Losing the oil endgame

A rather pessimistic video in which peak oil is predicted to occur in 2008/9. It should be noted that similar predictions have been made before and been wrong, but if no major new oil fields are being found then it's only a matter of time before supply peaks.

Real Oil Crisis

As the video points out there are a variety of alternatives and the main question is how quickly economies would be able to transition to these. Ideally the transition away from an oil based economy will be gradual, carefully managed by governments and happen in accordance with traditional market forces - somewhat like the transition from coal to oil or from horse power to steam power. In the worst case scenario powerful vested interests able to strongly influence the world's governments will cling tenaciously to their traditional sources of profit until the last possible moment, resulting in a relatively sudden and economically damaging energy shortage.

Little Brother

A new ebook by Cory Doctorow.

"It seems like the entire police and governance system in Britain has fallen in love with DNA swabbing, fingerprinting and videorecording everyone, on the off chance that someday you might do something wrong. In early 2008, the head of Scotland Yard seriously proposed taking DNA from fiveyearolds who display "offending traits" because they'll probably grow up to be criminals. The next week, the London police put up posters asking us all to turn in people who seem to be taking pictures of the ubiquitous CCTV spycameras because anyone who pays too much to the surveillance machine is probably a terrorist."

World oil supply

You can draw your own conclusions from this graph. A best fit line to the data would still give an increasing trajectory, although it looks like in the last three years production has plateaued. Common sense would dictate that production from a finite resource can't go on increasing indefinitely.

Listening to radio 4 during the week I overheard one government spokesman say that peak oil is estimated to occur in 10-15 years as an aside whilst talking about high fuel oil prices, although the opinions of politicians as we all know can be just as dodgy as any internet commentators.

Pyramids

A photo of my grandfather standing on an Egyptian pyramid during service with the RAF in WW2. Being such a nerd it immediately occurred to me that it probably will be possible in the not too distant future to identify the exact location and pose from which this photo was taken, using technologies similar to photosynth.

Is life a copyright violation?

An interesting Biota discussion about the origins of life can be found in this podcast. One observation is that given the estimated time elapsed since the supposed big bang and the time taken for life to progress from microbes to humans on Earth, life could have evolved several times and might be very ancient. This leads me to think of an alternative version of the Hitchhiker's guide to the galaxy, in which instead of being destroyed to make way for an intergalactic bypass the Earth is destroyed because it's an unauthorised copyright violation, with the Vogons being a kind of intergalactic RIAA enforcement agency.

"In summary, in spite of a number of simplifications and compromises necessitated by the limited available computing resources, we have demonstrated the spontaneous emergence of self-replicating computer programs, sexual reproduction, and evolutionary self-improving behavior among randomly created computer programs using a computationally complete set of logical functions" - John Koza

The idea that artificial life simulation might be used to investigate how life got started isn't a new one. I have a book by John Koza called "Genetic Programming", in which there is one chapter describing a simulation where self-replicating structures spontaneously emerge from an initially entirely random soup consisting of a few rules analogous to an artificial chemistry. A more interesting variant of this would be to replace the very abstract functions used by Koza and many other Alife researchers with functions representing real molecules and their known interactions (i.e. the rules of physics/chemistry). It might then be possible to search for niches in which self replication or autocatalysis spontaneously appears. The results of such simulations might be used to better inform SETI-like searches.

Autonomous vehicles: the road ahead

This is a nice video which explains quite clearly the trends in autonomous vehicle development and how it's likely to pan out over the next decade or two. The sensors recently used in the DARPA Urban Challenge all cost in the range of thousands of dollars, but if over time this comes down to hundreds then tens we can also expect to see them used on a smaller scale for autonomous vehicles (robots) inside businesses and homes. Interestingly digital imaging has already traversed this downward curve, with imaging sensors now being extremely cheap. If you can do the same thing with imaging that's possible with lasers (and I think you can) then we're already into the realm where autonomous navigation becomes viable for industrial/business use.

Ninja tunes

This weekend's sound track is Puffy64's Last Ninja II tribute.

OLPC goes over to the Dark Side

"Most importantly, the OS that OLPC ships should be one that embodies the culture of learning that OLPC adheres to. The culture of open inquiry, diverse cooperative work, of freely doing and debugging — this is important. OLPC has a responsibility to spread the culture of freedom and ideas that support its educational mission; that cannot be done by only offering a proprietary operating system for the laptops." - Ivan Krstić

It looks as if the one laptop per child effort has basically abandoned its open source operating system and is going over to Windows XP. Windows XP is not a bad operating system, and has a long track record, but I don't think it's especially suited to this sort of application which requires a very bare bones hardware specification. I think the OLPC processors are something like 433MHz, so XP is going to run like a snail on them.

Children using an OLPC with XP will suffer from the following problems:

• General security/reliability issues - what I refer to as "the security circus". Windows operating systems absolutely require some form of antivirus/antispam software if they're going to be connected to the internet and run a browser (which I believe the OLPCs are), and this can impose quite onerous responsibilities upon the user. Children aren't going to care or know the slightest thing about security, so I suspect these systems will soon get bogged down with all manner of malware. This is an unnecessary overhead for a system which is supposed to be simple and ultra reliable.

