of those same books in the 1980s - definitely the Space games one, and possibly also the fantasy games book. Those books were quite well written and certainly conveyed the impression of the computer as a kind of window into a universe of unlimited creativity, which as a kid and even as an adult is quite exciting. The user manuals which came with the BBC Micro were also quite useful for learning the BASIC programming language, and contained fairly clear descriptions of the syntax.
What's less easy to convey though is just how rare books about computer programming were in the 1980s - certainly to a kid of unremarkable means. Getting any information on computer software or hardware more advanced than the beginners books was either hard or impossible. Typically libraries contained only a handfull of computer related books, usually of the "gee wiz"/coffee table variety rather than technical manuals. So for most kids of my generation technical discoveries were made largely by a combination of experiment and social networking. Detailed information could be acquired through the British Library - if you knew what you were looking for - and this was a protracted and moderately costly process which could take anywhere from several weeks to several months. Imagine typing a query into Google and only getting an answer back three months later!
By contrast it's much easier for anyone of any age to learn programming today, with the information being mostly freely available online and truly gigantic in volume. Within my lifetime there has been a complete reversal from a situation of information poverty to one of information ubiquity.
Wednesday, September 30, 2009
Tuesday, September 29, 2009
The age of sonic weapons begins
A sonic weapon called the Long Range Acoustic Device (LRAD) is used for the first time against civilians at the G20 protests.
According to Wikipedia this is a relatively unsophisticated device which is not much more than a bunch of tweeters able to emit a sufficiently loud blast of sound to damage hearing.
The low tech nature of this makes me wonder. Traditional riot control methods such as water cannons and rubber bullets would be prohibitively difficult for a rioter to build or operate themselves, but a crude sonic weapon like the LRAD protentally could be constructed in a back yard using maybe a car battery and a bunch of commercially available speakers. Such devices might even become of interest to terrorists, since although they're non-lethal the threat of having your hearing possibly permanently damaged would be enough to cause panic and distress, and it would be fairly easy to bolt such a device onto a vehicle and quickly drive through a crowded area.
When the history books get written of course this will be just the very early beginnings of a whole range of directed energy weapons.
According to Wikipedia this is a relatively unsophisticated device which is not much more than a bunch of tweeters able to emit a sufficiently loud blast of sound to damage hearing.
The low tech nature of this makes me wonder. Traditional riot control methods such as water cannons and rubber bullets would be prohibitively difficult for a rioter to build or operate themselves, but a crude sonic weapon like the LRAD protentally could be constructed in a back yard using maybe a car battery and a bunch of commercially available speakers. Such devices might even become of interest to terrorists, since although they're non-lethal the threat of having your hearing possibly permanently damaged would be enough to cause panic and distress, and it would be fairly easy to bolt such a device onto a vehicle and quickly drive through a crowded area.
When the history books get written of course this will be just the very early beginnings of a whole range of directed energy weapons.
The benefits of vaccination
There's a flap going on in the mainstream media currently about the death of a girl, which was allegedly caused by being vaccinated against a virus which causes cervical cancer. Having watched some TV news item about it - and its a rare event for me to watch TV at all these days - one thing which is really obvious is the complete lack of any analysis of the figures or presentation of contextual information. To the naive viewer the take-home message is something like:
"OMG! My child could die after being vaccinated!"The figures presented were 1 death out of 1.4 million vaccinations (that's 0.00007%), but what they didn't bother to mention was how many of those girls would have subsequently died from this particular form of cancer had they not received the vaccine. I suspect that the percentage would be somewhat higher. The lack of this kind of information, presented on prime time TV in an simple way which most people can understand, means that disquieted parents will turn to Google, where they will quickly encounter the purveyors of half-truths, ignorance and pseudo-science, whose messages are often loud, contain entertaining tales of skullduggery, and of course are utterly untainted by any kind of credible evidence.
Monday, September 28, 2009
Sunday, September 27, 2009
Should you report your employer's pirated software?
If you're employed, that is. My answer would be that if you can't reasonably persuade them to use fully legal software, and in most cases it is possible to do that, then yes you should have absolutely no hesitation about reporting them. Strict observance of the terms of licenses and copyrights is in my opinion the best route to encourage far greater adoption of free and open source software in the workplace. As I've long observed, the traditional proprietary software model contains inherrent absurdities, which become clear if you strictly adhere to what the copyright terms say.
If initially raising the issue fails to get a result, or if they endlessly stall on making a decision, you can use diplomacy to steer them towards the right side of the law. Show them web sites where you can report software piracy and say something like "Anyone in this office could use this site, either at work or when they go home, and especially if they feel that their job might be at risk."
If initially raising the issue fails to get a result, or if they endlessly stall on making a decision, you can use diplomacy to steer them towards the right side of the law. Show them web sites where you can report software piracy and say something like "Anyone in this office could use this site, either at work or when they go home, and especially if they feel that their job might be at risk."
Saturday, September 26, 2009
When computer scientists protest...
Classic. More here.
Free the variables!
I like the little graph here.
There's some other reporting of the G20 protest here. I'm not really sure what the protesters are actually protesting about though, and pushing wheely bins around in the street seems like a bizarre persuit. Why are the police tear-gassing people pushing wheely bins? Shouldn't they just move out of the way and let the bins roll? Are the protesters trying to influence the opinions or decisions of the national leaders? If so, using wheely bins doesn't seem like a very efficient way of doing that.
Sometimes it seems like I'm an alien who has beamed down from the mothership and who is observing the strange behavior of the human species without being able to understand what it all means. Generally people don't do things without some kind of motivation which makes logical sense at least within their own world view.
Free the variables!
I like the little graph here.
There's some other reporting of the G20 protest here. I'm not really sure what the protesters are actually protesting about though, and pushing wheely bins around in the street seems like a bizarre persuit. Why are the police tear-gassing people pushing wheely bins? Shouldn't they just move out of the way and let the bins roll? Are the protesters trying to influence the opinions or decisions of the national leaders? If so, using wheely bins doesn't seem like a very efficient way of doing that.
