Considerably less than 2x the bandwidth, if you have an intelligent encoding/compression scheme that exploits the similarity between right and left visual channels. 2x would correspond to completely unrelated video streams to each eye.
Something to keep in mind is that Sony is vertically integrated here, and can unilaterally bring 3D to market. They make TVs, own the Blu-Ray standard, own the PS3, and even have their own movie production division. So they can bring at least the movie and gaming part of this to market without anybody's cooperation. Broadcast TV in 3D is IMHO much farther away; it took digital TV so long to become a standard that it was technically obsolete on launch day.
So Sony can do it. It's a separate question whether this kind of non-industry standard solution has a long term chance of success. Sony has a history of funky homegrown technologies that failed to become standards (e.g., Memory Sticks, Betamax, ATRAC). On the other hand, Sony is uniquely positioned among for example the console makers to be able to do this, so maybe they see it as a way to set their PS3 and TVs apart from the competition. Neither Nintendo nor Microsoft can bring a full solution to market, so if customers take to the idea of 3D for gaming and/or movies, it gives Sony a good competitive advantage.
Since you are a software engineer, I'm guessing you don't have deep contacts into the movie or games industries, the biggest potential users of this technology. The reason I bring it up is that unless you understand your customer really well, even if your technology is pretty good you will have slim chances of giving your customers a full solution they can use. If you know someone who knows mocap inside and out, and can interest them, then perhaps you can persuade them to be a partner.
If you're a relative outsider, I think your goal should be to sell your company (and technology/patents) to an established company in the industry. If the technology is good and you have a well-written patent, this could be fairly lucrative. I'm guessing that's what a company like Sixense is trying to do.
Now, if your goal is to sell your technology so that someone else can bring it to market, the key is to have a very compelling technology demonstration of what it can do. Millimeter precision in motion capture is something that magnetic sensors can achieve today, for example the systems from Polhemus. How are you better? Do you track over a larger volume, or with smaller markers? Or do you feel you can achieve a cost that's lower than anybody else, thereby opening up new markets? The technical characteristics will determine: (a) how distinctive your technology is, relative to others that exist, and (b) which companies could make the best use of your technology. Others have pointed it out, but I'll reiterate: Nothing sells technology like a good working prototype. If you don't have this and have already filed a patent, you are taking a big chance that your claims are written correctly.
So you can get to your data from anywhere. Data no longer lives in some particular spot, so you can get your contacts, notes, emails, etc. from any device, anywhere.
So you can easily collaborate with people. Anyone who's ever tried to collaborate on a complicated presentation or spreadsheet has lived the nightmare of version numbering, merging changes, etc. Problem completely solved.
So your data will stay safe. What if your laptop gets stolen, or worse yet, your laptop plus your backup server get destroyed because your house burns down. Cloud providers can spread your data across many facilities. Yes you could back up all your files regularly on an external hard disk that you leave in a different location (work, the trunk of your car, etc.), but how many of us really do this?
Lower cost through aggregation. This applies more to CPU cycles than storage. Your typical computer CPU is very under-utilized, even when you're actually using it. The time-averaged CPU utilization is probably less than 5% for most personal computers -- that other 95% is waste. A cloud provider can aggregate users, and to a large extent put that wasted resource to good use by serving other users. Example: Say an ultra-high-end gaming PC costs $0.20 per hour to own and operate over its lifetime. If you're a hardcore gamer you might use it, say, four hours a day on average -- in which case the true cost to you is $1.20 per hour of use. A cloud provider might be willing to rent that PC to you for $0.30 per hour of use, because they can sell it to other users during the 20 hours you're doing other things. Now your cost of doing high-end gaming has gone down by a factor of 4. (The big loser here is Intel, who has now sold far fewer CPUs than they otherwise might have.)
Obviously there are downsides to "cloud computing", so it's not a solution to all problems. But for some uses the above benefits are very real.
No. Patents by themselves are not evil. They serve an important purpose, which is why the Constitution of the United States explicitly provides for them. Unfortunately the system today is overwhelmed, trying to deal with very technical concepts that the typical examiner may not completely understand, and is therefore subject to abuse. It is this abusive behavior that is evil, not the idea of patents or the patent system.
Actually much of the abusive behavior is completely rational. Honestly I can't fault Amazon for getting the 1-click patent, if for no other reason than to keep someone else from getting it and suing them.
but I still think the best spot for observational astronomy has to be the far side of the Moon. You've got several thousand miles of light and EM shielding, and a good couple weeks' seeing a month when the Sun goes down. Once the 'scopes cool off, there's no warping. What's not to love?
