On the busses

16 cameras on a single London bus. Apart from the political dimensions (I think we're now well advanced into the era of ubiquitous surveillance), it's a highly visible indicator of the power of Moore's law, with the lowering cost and increasing performance of digital imaging and digital storage technology. Probably a couple of decades ago even having a single camera on a bus would have been a very expensive installation, with a fair amount of maintenance in terms of swapping video tapes.



I think historians will look back at the transition into a surveillance society with interest. When you examine the details there was really no decisive point at which parliament decreed that CCTV cameras should be erected in every street corner or inside busses. It was just something which happened gradually over the course of a couple of decades in a mostly bottom up fashion, driven by what I think is the especially British fear of dissorderly or unconventional behavior.

It's still true that most surveillance systems are privately owned, but I think it's only a matter of time before camera surveillance too, like other forms of digital information before it, heads into "the cloud". In a world where high speed wireless networks are available everywhere centralising the storage and processing of digital imagery will make good economic sense in the never ending drive for greater efficiency.

How long will the Singularity University last as an organisation?

The poll results are in, and the wisdom of the crowd has spoken.



The results are:

Less than 6 months 10%
1 year 10%
1-2 years 40%
2-5 years 30%
5-10 years 10%
10-20 years 0%
More than 20 years 0%

The consensus of opinion seems clear. According to the pollsters the Singularity University is only likely to last for around 2 years in its current form. After that time I expect that it will either be renamed and repurposed in some more focussed way, or disbanded altogether.

Interestingly, nobody expects the organisation to last for longer than a decade, as would be the case for a maturing and increasingly recognised scholarly body. However, this could be open to some interpretation. It might be that some people believe that a technological singularity will occur in ten years or less, in which case the Singularity University is rendered obsolete by its own success. In my opinion though this probably isn't what the poll is showing.

The future of energy

Some pretty awesome statistics on how much solar power would be required to fulfill the world's current and future needs can be found here. I've heard people like Ray Kurzweil talk about how solar energy could be more than sufficient to power the future, but the diagram shows the same kind of thing more easily in a visual way.

This assumes that the solar panels are 20% efficient at converting sunlight to usable electricity, which seems reasonably conservative. The best quality panels which you can buy today are about 15-20% efficient, and there are also reports of technologies in development which might push this up much further into the 30-40% range.

"According to the United Nations 170,000 square kilometers of forest is destroyed each year. If we constructed solar farms at the same rate, we would be finished in 3 years."

"The Saharan Desert is 9,064,958 square kilometers, or 18 times the total required area to fuel the world."

So even using a conservative estimate, which is likely to be exceeded as the technology is refined, it looks as if a future powered substantially by the sun is eminently doable provided that there is sufficient political motivation. Even if the idea of land based solar farms proves to be unpopular it would still be possible to build floating solar factories on the seas, with the added advantage that you can concentrate them around the equator to maximise efficiency and also use them to produce hydrogen by electrolysis at the same time.

Constructing solar farms in desertified areas of the world would make the most sense, since this land is usually neither of commercial nor wildlife interest. Large areas covered by solar panels would also mean that the energy which normally heats the ground is blocked, so the surrounding region and its atmosphere may be cooled, perhaps to the extent that former desert becomes suitable for farming again.

Silly jingoism from iRobot

A strange, and it would appear from the waving guns and flags highly jingoistic, video from iRobot. Presumably they're hoping that this gives their robot a sexy image amongst impressionable young american males. The music is actually ok, but from the rest of it I get the impression that some marketing person went way overboard.

Standard data sets for mobile robots

One thing which I've complained about consistently over the last decade or more is the lack of repeatability of academic artificial intelligence research. If you're claiming to be doing anything resembling real science empiricism dictates that any results you come out with should be independently verifiable. Otherwise the graphs, diagrams and tables shown in research papers could be entirely fabricated, or (as I suspect is often the case) merely depict cherry picked data showing a good looking result which is not actually statistically representative of the true performance of the system.

In the area of robotics a certain amount of non-disclosure is understandable, since technical advances may eventually lead to patentable inventions of commercial value, but one thing which has really been lacking are data sets taken during experiments, which can be used to independently validate the effectiveness of particular algorithms, or fairly compare the benefits of one type of system against another.