Sometimes it seems like I'm an alien who has beamed down from the mothership and who is observing the strange behavior of the human species without being able to understand what it all means. Generally people don't do things without some kind of motivation which makes logical sense at least within their own world view.
Plotting propositions
Closeup of a section of the Mindpixel GAC-80K corpus plotted in phonetic space. The horizontal axis corresponds to a phoneme (3-gram) index. The vertical corresponds to a standardised soundex index. Green pixels are probably true propositions. Blue pixels are probably false propositions, with pink/red being closest to zero probability. Black pixels are unmapped (no Mindpixels mapped to these locations).
The map is updated in a similar way to an occupancy grid, with each Mindpixel being represented as a small gaussian distributed probability value.
The complete corpus looks like this.
Islands of truth drifting in a sea of negative knowledge.
Ultimately there will be some mapping between speech sounds and the truth of statements made by an average English speaker (if there weren't, you wouldn't be able to answer questions), but it's going to be a very complex mapping and involve many more dimensions than just the two shown here.
The map is updated in a similar way to an occupancy grid, with each Mindpixel being represented as a small gaussian distributed probability value.
"MISTIC's primary purpose is to build a map of human consciousness. The only people that have ever seen an actual visual map (adaptive-subspace self-organizing map) of human consciousness are on the MISTIC team...
it's lowres right now, but we know what it looks like (very fractal)..." - K. Christopher McKinstry, 1997
The complete corpus looks like this.
Islands of truth drifting in a sea of negative knowledge.
Ultimately there will be some mapping between speech sounds and the truth of statements made by an average English speaker (if there weren't, you wouldn't be able to answer questions), but it's going to be a very complex mapping and involve many more dimensions than just the two shown here.
Thursday, September 24, 2009
On the busses part 2: Laboratory conditions
A BBC article describes a system which supposedly will be capable of detecting people being assaulted on busses. Having tried detecting people "in the wild" it's a notoriously difficult problem, so I'm somewhat skeptical that this is going to work well. On a bus you've got constantly shifting shadows as the vehicle moves, and people frequently moving about or fiddling with bags, coats, tickets and gadgets. A bus is at least a semi-structured environment, so the system would know where passengers were likely to be sitting, but probably many cameras would be needed to avoid excessive occlusions (a possible solution might be multiple omnidirectional cameras mounted on the ceiling).
Concievably with enough cameras having overlapping fields of view and the right software it might be possible to construct a real time 3D volumetric model of the environment and the passengers in it. Only then I think it might be possible to detect a salient event such as a handbag being snatched or someone being punched. But even by current standards this would be a very sophisticated computer perception and interpretation system - essentially solving, at least partially, the grounding problem and implementing some subset of commonsense reasoning.
More likely I think they'll try implementing a system which just scrapes the 2D pixel image looking for motion or moving segments/blobs, because this is much easier and less computationally demanding than volumetric modelling. Unfortunately though the 2D methods are fraught with ambiguity, because you're just dealing with a low dimensional shadow of what's really happening which can correspond to multiple possible 3D configurations.
So I think with enough effort over a five year period combined with a volumetric approach they might be able to pull it off, but the trouble is that this isn't really low hanging fruit and attempts at incremental improvements to 2D techniques currently used to process CCTV images won't deliver the kind of performance they're hoping for (I think they would really struggle to eliminate false positives).
Concievably with enough cameras having overlapping fields of view and the right software it might be possible to construct a real time 3D volumetric model of the environment and the passengers in it. Only then I think it might be possible to detect a salient event such as a handbag being snatched or someone being punched. But even by current standards this would be a very sophisticated computer perception and interpretation system - essentially solving, at least partially, the grounding problem and implementing some subset of commonsense reasoning.
More likely I think they'll try implementing a system which just scrapes the 2D pixel image looking for motion or moving segments/blobs, because this is much easier and less computationally demanding than volumetric modelling. Unfortunately though the 2D methods are fraught with ambiguity, because you're just dealing with a low dimensional shadow of what's really happening which can correspond to multiple possible 3D configurations.
"The system, part of which has already been tested in laboratory conditions..."To anyone who has done any amount of computer vision work this statement ought to immediately ring alarm bells, because there's always a massive difference between tests under controlled lighting (or worse, simulation) and performance in a much more unconstrained and cluttered visual environment filled with interestingly shaped objects (people and their luggage).
So I think with enough effort over a five year period combined with a volumetric approach they might be able to pull it off, but the trouble is that this isn't really low hanging fruit and attempts at incremental improvements to 2D techniques currently used to process CCTV images won't deliver the kind of performance they're hoping for (I think they would really struggle to eliminate false positives).
Tuesday, September 22, 2009
Stereo camera calibration with the Surveyor SVS
A video showing the calibration procedure, which is described here.
The rectification isn't totally perfect, but I aimed for something which is a balance between convenience and accuracy and gave reasonable looking stereo disparities. Probably better calibration is possible by integrating results over several frames. This is actually a single frame technique which picks the result with the smallest RMS error. There is camera calibration code within OpenCV, although there would probably be licensing issues with using that code within the Sentience project.
The rectification isn't totally perfect, but I aimed for something which is a balance between convenience and accuracy and gave reasonable looking stereo disparities. Probably better calibration is possible by integrating results over several frames. This is actually a single frame technique which picks the result with the smallest RMS error. There is camera calibration code within OpenCV, although there would probably be licensing issues with using that code within the Sentience project.
Whole brain emulation
Whole brain emulation sounds science fictional, but I think at least part of this programme - reverse engineering the structure of a significantly sized brain, such as a mouse - could be achievable on quite a short time scale of years rather than decades.
Whole brain emulation is the most conservative route to non-biological intelligence. In a sense WBE wouldn't really be "artificial intelligence", it would merely be biological intelligence transferred into a different physical substrate. Even with an emulated brain we might still understand little about how it actually works, and that might be a project which takes a significant amount of additional analysis.
