Singularity Indicators

At the risk of going out on a limb in the proper scientific circles to which I hope I belong(!), since 2006 I have begun to take very seriously the idea of the technological singularity: that exponentially increasing technology might lead to super-human AI and other developments that will change the world utterly in the surprisingly near future (i.e. perhaps the next 20--30 years). As well as from reading books like Kurzweil's 'The Singularity is Near' (which I find sensational but on the whole extremely compelling), this view comes from my own overview of incredible recent progress of science and technology in general and specificially in the fields of computer vision and robotics within which I am personally working. Modern inference, learning and estimation methods based on Bayesian probability theory (see Probability Theory: The Logic of Science or free online version, highly recommended), combined with the exponentially increasing capabilities of cheaply available computer processors, are becoming capable of amazing human-like and super-human feats, particularly in the computer vision domain.

It is hard to even start thinking about all of the implications of this, positive or negative, and here I will just try to state facts and not offer much in the way of opinions (though I should say that I am definitely not in the super-optimistic camp). I strongly think that this is something that scientists and the general public should all be talking about.

I'll make a list here of some 'singularity indicators' I come across and try to update it regularly. These are little bits of technology or news that I come across which generally serve to reinforce my view that technology is progressing in an extraordinary, faster and faster way that will have consequences few people are yet really thinking about.

January 2012 This `blog' (if that's what it was!) will probably stagnate because I am now posting the kind of stuff that was on this page on my Google+ page (which is publicly accessible even if you don't have a Google+ account).

October 20th 2011 Kurzweil responds to Allen and Greaves' article, linked below... rather convincingly I would say!

October 13th 2011 A friend (thanks Ian) pointed me to this article in Technology Review by Paul Allen and Mark Greaves about the Singularity, and arguing against predictions that it will happen within the time-frame Kurzweil predicts (mid 21st century). This is on the basis that the complexity of what would need to be achieved to create artificial human-level AI, whether via a brain scanning + simulation route of a pure AI route, is much greater than what is sometimes assumed and therefore that progress in mastering this complexity will not follow the same exponential curve as raw Moore's Law processing resources. They call this the "complexity brake".

Note that the article certainly does not argue that super-human AI and the Singularity are not on the way --- only that the timescale will be longer (they predict that in 100 years we will still be waiting).

October 11th 2011 I have just written a short article introducing the Singularity and describing the discussion we had about it at the Berlin Summit on Robotics. Any comments would be welcome.

October 10th 2011 An article from the Economist, about research from Stanford which shows that the energy efficiency of computing is doubling every 18 months. This is even more significant (though of course highly correlated to) the usual Moore's law type statement describing the amount of computation available per unit cost or similar. The article shows a great graph going back to ENIAC in 1945. I don't think that power problems are going to be the issue that stops AI research rolling forward!

September 21st 2011 Again not really about the Singularity, but an interested interview in Wired with James Dyson about Dyson, engineering and education (and the robot project that I am involved with).

August 20th 2011 A nice article in the Wall Street Journal by Marc Andreessen, co-founder of Netscape and tech investor, about how almost all industries are being taken over by software. This is definitely one of the pillars of the Singularity argument, since when technologies become information technologies they can experience exponential growth.

July 20th 2011 I was privileged to be invited to take part in the Berlin Summit on Robotics, event organised by Oliver Brock which brought together a collection of very interesting researchers working in robotics with the aim of discussing much more controversial or forward-looking topics than at a regular conference. There were great discussions on various topics, with a particular focus in the future of manipulation. While the past few years have seen great progress on perception problems like SLAM where the goal is to model aspects of the environment around a robot, there are still great challenges in turning those representations around into guides for action. Manipulating everyday objects in normal surrounding and with normal sensors is still very difficult. Clearly machine learning is an increasingly important part of the solutions now being proposed, but we discussed questions about how far learning could or should go. Are the human-designed models and representations we currently use in SLAM and other perception problems actually useful at all for manipulation, or should we let learning go the whole way and be free to design its own representations?

Rising computer power is clearly crucial here in changing the types of approaches which we can begin to consider in these problems. At the meeting, I made a short presentation about the Singularity because I was very keen to know what top roboticists thought about accelerating change and the long-term prospects of robotics. It was somewhat surprising and very interesting to me that the main view in the group was highly sceptical that we might be close to super-human AI, with the kind of issues I've described above about how difficult tasks like manipulation are not close to solution in most people's minds --- and that super-human computer power still does not take us much closer to super-human AI. Obviously there is the question of where the human-level software to run on these super-powerful computers will come from, which is still very much open to debate (and there are various possible routes, either from scanning or reverse-engineering biological brains or from a completely engineered AI approach); but my own feeling is that we can't say much about what will be possible with super-human computers until we actually have them. Perhaps the reason general AI doesn't work yet is just a question of scale and the software may drop into place relatively easily once the hardware is easily available.