A new web site called Rawseeds seeks to deal with this problem, by providing standard data sets for robots with various kinds of sensors. So you have laser scans, monocular and stereo/trinocular vision, IMU, wheel odometry and so on for mobile robots within various kinds of environment. I think this is a marvellous idea, and introduced not before time. With the GROK2 robot it must have taken me over a year of construction (and other fiddling about) just to get to the stage where I can even begin to think about gathering useful data sets, and this must be a common problem holding back progress in this area. If a researcher who doesn't possess an elaborate or expensive robot platform, or the skills needed to build one, can download data sets and test algorithms without the hardware then I think this is certainly going to improve the potential for progress.

Open source code commenting

The peak desnsity is around 20%, or one line of comments in every five lines. According to this there are some software projects with more than 80% comments. In that case you might as well just go the whole hogg and write a book instead of a piece of software. More graphs can be found here.

Matrioshka

An interesting speculative article on computational intelligence at the limits of physics is proposed by Robert Bradbury. The Matrioshka brain consists of a set of large solar powered computers orbiting a star at several different distances - hence the name derrived from the famous Russian dolls. It's hypothesised that civilisations who have advanced to a post-biological stage would be consitituted by such machines, and that their absorbtion of solar energy might explain the apparent deficit of matter within galaxies by dimming or blocking entirely the light from the host star.

It's a science fictional idea, but entertaining to think about nontheless. Here I am mucking around with robots and software, and the universe around me might actually be completely swamped with AIs, like microbes in a pond.

Forbidden sharing

Probably the most intelligent commentary on the proposed "cut off" of file sharing activity can be found here. The commentator in a little video clip in a recent BBC article I think didn't stress in sufficiently clear terms that what's being proposed would be unfeasible from a technical point of view.

The proposal, made by the UK government, is basically to "cut off" the internet connection of anyone found sharing files over peer-to-peer networks, in order to prevent violations of copyright. Unfortunately not all P2P activity is illegal. For instance BitTorrent is used to distribute versions of Linux and sites like Jamendo use it to legally distribute music under creative commons licenses.

So from an ISP's point of view how can you tell whether a given data stream contains illegal material? To do that would involve knowing the format of the data being transmitted and then perhaps comparing chunks of it to chunks of known music or movie files - a process similar to virus scanning or speech recognition. Any Windows user knows how arduous a full system virus scan can be, and for ISPs to do this on every internet connection would require significant investment in hardware and software.

Such scanning might be impractical and prohibitively expensive to implement now, but assuming that Moore's law continues for a while longer perhaps in five years time it might be feasible. Unfortunately the problem is more difficult than implementing something akin to virus scanning on a gigantic scale. I'm not an expert on the latest developments in P2P networks, but I would expect that at least some of them encrypt their data transmissions. Once encryption enters into the equation - even if it's not very strong encryption - then any hope of being able to detect illegal files moving over networks on a national scale and in a reasonably cost effective way vanishes entirely. They could try to infiltrate the P2P systems by implementing bogus clients, but unless such clients are also actively sharing files they're likely to be blocked or assigned a low rating, since these things tend to operate on a reputation basis.

So if ISPs are forced through legislation to start cutting off their customers what will they do? I think they'll do the only thing that is within their current capabilities and begin blocking systems such as BitTorrent entirely, regardless of whether the content being sent over them is illegal or not. This would mean that many legal activities would also be blocked too, which would be very counterproductive.

Looking beyond the next few years a regime under which all copyrights are rigerously and automatically enforced through some type of scanning system would be a pain, but such enforcement may provide extra efficacy for open source licenses and actually help creative commons media to compete more aggressively against traditionally copyrighted media. So the ultimate irony is that the measures being called for - if ways can be found to implement them - could end up hastening the demise of old business models.

Examining the runaway paperclip scenario

I was recently watching this video, in which Eliezer Yudkowsky talks about the runaway AI scenario. I've seen him make this argument many times, and in summary it's something like:

We don't want to build an AI which fills the universe with paperclips.

There appears to exist a gigantic theoretical oversight somewhere here. How would the paperclips actually be produced?