The most speculative aspect would be producing a functional model, since this requires having a detailed and accurate simulation of how neurons operate and grow. But at this point even a non-functioning, purely structural model of the complete connectome of a brain would be a big step forward.
Most of the techniques and technology required to scan a brain and produce a complete 3D model of every neuron exists today, but has typically only been used on very small sections. What remains is to scale up the effort.
Whole brain emulation is the most conservative route to non-biological intelligence. In a sense WBE wouldn't really be "artificial intelligence", it would merely be biological intelligence transferred into a different physical substrate. Even with an emulated brain we might still understand little about how it actually works, and that might be a project which takes a significant amount of additional analysis.
The most speculative aspect would be producing a functional model, since this requires having a detailed and accurate simulation of how neurons operate and grow. But at this point even a non-functioning, purely structural model of the complete connectome of a brain would be a big step forward.
Most of the techniques and technology required to scan a brain and produce a complete 3D model of every neuron exists today, but has typically only been used on very small sections. What remains is to scale up the effort.
mylifemyid: How young people rejected ID cards
The Register reports on how the Home office is refusing to publish research which it carried out last year with people aged between 16 and 25 on the subject of the UK ID card scheme.
The site was http://mylifemyid.org, and of course it's no longer available. It also seems that they were fairly determined to make sure that it didn't get archived.
"Anyone who had though of hacking the MyLifeMyID site would have come across a programmer’s note-to-self that read 'generally we don’t want to be indexed'. This note appeared in the 'robots.txt' file, a file that sat on the MyLIfeMyID web server and basically said to search engines 'move along now, there’s nothing to see here'"However, like just about everything else on the web it's hard to take down information without leaving a few traces behind, like this archived page. The blogosphere also remembers:
"Have you actually read the site? Have you seen the opinions voiced on there? Not one post other than by the administrators takes a positive view of the ID scheme. Even though the administrators are removing a lot of posts and banning a lot of people there has yet to be anyone standing up and saying they like the idea. And in general there’s a huge suspicion of the government claiming they’re just canvassing opinion. We’ve all heard the way Downing Street petition statistics are misused at PMQs and I have no doubt that the sole purpose of this site was to produce a statistic the Home Office can use to totally bend the truth and tell us we (the student population) are broadly in favour of something in actuality a huge number are completely opposed."Some other blog references:
Graceful exits
Skeptobot
The Future Place Blog
ID-watch
Tell Them What You Think
Some feedback on the running of mylifemyid, posted to the no2ID forum
"I made a thread asking how much the government was paying them, which was then immediately removed and I was banned.. I can’t remember all the posts I made because there is no record of them now, which is a crucial flaw of the forum. Many people’s opinions have been suppressed"All of this suggests that such public consultation exercises have no real intention of engaging with the public and listening to their views. When people express positive views about ID cards, they get onto the BBC web site. When they express negative views their opinions are erased from internet history. I assume that the plan here was to launch the site in a very low key way and get a few opinions which could be heavily moderated to look as if young people were on balance in favour of the cards.
As far as dispelling myths goes I think it's true that the proposed ID card will not carry bank details - initially. But this is the usual politicians slight of tongue because once the ID infrastructure is in place it will be very easy either to link your bank details to the card via a reference number or serial number, or to explicitly pass legislation requiring bank account numbers to be written onto the card's chip.
Even if the ID card scheme is scrapped in 2010 by the new administration issues surrounding data collection and data use will remain, and it does seem likely that huge quantities of personal information will be accumulated and searchable by various organisations, governments and companies (it is now, but the heap will grow much taller). There's no doubt that information is power, and if we're to maintain some kind of balanced society which isn't just a kind of digital tyranny (something like the world of GATTACA) the only way forward is to move towards equiveillance, such that no individual or group has a monopoly on information about people's lives. You might call such a situation a "global village", where everyone knows about the affairs of everyone else.
[Supplemental]
Upon further Googling it appears that some public spirited individual actually did archive the forums to the site before it went down. You can find the archives here. Looking through the text this does appear to be genuine, and there are some good points made. Also it does appear to be true that the comments range from ambivolence to outright hostility towards the scheme. There seem to be no commentators expressing support. Here's an example:
"The current system means that when I lose one document, I can still go about my life normally. The current system means that no one can impersonate me if they steal or find just one document. I don't use the same key for my house and car. Why would I want a single key to my life?"Also there are a few comments like this:
"Why was that post just edited?And here's another forum comment that I found amusing
What was wrong with what I said?"
"The creepy government ID card that the Home Office is paying you to research comes with possible £1000 fines if you lose it, damage it, or change your address without telling them.And a final comment made as the forum is being closed down.
It's all set out in section 11 of the Identity Cards Act, which you've conveniently linked from the front page of your site. Why not take a moment to go and read it? We can wait.
Back again? OK, answer me this - would you pay any money at all for something that could make you subject to repeated £1000 fines, and which you would never be allowed to hand back for the rest of your life? Or would you in fact go to almost any lengths to avoid being given such a thing in the first place?
I can answer for me - under no circumstances whatsoever will I ever have an ID card issued under the Identity Cards Act 2006.
Ever.
What about you?
Hope that's clear ..."
"Whatever happens, this forum has demonstrated that the 'yoof' of today isn't the stereotype the government thinks it is dealing with. We aren't a bunch of semi-literate, happy slapping twits with no knowledge of politics or advanced political opinions. Indeed, this forum has demonstrated that many here have fairly insightful and advanced political views."Having now reviewed much of the raw data (opinions given on various hypothetical scenarios within which ID cards might be used) I think the reason why the government is stalling on releasing the results is that they will not be able to honestly make a claim such as "many young people expressed positive views towards the scheme", or even "some young people...". With the possible exception of some leading opinions entered by the moderators the comments are universally expressing views which criticise ID cards - sometimes with a scathing level of articulacy.