July 19th 2011 An interesting counterpoint to Kurzweil's film is the film Plug and Pray, featuring AI pioneer Joseph Weizenbaum (who died during the making of the film) who had a much more pessimistic view of the prospects fo technological progress. It is not that he didn't think a Singularity type event would happen, I think; more that he was worried about what it would mean. The film features various current robotics projects and shows the researchers working on them apparently not thinking much about the long-term implications of their work; something I very much recognise that we usually do. I certainly share many of his fears about what will happen in near future of technology. However it is not clear to me that his suggestion to stop working on potentially dangerous projects is the way forward.

11th April 2011 Last week I went with Murray Shanahan to a showing of the film Transcendent Man at the Science Museum here in London, which is about the Singularity and the life of Ray Kurzweil. The showing included a Q&A session with Kurzweil himself. It's definitely worth seeing and I think backs up the ideas in The Singularity is Near strongly. I still find it hard to pick holes in the key arguments of accelerating change, although I don't completely agree with his very optimistic viewpoint.

A good way to judge a `futurist' such as Kurzweil is to look at how he's done at making predictions in the past. In this document you can see his assessment of how the predictions made in his early books written in the 80s and 90s are now faring. In particular his book `The Age of Spiritual Machines', written in the late 90s, made 147 predictions for 2009 and his evaluation is that 86% of those are mostly now true. I think this is a reasonable evaluation on the whole; and quite a lot of those predictions were certainly about non-obvious things that have only showed up in the past year or two with iPhones, eBooks, etc. which probably to me seemed unlikely only five years or so ago.

13th January 2011 I was lucky enough yesterday to be able to visit the central distribution centre of the online supermarket company Ocado in Hatfield, just north of London. This is claimed to be the most automated facility of its type in the world, and I was extremely impressed. (Look at a video of it here). Every online order they receive is assembled specifically in this centre before being sent out in boxes in vans to homes around the country (sometimes via regional distribution centres). The level of automation is very large: the boxes travel around the centre automatically on a network of conveyor belts, the locations of every one tracked by sensors. They arrive at picking stations where at least for now human workers place a few close at hand items in each before they move off to another station and eventually are taken down to the place where the vans are loaded. Of course there is huge IT complexity behind making this all run smoothly and many interesting optimisation problems to solve to maximise efficiency, and they have a large team developing the software required. It's hard to believe that this isn't the way forward for all types of shopping!

20th November 2010 There seem to be plenty of exciting things going on at the moment; one that we are all talking about is the low-cost availability of the Kinect device from Microsoft, released as a peripheral for XBox (which is a hugely ambitious step in putting real computer vision into the living room) but which will also revolutionise various areas such as robotics and augmented reality. The quality of the depth map data it gives easily surpasses that available from much more expensive time-of-flight depth cameras which have up until now only been a niche interest in research labs. Kinect's depth mapping works using the old idea of structured light --- an infra-red dot pattern is projected onto the scene, and observed by an offset infra-red camera which measures offsets to calculate a depth value for every pixel at 30Hz.

Another sign of the singularity idea being taken seriously by academics is this paper The Singularity: A Philosophical Analysis, by David Chalmers, which is well worth a look. We will be discussing it in our reading group in a couple of weeks.

19th May 2010 Excuse the indulgence of mentioning my own group's work on this page, but the research Richard Newcombe (a PhD student I co-supervise with Prof. Murray Shanahan) has been leading on Live Dense Reconstruction with a Single Moving Camera I think qualifies in terms of the fact that even though I see it running nearly every day it still barely seems possible to me. There is now vast computer power available for real-time computer vision from Graphics Processing Units (GPUs), now available for use as general purpose processors via languages such as CUDA and OpenCL, but it wasn't immediately obvious to me how that power could be harnessed to do something new in the real-time 3D vision area I am working in. This new work shows exactly what kind of possibilities are opened up, as we demonstrate that monocular SLAM can be augmented with rapid and accurate dense scene reconstruction. The applications of this to robotics or other areas such as augmented reality are obvious I think.

3rd November 2009 I finally gave in a couple of weeks back and got myself a smartphone. I went for the HTC Hero rather than the obvious choice of an iPhone in the end because I preferred the idea of the more open Android operating system and I have to say I like it a lot. I've never really been interested in mobile phones before but finally these ones are much closer to a PC in what they can do than they are to any phone I've had before --- really general purpose computing in your pocket. The only reason laptops will survive a bit longer is that the interface problems of keyboard and large screen still haven't been solved for pocket-sized devices --- I am still many times more efficient going through my email on a laptop than on the phone. But I'm sure someone will come up with good solutions for these problems soon.

This is old news for those who've had an iPhone for a year or two, but it is remarkable to think this this device makes nearly all of the gadgets I've bought over the last 20 years instantly obselete, while being smaller and lighter than nearly all of them. It has wi-fi, GPS, mobile network, Bluetooth, a camera, tilt sensor, etc., all integrated behind a very nice touchscreen interface. So the things it can work as include (just for starters): a watch, diary, still camera, camcorder (although the camera quality is not very good so I'll still carry a normal camera), satnav (with streaming Google maps!), radio (live streaming stations), portable TV (with BBC live channels, iPlayer, youtube), mp3 player (or with the Spotify app, access to most of the music in the world, anywhere), calculator, voice recorder, portable gaming console (though I haven't found anything brilliant yet), weather forecaster (not via local sensors, but it knows where you are and gets weather info. from the web), pretty usable web browser, email, etc... if computers have come from early 80s 8bit micros to this in only 25 years, I can't begin to imagine what we'll have in 2034...