Filling the universe with any kind of monoculture seems like a fairly unintelligent thing to do, but suppose for sake of argument I was to construct an autonomous paperclip manufacturing machine and start it going by winding one of those hand cranks so often used on early 20th century automobiles. Would such a machine really pose a serious threat to the universe requiring well educated commentators to lose sleep and dedicate entire careers to contemplation of the wider existential implications?

Suppose that my machine is modestly high tech. It could for instance mine iron ore from the soil using some incredibly vaguely defined nanotechnological paraphenalia. To mine minerals at any spatial scale requires firstly that you have somewhere to store the stuff you've extracted, and secondly that you have somewhere to dump the stuff which you don't want. Obviously if the machine were to dump waste products at the same location as the mining was taking place the unwanted buildup would soon lead to frustrations and reduce the efficiency of operations. One way to avoid efficiency-depleting deposits might be to fling soil out of the back of the machine, and have the machine itself parambulate slowly across the terrain - essentially performing open cast mining. This might be ok if the terrain in question was uniformly rich in iron deposits, but this is rarely the case. Concentration of minerals at specific geological sites would mean that the aforementioned flinging strategy might not suffice to dispose of waste materials for very long.

Already, our superintelligence is facing an increasingly accumulating heap of trouble, even before it has arrived at the enviable position of being able to manufacture its very first paperclip. But superintelligences don't rest for long. A secondary plan, which performs the waste disposal job more efficiently would be to have trucks into which soil can be loaded. The trucks could then carry the waste soil to more distant locations where it could be dumped. Unfortunately in the classic runaway scenario the machine is unaided by human workers, and is instead some completely autonomous device, so the trucks need to themselves be manufactured before they can play any non-trivial role in this whole drama. In the initial stage of development therefore the machine needs to concentrate upon building infrastructure rather than directly carrying out its primary goal (the making of paperclips). Building infrastructure takes time, energy, and above all planning permission from the relevant authorities. Even if the requisite paperwork is fast-tracked - or waived entirely - we still need energy to move atoms from one place to another, and in a terrestrial environment the maximum rate of movement is going to be constrained by a variety of physical factors such as friction and gravity - no matter what kind of transportation technology is actually used.



So, although automated construction of mining facilities could proceed somewhat expeditiously a variety of physical constraints will mean that the rate of manufacture is going to be limited to the sorts of time scales where humans would be likely to notice what was going on and possibly intervene to halt the flagrant violation of local planning regulations. For instance, it seems unlikely that the laws of physics would permit a large open cast mining operation to be set up within 60 seconds, or even a few hours, if we assume that the initial paperclip machine is a metre or two in diameter and that it's using the most advanced nanotechnology available.

Perhaps I'm missing something, but it just seems unlikely that anything resembing the paperclip scenario could occur without the active collaboration of humans in the process and a good deal of manufacturing infrastructure already in place. Such scenarios don't appear to have been thought through in sufficient detail.

Apologising to Turing

Will it be possible to get an official apology for the persecution of Alan Turing as a homosexual in the 1950s? Maybe. Maybe not. It depends upon how fashionable Turing is considered to be, and whether the government judges places like Bletchley Park to be bringing in significant revenue from tourism.

Under normal conditions it's extremely difficult to get an official apology for anything, even if the event in question was an unambiguous transgression of justice. If I were a whitehall bureaucrat my opinion would probably be that Turing was just a single high profile individual, and that there must have been many other less illustrious folk who were persecuted in the same manner prior to the late 1960s. Many of those people will still be living, and if an apology is given that might open the floodgates to compensation claims. In legal terms I think an apology counts as some admission or indication of liability.

Outside of the Artificial Intelligencia, mathematicians and tech geekdom actually I think Alan Turing is not all that well known. Probably the average man on the Clapham Omnibus might not know who Turing was. So the chances of apology are probably slim, but still there's nothing to be lost by trying.

Trees come out of the air

True names

A fun episode of Skeptoid uncovers which names associated with particular products are real or merely marketing inventions. Many of these are uniquely American, so I've never heard of them anyway.