Sunday, September 20, 2009
Beer Zone
An amusing anecdote from Dan Lynch describing drinking beer in America. I also noticed what seemed to be "ghosts of prohibition" when I was in the US, although I assumed it was just some cultural affectation of Pittsburgh. I remember passing a few places with frosted over windows which looked extremely dodgy - like 1920s lock-ins. I assumed they were pubs, but they looked so inhospitable that I never dared to enter one, for fear of being clobbered over the head by an enraged Glaswegian (or the US equivalent thereof). They looked something like this:
Saturday, September 19, 2009
Clarke's laws
1. When a distinguished but elderly scientist states that something is possible, he is almost certainly right. When he states that something is impossible, he is very probably wrong.
2. The only way of discovering the limits of the possible is to venture a little way past them into the impossible.
3. Any sufficiently advanced technology is indistinguishable from magic.
2. The only way of discovering the limits of the possible is to venture a little way past them into the impossible.
3. Any sufficiently advanced technology is indistinguishable from magic.
Are privacy policies a big waste of time?
Interesting studies on online privacy policies.
Basically the conclusion is that very few people actually read the privacy policies on web sites, and of those that do the most savvy users only have an average of about 60% comprehension of what's written.
I'm definitely an average user here, and I don't think I've ever actually read a privacy policy. This isn't because I'm not interested in privacy, but that I expect that these things are written in convoluted legalistic language (which seems to be correct according to the comprehensibility measures shown in the above study) such that even if I invested the time I'd probably be no wiser at the end of it.
One factor not mentioned in the study which I think is relevant is that such policies are often not intended to be understood by the user. That is, they're obfuscatory and designed to conceal activities which if more clearly phrased might cause concern and possibly reduce the popularity of the site or service being provided.
As far as interacting with web sites is concerned I think it's prudent to just assume that there is no privacy and that whatever data you enter - with the possible exception of credit card numbers - will be shared around, hacked, mashed up, left on USB sticks or laptops and then lost, crawled, burned onto DVD and posted by snail mail, data mined and keyword searched by multiple systems, companies and governments.
If there are serious concerns about online privacy I think a better solution than privacy policies which almost nobody understands would be to have a site which permitted digg-like privacy ratings for different sites, because fundamentally it's a question of how much or how little you trust a given organisation with your data or financial transaction. ebay has this sort of trust mechanism. If enough users voted it would then become clear which companies were more or less trustworthy. If there were to be an element of legislation it might be that sites must link to the privacy rating site so that their current trust rating is displayed. The big flaw with this otherwise cunning plan though is that internet voting systems can usually be easily gamed.
Basically the conclusion is that very few people actually read the privacy policies on web sites, and of those that do the most savvy users only have an average of about 60% comprehension of what's written.
I'm definitely an average user here, and I don't think I've ever actually read a privacy policy. This isn't because I'm not interested in privacy, but that I expect that these things are written in convoluted legalistic language (which seems to be correct according to the comprehensibility measures shown in the above study) such that even if I invested the time I'd probably be no wiser at the end of it.
One factor not mentioned in the study which I think is relevant is that such policies are often not intended to be understood by the user. That is, they're obfuscatory and designed to conceal activities which if more clearly phrased might cause concern and possibly reduce the popularity of the site or service being provided.
As far as interacting with web sites is concerned I think it's prudent to just assume that there is no privacy and that whatever data you enter - with the possible exception of credit card numbers - will be shared around, hacked, mashed up, left on USB sticks or laptops and then lost, crawled, burned onto DVD and posted by snail mail, data mined and keyword searched by multiple systems, companies and governments.
If there are serious concerns about online privacy I think a better solution than privacy policies which almost nobody understands would be to have a site which permitted digg-like privacy ratings for different sites, because fundamentally it's a question of how much or how little you trust a given organisation with your data or financial transaction. ebay has this sort of trust mechanism. If enough users voted it would then become clear which companies were more or less trustworthy. If there were to be an element of legislation it might be that sites must link to the privacy rating site so that their current trust rating is displayed. The big flaw with this otherwise cunning plan though is that internet voting systems can usually be easily gamed.
Friday, September 18, 2009
Thursday, September 17, 2009
2050
An amusing romp into the hypothetical world of 2050 and its social, political and technological issues. Although many aspects of this future are debatable I think it's a more realistic view than some being promulgated by popular futurists today.
In this future technology has not solved all problems, and this isn't a singularitarian techno-utopia. People in the wealthier countries are living longer, but certainly not indefinite, life spans. Those elsewhere have been suffering from resource wars and environmental disasters which have resulted in mass migrations. Most of the population, apart from a few elites, are implanted with an "oracle" chip which is the ubiquitous means of providing communication, information and entertainment. There is also a world government. Colonies have been established on the moon and mars, but they're small and considered excessively expensive and unsustainable.
In this future technology has not solved all problems, and this isn't a singularitarian techno-utopia. People in the wealthier countries are living longer, but certainly not indefinite, life spans. Those elsewhere have been suffering from resource wars and environmental disasters which have resulted in mass migrations. Most of the population, apart from a few elites, are implanted with an "oracle" chip which is the ubiquitous means of providing communication, information and entertainment. There is also a world government. Colonies have been established on the moon and mars, but they're small and considered excessively expensive and unsustainable.
Lunar rovers
Maybe they could join several of these together and make a lunar centipede, or even a lunar millipede.
Wednesday, September 16, 2009
Overconfidence is not always a good thing
AGI can be achieved within ten years, according to Itamar Arel.
Well this all sounds like very high level flim-flam, and the guy seems to be making some very confident assertions, such as "will possess much of what we associate with consciousness". I've heard statements like these innumerable times on AI related forums. So he can talk the talk, but what does his walk look like? Reading a few of his papers this looks like fairly conventional reinforcement learning stuff. I'm not seeing anthing here that really seems to justify that a breakthrough has taken place, or is about to take place within a few years worth of engineering refinement combined with some appropriate level of funding. Nor am I especially persuaded that just taking some classical neural net and RL algorithms and hardware accelerating them is necessarily going to produce an AGI.