• An already slow machine is going to get slower over time. The ancient NTFS or FAT file systems become increasingly fragmented and performance noticeably degrades over time. Are OLPC users really going to be expected to install and run defragmentation programs? This isn't really an issue on GNU/Linux systems.

• The extra cost for the XP operating system is a substantial chunk of the overall cost of the system, potentially compromising the target price. $20 is nothing to the likes of you and me, but on a machine which is supposed to cost no more than $100 it's 20% of the price! For this type of application paying money for software licences when you don't absolutely need to do that is a waste of resources. Also don't forget that Microsoft can arbitrarily hike up the price at any time, or withdraw support for the OS entirely.
  

• The Windows operating system reduces educational opportunities. Since this is a closed source system there is no opportunity to find out how it's working or to make changes to it at a fundamental level. Admittedly the number of children who will become computer programmers is going to be small, but that small group will have a disproportionately large influence on the rest. Some kids are competent programmers by the time they're about 13/14, and so could potentially be used as a resource to help schools or local communities become self sufficient in terms of software needs, rather than having to rely upon outside support (with the associated costs).

For more inside information about the OLPC project and its woes see this blog post.

GROK2 motion buffering

Here I'm testing a series of movements inserted into something called a move buffer. In this test I send a single command to move forward, then whilst that's running I also send a few other commands via TCP/XML which are put into the queue. This ensures that the robot can receive commands (either from an onboard computer or possibly from one or more remote computers) and schedule them in a sensible way without getting confused. This sort of capability is standard issue on most industrial motion control hardware used to control robots or CNC machines.



Repeatability of movements looks generally good, although I would expect this when only moving over very short distances. This probably means that I won't need massive numbers of particles to model the pose uncertainty.

Things to be implemented soon are:

• The ability to pause and resume the motion, ramping down to zero speed then ramping up again to avoid jerky motion.

• The ability to merge multiple motion commands. So for example if several forward motion commands are requested the system joins these together into a single continuous motion rather than starting and stopping several times.

• Joystick control. This will simply push commands into the move buffer as appropriate.

You could roughly categorize these kinds of functions as being analogous to the operation of a cerebellum in animals.

It may be a dirty job, but some bot's gotta do it

"We are all in the gutter, but some of us are looking at the stars"

- Oscar Wilde

Jim Lynch talks about the Looj: a gutter cleaning robot made by iRobot. An interesting point about this is that just like the Roomba a robot like this could have been made and marketed at a reasonably economical price at any time within the last 25 years. The Looj is even simpler than a Roomba, being basically similar to a specialised kind of radio control car, so devices like this could have appeared in the 1970s (perhaps tethered rather than radio control). So why wasn't something like this available long ago? My guess would be that probably it was, but that whoever invented it took it to a company or a venture capital organisation and got laughed out of the room. It seems that for a robotic product to be successful not only the technology has to be in place at the right cost but also there needs to be a suitable cultural atmosphere within which these things aren't regarded only as being purely science fictional fantasies. Public expectations of what is feasible can be slow to catch up with the pace of technology.

Artificial companions

• Have you ever seen someone shut a door by grasping the edge of the door and pushing or pulling it shut?

• Would you recommend that strategy?

• What will happen to someone who tries?

• How do you know?

In a recent talk called What designers of artificial companions need to know about biological ones Aaron Sloman considers the requirements for AIs (presumably robotically embodied) capable of performing some domestic care worker type duties. There's a lot of stuff in this talk which I'd agree with, and which also relates to the kinds of issues mentioned in the AGI-08 discussion on future directions for AGI, such as whether it's useful to have a "roadmap" or benchmark tests to judge performance.

"It is specially important not to use fixed narrowly-focused benchmark tests to drive research: all that will produce is (at best) systems that are good on those benchmarks. Benchmarks can be very useful for engineering projects. Scientific AI research has been seriously harmed by benchmarks."

Benchmarks which are not narrowly focussed could be beneficial though. One such situation might measure how quickly the system can learn to play any arbitrary game whose rules can be described in an ordinary human language, where the range of permissible games is not defined in advance. So with appropriate instruction could the system learn to play chess or hopscotch or rubik's cube or hide and seek. The actual performance on the game would not be important, only the fact that it could play in accordance with the rules to a level similar to a naive human of average intelligence given identical instructions. This would really be a test of the system's ability to construct, refine and act in accordance with mental models.

Robosoccer: an AIBO's eye view

In this sort of game the visual processing is pretty trivial, since the bright colours of the pitch, ball and goals can all be easily segmented. Things probably get more interesting though when there are other robot players.

Sparkie's dream

Some background music whilst testing robots at the weekend. Mostly I'm still debugging the motion control and joysticking, and testing emergency stops.

First life meets second life

This is obviously moving closer to the "popstars in your living room" idea. Many forms of entertainment or communication could be delivered in this way, and the open source nature of the Second Life client software makes this more feasible.



The VR display used here is quite crude. A better solution would be something like an Eyetap display.