This got me thinking. Did Mr Kipling actually exist, and if so did he really make exceedingly good cakes? Alas, according to Wikipedia Mr Kipling:

"exists in the hearts and minds of Manor Bakeries' employees and all cake lovers, but not in fact as a real person"

I'm gutted. And to think all this time I had been harbouring a rustic idyl of Mr Kipling as a kindly old gent baking cakes with cherries on top, then force-feeding them to an assortment of grandchildren in a slightly out of focus 1980s TV advert.

Software patent snafu

Further proof, if it were needed, of just how much of a bad idea patenting software really is. It's the ultimate irony that Microsoft - probably the biggest proponent of software patents in the tech industry - has been stung by a patent troll to the tune of something in the region of $40 million in damages. Just to add insult to (financial) injury, the patent in question is some utterly trivial XML quirk involving storing "custom" (presumably non-xml formatted) content inside an XML document.

More artwork

This is entitled "Protohuman unconference light refreshments"

Another robotics web site

Another new robotics related web site, showcasing various academic research going on around the world. I quite like the NREC container handling system, which appears to be a 3D version of the sort of automation I did in the glass industry, except with potted plants instead of glass bottles.

Laws of war

Of course the notion of "laws" applying to warfare, given the effects and events which war produces, seems nonsensical. But be that as it may, there's another quite lengthy article discussing the use of military robots and how this might alter "the laws of war".

If anything, the article understates the stupidity of likely robotic combat operations within the foreseeable future. An example where a robot might be able to fire at a target autonomously would be if there was incoming fire from a particular location. This might be detected both from the sound of gunfire and from visible tracer flashes. But on a moving ground robot, which is basically just a small tank, listening for external sounds and filtering them out from internally produced noises of tracks, and engine, and other ambient noises might be difficult. Difficulties with automatic speech recognition provide some indication of how hard identifying particular noises can be. Also, flashing lights do not necessarily mean incoming gunfire. They could be transcient reflections from windows or water, electronic gadgets or christmas decorations. In an urban kind of environment information is much more ambiguous than you might at first think.

Even if someone is standing a few metres from the robot, and it scans them with a laser producing a millimetre precision 3D model of them, would it be able to determine whether they were carrying a gun, or a shovel, or merely just pointing their finger in a particular way? Is the object moving towards it a bomb, or a mother pushing a baby's pram? Unfortunately automatically identifying enemies at street level is a task which requires solving hard AI problems which have so far evaded solutions.

Then there's the issue of accountability. The article says:

"the human creators and operators of autonomous robots must be held accountable for the machines’ actions. (Dr. Frankenstein shouldn’t get a free pass for his monster’s misdeeds.) If a programmer gets an entire village blown up by mistake, he should be criminally prosecuted"

Unfortunately, software systems are rarely written by a single programmer these days. Often software is written by multiple people from multiple companies at multiple different levels (drivers, operating system, applications, databases, etc). In the instance of an accident identifying someone to blame at an individual level might be very difficult, especially considering the highly complex interactions between logic driven systems and the real world. If the software driving the robot is partly open source - and that's increasingly likely - there may be tens, hundreds or thousands of original authors. Open source systems may even become the norm within the military robot realm, due to the explicit waiver of claims that the software is "fit for any particular purpose" within the license - providing a degree of legal protection or at least plausible deniability for suppliers.

Given the technical difficulties combined with the political drive to reduce costs and allied casualties I think a more realistic scenario for autonomous military robots is that a specific geographic area is simply declared to be a "no go zone", with robots sent into the zone firing at anything with an approximately rectangular shaped infrared heat signature. Such a scenario would be considered to be absolutely ideal by despots seeking to prosecute genocidal campaigns of ethinic cleansing.

A history of blogging

More accidental art

Continuing the series of image processing mistakes as artform, this one is entitled "Appointment with an autocrat at a dilapidated concrete government building".

Virgin Galactic mothership

Growing demand for the Uber-Geek

Does having Linux related skills and knowledge make you more employable? This article suggests that it does.

I've had a couple of job interviews in the last two months, which was more than I had been expecting given the state of the economy. In the first interview they didn't really seem to care what software or operating systems were used, but that was a strange interview anyway for other reasons (spooks and "security" nonsense). In the second interview the technical manager and chief "software guy" displayed open hostility to the notion of using Linux or even open source software generally, despite the fact that their core business was embedded systems. They said that "under no foreseeable circumstances" would I be permitted to use Linux if I got the job, even if I used my own laptop so that they didn't need to supply any additional equipment or make any software changes to existing systems.