Overconfidence is a classic trap for AI researchers. If you're going to make big claims that AGI - that is, something with human-rivaling general intelligence skills - is going to happen within a few years you really need to be prepared to give a detailed breakdown of why you think that's likely to happen. If not, then it's just yet more hand-waving and vague references to Moore's law, and rest assured that folks will resurrect old forum postings or videos a few years down the line which make you look silly if the grand prophesies don't pan out.
Reinforcement learning would seem to be only part of what animals and humans do. If you look at the capabilities of humans and also some birds, novel tasks can be accomplished without trial and error in a one-shot operation. It might be argued that what's going on in these cases is an internalised version of reinforcement learning performed within a simulated environment, but at present whether this really occurs seems to be unknown.
Well this all sounds like very high level flim-flam, and the guy seems to be making some very confident assertions, such as "will possess much of what we associate with consciousness". I've heard statements like these innumerable times on AI related forums. So he can talk the talk, but what does his walk look like? Reading a few of his papers this looks like fairly conventional reinforcement learning stuff. I'm not seeing anthing here that really seems to justify that a breakthrough has taken place, or is about to take place within a few years worth of engineering refinement combined with some appropriate level of funding. Nor am I especially persuaded that just taking some classical neural net and RL algorithms and hardware accelerating them is necessarily going to produce an AGI.
Overconfidence is a classic trap for AI researchers. If you're going to make big claims that AGI - that is, something with human-rivaling general intelligence skills - is going to happen within a few years you really need to be prepared to give a detailed breakdown of why you think that's likely to happen. If not, then it's just yet more hand-waving and vague references to Moore's law, and rest assured that folks will resurrect old forum postings or videos a few years down the line which make you look silly if the grand prophesies don't pan out.
Reinforcement learning would seem to be only part of what animals and humans do. If you look at the capabilities of humans and also some birds, novel tasks can be accomplished without trial and error in a one-shot operation. It might be argued that what's going on in these cases is an internalised version of reinforcement learning performed within a simulated environment, but at present whether this really occurs seems to be unknown.
Another BBC snafu (aka the hunt for the missing copy editors)
From this article.
"Critics of the BBC's request say that, because open source firms do not use licenses, it would be harder for them to obtain the necessary permissions from the BBC, effectively pushing them out of the market."I think it's fair to say that open source firms are obsessed by licenses and licensing.
Tuesday, September 15, 2009
Printing the statue of liberty
Another baby step towards the manufacturing of items in the home, using the makerbot. I think the breakthrough point will come when a machine like this can make something which is undisputably of economic value - such as printing parts which can be assembled into a chair that you can sit on.
Monday, September 14, 2009
Motion models
Motion model for a mobile robot, which is part of a unit test. This is in preparation for turning the DP-SLAM code into a single standalone component (maybe a SLAM server), which can then be used with the Surveyor SVS, and other robots. I'm attempting to make the motion model cope with full 6D poses, so that it can also be used on the Rodney humanoid.
This is open loop, with the uncertainty increasing over time - hence the expanding cloud of poses.
Here uncertainty ellipses are shown.
This is open loop, with the uncertainty increasing over time - hence the expanding cloud of poses.
Here uncertainty ellipses are shown.
Omnidirectional vision
A description of obstacle avoidance using Pi Robot. The results look ok.
I've read about omnidirectional vision, but never really tried it because typically a very oddly shaped mirror is used, which are impossible to source and presumably hand made. What I liked about this robot is that they're just using a reflective christmas decoration as the mirror. Although this doesn't have the same shape used by other omnidirectional vision systems, it seems to work ok.
You can see the ground plane features which are used in this video.
The main problem with this edge based system is that it assumes a homogeneous ground plane appearance. If there were multiple carpets with patterns, or even just sunlight from the windows casting shadows on the floor then this strategy wouldn't work. However, with the addition of wheel encoders and some extra tinkering this probably could be made to work much better, and possibly also yield 3D information about the environment.
Wheel encoders can be used to get an estimate of how far the robot has moved, using a motion model. Because of the spherical shape of the mirror we have an idea of the pan and tilt direction which rays of light are coming from, and by performing data association on the edge features between successive frames as the robot moves a disparity figure can be found. With the mirror at a given height above the ground and knowing the approximate movement distance of the robot it should be possible to classify edge features by their disparity as belonging to the ground plane or not (assuming that the ground is flat, which will be true most of the time).
With more work you can probably obtain ranges for any observed features which can be tracked from one frame to the next, and begin to construct a 3D map of the environment. This would be highly dependent upon the accuracy of the wheel encoders, but if a suitable SLAM method is used together with a simple sensor model (it would be much simpler than for stereo vision) the uncertainty could be handled.
I've read about omnidirectional vision, but never really tried it because typically a very oddly shaped mirror is used, which are impossible to source and presumably hand made. What I liked about this robot is that they're just using a reflective christmas decoration as the mirror. Although this doesn't have the same shape used by other omnidirectional vision systems, it seems to work ok.
You can see the ground plane features which are used in this video.
The main problem with this edge based system is that it assumes a homogeneous ground plane appearance. If there were multiple carpets with patterns, or even just sunlight from the windows casting shadows on the floor then this strategy wouldn't work. However, with the addition of wheel encoders and some extra tinkering this probably could be made to work much better, and possibly also yield 3D information about the environment.
Wheel encoders can be used to get an estimate of how far the robot has moved, using a motion model. Because of the spherical shape of the mirror we have an idea of the pan and tilt direction which rays of light are coming from, and by performing data association on the edge features between successive frames as the robot moves a disparity figure can be found. With the mirror at a given height above the ground and knowing the approximate movement distance of the robot it should be possible to classify edge features by their disparity as belonging to the ground plane or not (assuming that the ground is flat, which will be true most of the time).
With more work you can probably obtain ranges for any observed features which can be tracked from one frame to the next, and begin to construct a 3D map of the environment. This would be highly dependent upon the accuracy of the wheel encoders, but if a suitable SLAM method is used together with a simple sensor model (it would be much simpler than for stereo vision) the uncertainty could be handled.