29th July 2009 There was a major article in the New York Times this weekend about the Singularity and related matters, with various different opinions. Significantly (to me), those of us former and current members of Oxford's Robotics Research Group were alerted to this article by the head of the group Prof. Sir Mike Brady himself, showing that thinking about these ideas is really becoming mainstream.

23rd June 2009 It's been too long since I've posted to this page but I continue to observe technology racing into the future. Most of the following are very well known but still remarkable. Obviously Spotify, immediately spoiling all the fun of buying and owning music, finding something you've been looking for for ages in a record shop or recording things off the radio... but impossible to resist, and I still can't figure out how they can get away with playing so few adverts. Actually I have found myself going back to because I prefer the random element when I am working.

The new iPhone, devices which really now at last are delivering general purpose, always-on and hand-held computing (and see how Georg Klein in Oxford has just managed to make the PTAM camera tracker run on the old one); new game interface technologies from Sony and especially Microsoft (finally truly advanced computer vision for the living room). In robotics, Willow Garage seem to be right on the money in targeting home applications through advanced computer vision and manipulation. Actually I'm wondering if we are entering an era where the best technologies for computer vision and similar are going to be developed and sold by companies rather than academic projects, due the the large component of systems type work needed to develop such things.

3rd December 2008 I am currently really enjoying Douglas Hofstadter's recent book I am a Strange Loop which is about the mind and his idea that a person's sense of self has the properties of a strange recursive loop of signals and patterns in the brain. It is a very interesting read with lots of nice analogies and personal stories.

5th October 2008 Have a look at Dropbox. It is a service which offers automatic syncing of any type of files across multiple desktop machines (Windows, Linux or Mac) and a private space on a web-server in the cloud, all for free (at least for the moment). Incredible that there is now such a simple, cheap solution for what was a tricky problem not so long ago; this says something about the remarkable scaling of information and communication technologies.

29th September 2008 I was lucky enough to visit the team at Microsoft Live Labs this week which created Photosynth. This is a great service which features state of the art image matching and structure from motion to perform 3D reconstruction and registration from arbitrary sets of user-submitted photos. This really proves how mature 3D vision technology has become.

9th July 2008 I saw many impressive presentations at Robotics: Science and Systems 2008 in Zurich the week before last, but perhaps the most thought-provoking was the invited talk by Miguel Nicolelis from Duke University on his work on neural ensemble interfacing with the brains of monkeys. Essentially, by simultaneously monitoring the activity of around 500 neurons in different parts of the brain via implanted probes, it proves to be possible over a period of weeks to train the monkey to control a computer cursor or robot arm completely instinctively, as it it were another limb. On seeing these results it is hard not to believe that practical human brain-computer interfaces are coming in the near future.

11th June 2008 A great piece of work published at the recent conference ICRA 2008 by Mark Cummins and Paul Newman of the University of Oxford. They demonstrate in their FAB-MAP and ABFAB-MAP work that a principled probabilistic approach to appearance-based mapping permits reliable robot mapping and prior-free relocalisation on city scales without the need for geometric maps. See the videos and also now code which is available from Mark's homepage.

6th June 2008 There is a collection of very interesting articles about the technological singularity on the IEEE Spectrum website which show that the idea is getting taken more and more seriously. Comments from most of the usual characters, plus a few unusual ones. I was surprised to see Steven Pinker, someone whose views with which I'm generally completely in agreement, reject the singularity idea completely (here). He points to technologies that have been predicted which have never arrived (jet-pack commuting, underwater cities...) --- but I think there are other much more remarkable technologies which have suddenly popped up and become completely ubiquitous, mostly over the last 10--15 years (mobile phones, Google, GPS...). Rodney Brooks seems to want to appear wary but in the detail of his comments is very much in agreement with what Kurzweil says in his book about the ideas of accelerating change, humans merging with their technology and so on.

30th April 2008 Not really to do with the singularity (I don't think) but I just saw a talk by Prof. David Mackay (Department of Physics, University of Cambridge), whose work (and book) on inference and information theory I have long admired. His current project is a pragmatic, numerate investigation of sustainable energy and how we could live without fossil fuels and I think it is well worth looking at the current snapshot of the book he is working on called Sustainable Energy - Without the Hot Air. On the singularity theme, he does point out the exponential rise in the use of coal from the late 18th century to the present day.

27th March 2008 DARPA's Robot Pack Mule has incredible animal-like movement.

2nd March 2008 Below is an Extinction Timeline created jointly by What's Next and Future Exploration Network. Click on the image for the detailed timeline as a pdf (1.2MB). OK, this is a bit of a joke but it amazes me that things I took as essential tools as a child (maps? libraries? physical money?) are now become obsolete faster and faster as useful technologies...


Andrew Davison