A month ago, in The Swarm is Watching, I said that random, small groups would start building 3d models of the world, because they could. Building Rome in a Day is a project out of the University of Washington to do just that:

Dubrovnik

At the time of our experiments, there were only 58,000 images of Dubrovnik on Flickr. For this city we were able to experiment with the entire collection. Matching took only 5 hours on 352 compute cores. The largest and most interesting component corresonds to the old city. It is interesting that the reconstruction time for Dubrovnik is so much more than that for Rome. The reason lies in how the data sets are structured. The Rome data set is essentially a collection of landmarks which at large scale have a simple geometry and visibility structure. The largest connected component in Dubrovnik on the other hand captures the entire old city. With its narrow alley ways, complex visibility and widely varying view points, it is a much more complicated reconstruction problem, and this is reflected in the time it took to solve it.

Also worth noting is the fact that the reconstruction is not restricted to the city itself, as can be seen in the video below, it also contains the hills surrounding the city and part of Lokrum island which is south east of the city.

Labels: faster is different

Today I'd like to talk about two papers of moderate length on the subject of Engineered Intelligence. Reading both of them should take you an hour, maybe an hour and a half. Both are frequently cited, both discuss the relationship between intelligence and computation. And both are in the canon of the future.

If you are interested in Artificial Intelligence, or Augmented Intelligence, or Machine Learning, or even The Singularity; you need to read these two papers. They are short, well written, and accessible. Both papers are frequently cited, so many people feel that they are "familiar" with the papers. I'd like to convince you that these are both well worth your time, and I guarantee you'll learn a thing or two.

The first paper is "Computing machinery and intelligence", by Alan Turing. It was published in the journal Mind in 1950, but there are many copies of it bouncing around the net. Here are a few links:

• A PDF Scan of the Original Paper
• An HTML Translation on a tacky website with tables from the paper
• An HTML Translation with no tables
  

Alan Turing was a researcher in computation. He was involved in a great deal of the early work on what it meant to compute things, and what was and wasn't possible.The British government forced him to take hormones (in the 50's!) to make him more "manly", in the hopes they could turn him straight, and it seems likely that this was a major cause of his suicide. We'll never know how much more we missed because of this government sanctioned homophobia.

"Computing machinery and intelligence" is a wonderfully written and well articulated work on what we mean when we say "intelligence", and what it would mean for a computer to be "intelligent". It is also the source for the silly game called the "Turing test", which Turing himself only introduced to make a rather nice point about personal bias. Most of this paper is devoted to the arugments against engineered intelligence, and the various flaws in those arguments.

The second paper for today is "The Coming Technological Singularity: How to Survive in the Post-Human Era" by Vernor Vinge in 1993. This paper is much easier to track down, being a product of the web era:

• Vinge's official copy
• A copy with a nicer font
  

Vernor Vinge is a professor of computer science and mathematics who decided to pursue science fiction. He's got a good deal of work out now, I highly recommend his novel "Rainbow's End".

Vinge was one of the first author of the modern era to tie the old thoughts about machine intelligence with modern observations on acceleration. Unlike Ray Kurzweil, he isn't a shameless self promoter, so Ray gets more press. But I think Vinge is much more reasonable.

In "The Coming Technological Singularity", Vinge lays out the case for the inevitability of massively disruptive change brought on by engineered intelligence. Like Turing's paper, Vinge devotes most of this article to examining the arguments against engineered intelligence, and the problems with those articles.

Labels: faster is different, things worth reading

This is a post about how Faster is Different. This is a topic I'll return to frequently, but I want to make a point about it up front. Programmers think about the resources available to them in terms of space/time. This isn't the Physicist's space/time; this is about computers.

Computers process stuff (in time), and they store stuff (in space). So, roughly, processors = time, and memory/disk = space. Processors come at different speeds (operations per second), and with different numbers of cores. More of either gives you more time, so much that you can have more than 1second per second of time if you have more than one processor. More memory gives you more space.

This is interesting, because we (programmers) measure how expensive a given algorithm is in terms of space/time; and we measure how hard a given problem is in terms of space/time as well. It might seem strange to make the distinction between algorithms and problems, but sometimes you don't use the "best" algorithm to solve a problem. Generally, a problem is defined to be as hard as the absolute best algorithm that will solve it. If you don't have enough space/time to solve a problem, then you can't. This is the basis for cryptography, very hard problems.

Anyway, back to the main point: Faster is Different.