Sunday, September 13, 2009
Armadillos in space
A competitor in the Northrop Grumman Lunar Lander Challenge. This is all about balance and control. The machine is balancing on a single point, rather like trying to balance a full beer bottle upside down on the end of your finger, with the angle of thrust changing direction to maintain stability.
Saturday, September 12, 2009
Can surveillance mania be stopped?
An organisation called Freedom Not Fear aims to stop the surveillance mania. Here in the UK we're as manic as it gets, but can the wave of surveillance technologies and their uses really be stopped?
I think most likely the answer is "no", and this is because:
A good example of equiveillance was the death of a newspaper salesman at the G20 protests. Had it not been for a few individuals with video cameras (probably on mobile phones) I think that case would never have come to light, and the official story of death by natural causes would have been believed.
I think most likely the answer is "no", and this is because:
- Certainly in the UK, and possible elsewhere too, surveillance seems to be explicitly welcomed by most people, usually because it provides them with a sense of safety. That is, safety from criminals, terrorists, or just generally unconventional behavior.
- Most people believe that the surveillance technologies will only be used against bad people doing bad things. They have an naive notion of what qualifies as "bad", and don't believe that they could ever fall into that category themselves.
- In countries with a long history of democratic rule there is no historical memory of forms of governance which are less than benevolent. For example, in such states people are not occustomed to being persecuted on the grounds of ethnic identity, religious affiliations or political opinions.
- Surveillance technologies are cheap, and easily available. It doesn't cost a company or a government agency very much money to install and maintain, compared to other expenditures.
- Increasingly, integration of IT systems is becoming easier, such that data sharing takes little effort.
- Moore's law means that it's cost effective to store and process what until recently would have been regarded as unfeasibly large volumes of data.
- Data mining technologies are becoming available, to trawl through large data sets and uncover possibly salient patterns of activity. There's an entire business model right there.
- From the government perspective, greater knowledge of the daily activities of citizens opens up opportunities for taxation aimed at controlling behavior at a personal scale. For example, levying "carbon" taxes depending upon how many times a person goes shopping by car, and so on. There is also the alure of potentially near perfect enforcement of some existing rules and taxes - something which has not been practically achievable in the past.
A good example of equiveillance was the death of a newspaper salesman at the G20 protests. Had it not been for a few individuals with video cameras (probably on mobile phones) I think that case would never have come to light, and the official story of death by natural causes would have been believed.
Counting coffee cups
An interesting talk about the Stanford AI Robot (STAIR). I think they use edge detection followed by local symmetry estimation to find regions of interest which can be foveated. This is quite a Brooksian approach, which doesn't attempt to understand very much about the structure of the environment, other than perhaps a 2D map created using a laser scanner, but still manages to achieve a degree of competence on various kinds of tasks.
Demos of this kind can however be misleading, and often what's not mentioned is how repeatably the tasks can be performed. If the robot can fetch the desired object or carry out an inventory of cups with 90% accuracy then that might be good enough for some commercial applications.
In my opinion the Brooksian approach is only going to stretch so far, and for more complex tasks something resembling a 3D model, or partial model, of the robot's surroundings is going to be a requirement.
Demos of this kind can however be misleading, and often what's not mentioned is how repeatably the tasks can be performed. If the robot can fetch the desired object or carry out an inventory of cups with 90% accuracy then that might be good enough for some commercial applications.
In my opinion the Brooksian approach is only going to stretch so far, and for more complex tasks something resembling a 3D model, or partial model, of the robot's surroundings is going to be a requirement.
Friday, September 11, 2009
Ideas for Facebook quizzes
Some suggestions for facebook quizzes:
What command line program are you?
Which assembler mnemonic are you?
What extremophile are you?
Which semiconductor material are you?
What GEM desktop icon are you?
Which ediacara biota morphology are you?
Which intuitionistic type theory connective are you?
Which physical constant are you?
What type of vacuum tube are you...?
Which software license are you?
Which Darwin awards winner are you?
Which American revolutionary war pamphleteer are you?
What electron shell do you belong to?
Which food scare are you?
What cell organelle are you?
Which systems theorist are you?
What command line program are you?
Which assembler mnemonic are you?
What extremophile are you?
Which semiconductor material are you?
What GEM desktop icon are you?
Which ediacara biota morphology are you?
Which intuitionistic type theory connective are you?
Which physical constant are you?
What type of vacuum tube are you...?
Which software license are you?
Which Darwin awards winner are you?
Which American revolutionary war pamphleteer are you?
What electron shell do you belong to?
Which food scare are you?
What cell organelle are you?
Which systems theorist are you?
Thursday, September 10, 2009
Bundestrojaner interview
An interview with the author of the soon to be GPL'd Skype trojan, which sheds some tiny chink of light upon the murky associations between government agencies and malware writers.
Since this is going to be made open source I think it's fair to assume that it's no longer in use by those who originally sponsored its development, and has already been replaced by something else.
In principle the most efficient way to propogate government malware to citizens, enabling VoIP interception of the majority of the population, might be to incorporate it into a commonly viewed government related web site as part of a closed source media player such as flash, realplayer or iplayer. Deep packet inspection might also be used, but that comes with the problem of identifying which data is relevant out of a large data stream and also having to deal with encryption and varying transmission formats. Taking into account various anecdotal stories that I've heard it sounds as if malware is currently the preferred option for VoIP and other kinds of digital communications interception.
I think it is also much more possible for there to be associations between governments and antivirus companies than the interview dismissively suggests. I'm sure that companies, in Russia or elsewhere, would be more than happy to accidentally overlook certain pieces of malware in exchange for a sizable "subscription fee". No matter where you are in the world, money talks. I'm no expert on antivirus software, but a giveaway indicator that something like this might be going on would be if antivirus programs were designated for sale only within a restricted geographical area (i.e. a European version, a US version, an Australian version etc). Whether that happens or not, I don't know.
Since this is going to be made open source I think it's fair to assume that it's no longer in use by those who originally sponsored its development, and has already been replaced by something else.