Luddites notwithstanding, it seems fairly clear which direction things are going and I expect that Linux/FOSS adoption will continue to increase in the next few years due to the "perfect storm" factors described in the article. It's a little like the olden days of the early 1990s, when there were some people saying "Windows and graphical user interfaces are the future" and others saying "all we need is DOS". There was a lot of support for DOS, and some developers/businesses clung tenaceously to DOS for purely ideological reasons even into the late 1990s by which time it was clearly obsolete. One of the first companies I worked for in the mid 1990s had been started because the founder had been forbidden from using Windows 3.1 as a development platform at his previous company, so had resigned and started his own business. Similar things were said back then about Windows 3.1 as used to be said about Linux a few years ago, such as "sure it looks nice and it's ok for hobbyists or casual users, but you could never use it for serious business applications".

So at present I'm not seeing any growing demand for Uber-Geeks, but I expect that once the economy begins to improve there will be.

More bird intelligence

Yet more examples of birds displaying intelligent behavior. The Aesop fable goes like this:

"A Crow, half-dead with thirst, came upon a Pitcher which had once been full of water; but when the Crow put its beak into the mouth of the Pitcher he found that only very little water was left in it, and that he could not reach far enough down to get at it. He tried, and he tried, but at last had to give up in despair. Then a thought came to him, and he took a pebble and dropped it into the Pitcher. Then he took another pebble and dropped it into the Pitcher. Then he took another pebble and dropped that into the Pitcher. Then he took another pebble and dropped that into the Pitcher. Then he took another pebble and dropped that into the Pitcher. Then he took another pebble and dropped that into the Pitcher. At last, at last, he saw the water mount up near him, and after casting in a few more pebbles he was able to quench his thirst and save his life.

Little by little does the trick."

This doesn't appear to be a series of random attempts, eventually chancing upon a successful strategy, as would be expected in trial and error learning. Instead, it resembles judgement and intuition. Whilst not all of the birds solved the problems a sufficient number of them did in order to rule out the possibility of this just being an isolated and exceptionally rare case of a super smart individial (a sort of avian Einstein).

Birds, being descended from dinosaurs, basically have a reptilian brain. So this suggests that at least some aspects of high level planning and spatial reasoning can be carried out without a neocortex.

DOOM as a psychological phenomena

Tim Tyler makes some comments about doomsters, or collapsitarians as I now like to refer to them.



Proclaimations of doom are extremely common throughout history, and the newfangled invention of the internet allows collapsitarian memes to propogate with unprecedented efficiency from one host to another. Studying the spread of particular flavours of doom, for example through YouTube videos, blog entries or twitter feeds, would itself be an interesting exercise in understanding how ideas spread between people. Are there particular words or combinations of words which trigger a certain kind of collapsitarian thinking? What psychological levers do collapsitarian gurus pull in order to convert gullable members of the public into fee paying followers?

Collapsitarianism is almost exclusively parasitic, and simply wastes people's resources in various ways. Often its chief proponents make a comfortable living out of selling books, DVDs, subscriptions to bulletins, survival paraphenalia or other forms of merchandise. Looked at from a different perspective, it's usually just another pyramid scheme, with a select few individuals at the apex profiting from the enterprise.

Does this mean that we can be entirely complacent about IMPENDING DOOM? Clearly, this strategy didn't work for the dinosaurs or for anaerobic bacteria that existed before the Oxygen Catastrophe. We can immediately discount religiously prophesied forms of doom out of hand as pure nonsense, but there are I think a few genuine threats to life on Earth (or at least human life), such as nuclear weapons proliferation, collisions with asteroids or comets, climate change (for example a new ice age, although not actually an existential threat, would cause big problems for human societies), excessive ecological damage, over-dependency upon non renewable resources and so on. But with the exception of nuclear weapons, most of these topics are not very glamourous and are complex or difficult to evaluate so can't easily be used as a way to shift merchandise or charge subscription fees.

David and Goliath

The Surveyor robot with stereo camera attached is dwarfed by GROK2.