More space/time doesn't just make your existing stuff faster, it sometimes changes a problem with no solution (because it was too expensive) into a problem with a wildly expensive solution. And problems which are wildly expensive today will be very expensive a year from now, quite expensive 3 years from now, and sorta cheap in 5 years. 10 years from now, they'll be free. All thanks to Moore's Law, and its impact on our future supply on space/time, assuming it doesn't stop. Now, there's reasons to question Moore's Law, but that's for another post.

In this installment of Faster is Different, I want to talk to you about The Swarm. The Swarm is, roughly, everybody and their computers. The Swarm can do a lot of interesting things, and the Swarm is always watching you. Right now, that doesn't really matter, because it doesn't have a lot of space/time available to it. But that's gonna change.

My employer, Google, just posted some interesting research results in computer vision. I wasn't involved in any way, but its germane to the topic of this post:

Official Google Blog: A new landmark in computer vision
Our research builds on the vast number of images on the web, the ability to search those images, and advances in object recognition and clustering techniques. First, we generated a list of landmarks relying on two sources: 40 million GPS-tagged photos (from Picasa and Panoramio) and online tour guide webpages. Next, we found candidate images for each landmark using these sources and Google Image Search, which we then "pruned" using efficient image matching and unsupervised clustering techniques. Finally, we developed a highly efficient indexing system for fast image recognition.

While we've gone a long way towards unlocking the information stored in text on the web, there's still much work to be done unlocking the information stored in pixels. This research demonstrates the feasibility of efficient computer vision techniques based on large, noisy datasets. We expect the insights we've gained will lay a useful foundation for future research in computer vision.

If you're interested to learn more about this research, check out the paper.

Now first, I want to say that this is really neat research. And the math is cool. And the people involved have every reason to be proud of themselves.

Microsoft has an image group which regularly produces equally cool results. A while back, they came up with an application that was so popular, they made it a publicly accessible toy:

Microsoft Live Labs: Photosynth
You can share or relive a vacation destination or explore a distant museum or landmark. With nothing more than a digital camera and some inspiration, you can use Photosynth to transform regular digital photos into a three-dimensional, 360-degree experience. Anybody who sees your synth is put right in your shoes, sharing in your experience, with detail, clarity and scope impossible to achieve in conventional photos or videos.

So, what properties would we expect from the composition of these services? It would find images of commonly photographed sites, and would then abstract them into 3d models of the space, with textures. All of this without sending anyone out to scan or photograph the site. Now, add a few computational generations to this, say 5 years; and people will be doing this for video.

And if you'll recall, pretty much everyone's launched facial recognition software in the past year. Apple did it, Google did it, I lose track. It still isn't great, but its pretty good, and more space/time and samples can only make it better.

But because Faster is Different, what Google and Microsoft can do today, pretty much any company can do in 3-5 years, and small organizations will be able to do in 7-8 years.

So I want you to forget about Big Brother, or Evil Corporations, or The Bad Guys. Over the next 10 years, The Swarm is gonna map the entire world, and it is going to be able to tell, from the pixels alone, who's in any photo on the web, and where it was taken, and probably to some extent, when it was taken. You cannot stop this from happening, because as it gets cheaper, more and more organizations will independently do this themselves. This is inevitable and implied, because:

• People want to know,
• People profit from knowing, and
  
• Faster is Different.

Labels: faster is different, surveillance

Charles Stross, one of my favorite authors, made an excellent keynote about the future of technology at the LOGIN 2009 conference. Keep in mind, Stross is openly taking a deliberately conservative view, assuming essentially no disruptive technologies.

Key points:

• Expect only one to two orders of magnitude (10x to 100x) improvement in processing power and power consumption.
• Expect ~ 1tb/s wireless data, but no more.
• Expect your mobile phone/computer/thingy to be as powerful as your current desktop.
  

I strongly recommend you read the whole thing:

LOGIN 2009 keynote: gaming in the world of 2030

"Let me give you a handle on this device, the gadget, circa 2020, which has replaced our mobile phones. It's handheld, but about as powerful as a fully loaded workstation today. At it's heart is a multicore CPU delivering probably about the same performance as a quad-core Nehalem, but on under one percent of the power. It'll have several gigabytes of RAM and somewhere between 256Gb and 2Tb of Flash SSD storage. It'll be coupled to a very smart radio chipset: probably a true software-directed radio stack, where encoding and decoding is basically done in real time by a very fast digital signal processor, and it can switch radio protocols entirely in software. It'll be a GPS and digital terrestrial radio receiver and digital TV receiver as well as doing 802.whatever and whatever 4G standard emerges as victor in the upcoming war for WWAN preeminance."

Labels: faster is different