In principle the most efficient way to propogate government malware to citizens, enabling VoIP interception of the majority of the population, might be to incorporate it into a commonly viewed government related web site as part of a closed source media player such as flash, realplayer or iplayer. Deep packet inspection might also be used, but that comes with the problem of identifying which data is relevant out of a large data stream and also having to deal with encryption and varying transmission formats. Taking into account various anecdotal stories that I've heard it sounds as if malware is currently the preferred option for VoIP and other kinds of digital communications interception.
I think it is also much more possible for there to be associations between governments and antivirus companies than the interview dismissively suggests. I'm sure that companies, in Russia or elsewhere, would be more than happy to accidentally overlook certain pieces of malware in exchange for a sizable "subscription fee". No matter where you are in the world, money talks. I'm no expert on antivirus software, but a giveaway indicator that something like this might be going on would be if antivirus programs were designated for sale only within a restricted geographical area (i.e. a European version, a US version, an Australian version etc). Whether that happens or not, I don't know.
Tuesday, September 08, 2009
AGI roadmap
Reading the AGI roadmap there is some great flowery language at the beginning:
"As we approach the twilight of the first decade in the twenty-first century..."The twilight!
"one may feel inclined to settle with the great achievements of man that shine forth..."Shine forth, man.
"the stunning marvel of the Internet..."Some of the internet is impressive. Many other parts are, shall we say, less stunning.
Ethics and autonomous systems
A report from the Royal Academy of Engineering points out that autonomous road vehicles are near to being technically feasible, and that preparations will need to be made to facilitate the change. It suggests that market forces should be the dominant factor in allowing driver safety systems and inter-vehicle communication to develop, but that at some point a centralised system should control the flow of traffic.
Such a centralised system might work, but you can see a precedent for this in rail travel. Railways were, at least in theory, centrally controlled in the UK up until the mid 1990s, with a single national timetable. The penalty for centralisation is inflexibility, since elaborate pre-planned timetables can be difficult to reschedule if some unforseen event occurs.
A compromise which I think avoids the political hot potato of who owns the roads or who takes charge of and/or regulates the central planning system, is to have autonomous vehicles cooperate locally with each other and have the flow of the entire system be an emergent property. With an emergent traffic flow system you get the benefits of cooperative behavior similar to that which might be seen under centralised control but also with flexibility for the entire system to adapt naturally to changes such as road works, fallen trees, accidents and so on. In an emergent system any driver could take manual control of their vehicle and the rest of the system would adapt around that decision. Even when autonomous driving becomes a mature technology there may be good safety reasons for having a manual override.
As a bit of speculation it might be imagined that drivers who choose to use old-fashioned manual control perhaps would be taxed (or tolled) at a higher rate than other road users, as a response to the reduced overall traffic flow efficiency which the decision causes. This would provide a social mechanism which encourages people to use autonomous driving as the default option, only resorting to manual operation in emergencies. There's an existing tax, called vehicle excise duty, which could perhaps be adapted to reflect the amount of flow disruption that is caused.
Another disruption which automous driving might cause is an end to the revenue stream which results from levying fines on drivers who exceed speed limits. Local government or policing institutions who currently rely upon this sort of income will need to find alternative ways of making money. The biggest disruption though will be to the insurance industry, which currently makes a great deal of money out of the unreliability of human drivers. These industries aren't going to react favourably to being automated out of existance, but I expect that the move towards greater vehicle autonomy over the next few decades will be sufficiently gradual to allow them to wane slowly.
Such a centralised system might work, but you can see a precedent for this in rail travel. Railways were, at least in theory, centrally controlled in the UK up until the mid 1990s, with a single national timetable. The penalty for centralisation is inflexibility, since elaborate pre-planned timetables can be difficult to reschedule if some unforseen event occurs.
A compromise which I think avoids the political hot potato of who owns the roads or who takes charge of and/or regulates the central planning system, is to have autonomous vehicles cooperate locally with each other and have the flow of the entire system be an emergent property. With an emergent traffic flow system you get the benefits of cooperative behavior similar to that which might be seen under centralised control but also with flexibility for the entire system to adapt naturally to changes such as road works, fallen trees, accidents and so on. In an emergent system any driver could take manual control of their vehicle and the rest of the system would adapt around that decision. Even when autonomous driving becomes a mature technology there may be good safety reasons for having a manual override.
As a bit of speculation it might be imagined that drivers who choose to use old-fashioned manual control perhaps would be taxed (or tolled) at a higher rate than other road users, as a response to the reduced overall traffic flow efficiency which the decision causes. This would provide a social mechanism which encourages people to use autonomous driving as the default option, only resorting to manual operation in emergencies. There's an existing tax, called vehicle excise duty, which could perhaps be adapted to reflect the amount of flow disruption that is caused.
Another disruption which automous driving might cause is an end to the revenue stream which results from levying fines on drivers who exceed speed limits. Local government or policing institutions who currently rely upon this sort of income will need to find alternative ways of making money. The biggest disruption though will be to the insurance industry, which currently makes a great deal of money out of the unreliability of human drivers. These industries aren't going to react favourably to being automated out of existance, but I expect that the move towards greater vehicle autonomy over the next few decades will be sufficiently gradual to allow them to wane slowly.
Monday, September 07, 2009
More crackpottery from Redmond
To be a Microsoft expert these days it seems that you're required to display a certain level of ignorance with regard to Linux. For example:
I've been using Ubuntu for the last few years, and the "hundreds of updates per month" is certainly false. If I had been continually bombarded with updates that would have been offputting, but that just doesn't happen. Typically there might be a few updates every week or so, but not hundreds.
On Linux systems all updates are optional. Unlike on Windows you're not forced to install certain updates (such as Windows Genuine Advantage).
And as far as consuming a lot of time to maintain goes, Windows is definitely the high maintenance option with its antivirus, spyware/adware and other security circus applications. Often these applications don't add a great deal of security, bombard the user with confusing popups and can add an enormous drag on performance. As the neigbourhood "computer person" I've seen some real horrors in terms of security software, with naive users being seriously exploited by fearmongering companies pushing "security" products. When using Linux you don't need to deal with any of that nonsense.
I havn't used Windows 7, but I have used just about every other Windows version, including Millenium Edition (which eventually self destructed on me) and even Windows 3.0. Current versions of Linux compare well to XP and especially to Vista on all of these factors with the exception of hardware. Availability of hardware drivers is much less of an issue than it once was (particularly for webcams), with the main offenders being graphics card manufacturers. On the interface design and personalization aspects I think Linux has already overtaken the Windows 7 demos that I've seen, with all sorts of 3D effects and themes. Effects like these are pretty standard now, whereas the Windows 7 interface looks fairly clunky and not much of an advance upon Vista.
One really obvious area where Ubuntu needs to improve is the default desktop appearance. Most people, including myself, think that the brown theme is very uninspiring. For new users who aren't really all that interested in the underlying technology, especially when coupled with the above disinformation recited verbatim by a retail store assistant, the brown theme may be sufficiently offputting to choose Windows 7 instead.
I've been using Ubuntu for the last few years, and the "hundreds of updates per month" is certainly false. If I had been continually bombarded with updates that would have been offputting, but that just doesn't happen. Typically there might be a few updates every week or so, but not hundreds.
On Linux systems all updates are optional. Unlike on Windows you're not forced to install certain updates (such as Windows Genuine Advantage).
And as far as consuming a lot of time to maintain goes, Windows is definitely the high maintenance option with its antivirus, spyware/adware and other security circus applications. Often these applications don't add a great deal of security, bombard the user with confusing popups and can add an enormous drag on performance. As the neigbourhood "computer person" I've seen some real horrors in terms of security software, with naive users being seriously exploited by fearmongering companies pushing "security" products. When using Linux you don't need to deal with any of that nonsense.
I havn't used Windows 7, but I have used just about every other Windows version, including Millenium Edition (which eventually self destructed on me) and even Windows 3.0. Current versions of Linux compare well to XP and especially to Vista on all of these factors with the exception of hardware. Availability of hardware drivers is much less of an issue than it once was (particularly for webcams), with the main offenders being graphics card manufacturers. On the interface design and personalization aspects I think Linux has already overtaken the Windows 7 demos that I've seen, with all sorts of 3D effects and themes. Effects like these are pretty standard now, whereas the Windows 7 interface looks fairly clunky and not much of an advance upon Vista.
One really obvious area where Ubuntu needs to improve is the default desktop appearance. Most people, including myself, think that the brown theme is very uninspiring. For new users who aren't really all that interested in the underlying technology, especially when coupled with the above disinformation recited verbatim by a retail store assistant, the brown theme may be sufficiently offputting to choose Windows 7 instead.
Saturday, September 05, 2009
Can the A-Prize be won?
Listening to the Biota Live podcast the "A-Prize" organised by a group called the Lifeboat Foundation comes in for particularly stringent analysis. Having a look at the relevant web page it does indeed appear that the A-Prize might be so vaguely defined that, somewhat like the Longitude prize - it might be incapable of actually being won.
From the A-Prize site:
In summary, the prize criteria do seem sufficiently vague that winning could be indefinitely debated and defered. Also it's unclear what bounty the winner would actually acrue, other than the satisfaction of having made the achievement. Are there gradations of prizes, as for example is the case in the Loebner competition?
Ignoring the merits or otherwise of this particular prize - and I should say that I have no particular interest in or intention to enter into the A-Prize - I think prizes can provide a useful motivation to inspire people to solve problems or create new inventions. I'd certainly like to see more AI or Alife related prizes. But entrants for technology prizes should always check that there are some definite and reasonably clear criteria by which winning and losing can be delineated.
Incidentally, anyone with an interest in animats, might like to check out the CCNR talks. Animat studies are an interesting area at the intersection between biology, simulation and robotics.
From the A-Prize site:
"It is unlikely that a fully functional Synthetic Nonbiological Life Form (a.k.a. an Animat — see Type 4 below) will come into being in the near future."This seems curious because I have a number of books in the From Animals to Animats series, which almost exclusively describe fully synthetic creatures of various kinds and implemented at various levels of abstraction.
"Therefore, the Carbon BarrierTM is defined as the moment in the evolution of human Molecular Engineering when we first create an organism that must execute at least one synthetic nonbiological operation in order to complete its life cycle. The person or group that verifiably creates such an organism with an emphasis on the safety of the researchers, public, and environment, OR the person or group that provides information leading to the discovery that such an organism has been created will win the A-PRIZE."Ok, so by this I assume that we're taking an existing piece of biological wetware, hacking a chunk out of it, and replacing the missing component with some artificial system. This seems pretty vague though. Does the organism need to be a single cell, or can it be multicellular? If I were to have some organ of my body replaced with an artificial version - say a heart or a hippocampus - and that permitted me to complete my life cycle, would that mean that the inventor of the artificial organ could claim the prize? Then down at the opposite end of the scale, what if I were to insert an arbitrary RNA sequence into a virus and have it complete its life cycle - would that count?
In summary, the prize criteria do seem sufficiently vague that winning could be indefinitely debated and defered. Also it's unclear what bounty the winner would actually acrue, other than the satisfaction of having made the achievement. Are there gradations of prizes, as for example is the case in the Loebner competition?
Ignoring the merits or otherwise of this particular prize - and I should say that I have no particular interest in or intention to enter into the A-Prize - I think prizes can provide a useful motivation to inspire people to solve problems or create new inventions. I'd certainly like to see more AI or Alife related prizes. But entrants for technology prizes should always check that there are some definite and reasonably clear criteria by which winning and losing can be delineated.
Incidentally, anyone with an interest in animats, might like to check out the CCNR talks. Animat studies are an interesting area at the intersection between biology, simulation and robotics.
Friday, September 04, 2009
Subscribe to:
Posts (Atom)