For radio astronomy you may have a point. For optical astronomy, (a) EM shielding is irrelevant, and (b) you have the dust problem. The lunar dust is a big challenge to any precision work on the surface. Through mechanisms that are not well-understood, the dust appears to develop and retain a static electrical charge in sunlight, at which point it sticks to anything and everything. Think of those foam packing peanuts and how hard they can be to manage, then apply it to micron-sized particles. The Apollo astronauts had a hard time dealing with the dust during their excursions on the surface. Also the electrostatic repulsion is so strong that it is observed to loft dust as much as several kilometers above the surface, as "dust fountains". In general the surface of the moon is a pretty dirty place for doing precision optics, so you need a good reason to be there.
Potentially one such good reason would be to build a very large optical interferometer. The moon is geologically very stable, so conceivably one could build multi-kilometer synthetic apertures at optical wavelengths. The question is whether one could achieve something similar with a spacebourne constellation of telescopes in precise formation (e.g., Darwin, LISA), which would be much cheaper.
Since it's so calm, it'll just accumulate, then condense on the cold optics. Have fun seeing when your mirror's frosted over with an inch of rime.
Presumably you put the generator a sufficient distance away to minimize any disturbance to the optics, or to seeing quality. The area gets almost no precipitation and probably no animal life, so anything you lay on the ground will remain undisturbed. In this sense it seems like an ideal place to run an automated telescope, if you can get past the somewhat difficult access issue.
Regarding the "condensing on the optics" problem, astronomers have hundreds of years of experience dealing with this issue. The simplest approach is to slightly warm the optics using resistive heaters. As long as the optics are slightly warmer than the surroundings, any water in the air will condense somewhere else. You don't want too much heating, since then you form convective air currents above the mirror that harm the seeing conditions. However with some reasonably accurate temperature sensors and a feedback controller, the condensation problem is straightforward to solve.
The transistor is probably the best example, but there's plenty of others. It didn't make AT&T the richest company or most powerful company on the planet, but it changed our world.
This to me highlights the real questions: Did AT&T actually make a positive return on their investment in Bell Labs? If not, then why did they think it was in their best interests to fund Bell Labs in the first place?
The funny thing about technology is that the big payoff doesn't usually go to the inventor, or the scientist who did the basic research that led to the invention. People assume this, but it really isn't the case. The original inventor may receive patent royalties (if they're lucky), but the real payoff is in commercialization. Bell Labs invented transistors, just as Xerox invented ethernet and the GUI, and Sun invented Java. However none of these companies especially benefited from their innovations. We should expect the norm to be no large-scale corporate funding of basic research, not the reverse.
I suspect AT&T and RCA did what they did in order to mollify people who felt they were monopolies in their respective areas. I.e., they did this as a public good, in an effort to improve their public image and discourage regulators from brining them down. In this sense perhaps Bell Labs was a good investment for AT&T, if it delayed their breakup.
Seriously, I'd much prefer an open database of scanned works rather than letting one company negotiate a deal.
It is a nontrivial exercise to obtain high-quality scans of 20+ million books. The scanning must be done non-destructively, since nearly all of these books are out of print. This means someone/something turning pages and taking pictures. It costs most archivists hundreds of dollars to scan each book this way. Which is fine if you're the Brewster Kahle trying to compile a very small collection. If you want to do a complete job of it, it costs hundreds of millions of dollars, if not billions -- and that's if you get the scanning technology and QC pipeline right.
The question is: Who pays all that money to do the scanning?
I'm guessing Brewster Kahle would prefer that the US Government fund it. Maybe that would be nice, but I don't think it's particularly realistic. Other than that, only Google has stepped up to this effort. Microsoft quit theirs last year. If Google thought they had no legal basis to use this material, or make any money from it, I guarantee they would stop the scanning in an instant. They aren't stupid after all.
I'm guessing the "Open Book Alliance" has no intent to invest the scale of effort needed to pull this off. They're just trying to shoot torpedoes at Google.
That kind of clears up what orphan books are, but still, if no one knows who owns the rights, who are the people complaining?
I gather much of the issue some authors have with this agreement stems from a couple of concerns:
1. They feel they're being steamrollered into giving up some rights. For example, the agreement has provisions that allow Google to sell electronic access to works, and share the revenue with the authors. By signing on, they lose the right to negotiate individually with Google. Also, each public library will have one terminal where the full text of all works will be visible; some authors may perceive this as "giving away" their work. The interesting thing is that it's very likely in every author's best interest financially to participate. Nevertheless I imagine some fraction will (irrationally) opt out, just to make a statement.
2. They are concerned that Google will gain too much power within the publishing industry. The settlement appears to give Google a preferred position with regard to distributing these works. Maybe somebody else will come along and try scanning all of the same books, and negotiating a similar settlement, but then again they might not. Perhaps a lot of authors are wary of being trampled by yet another Big Corporation.
Also, to clarify with regard to orphaned works: The ownership of these xxM works is not necessarily unknowable, it just isn't known at present. You could send investigators after each work in turn, but this would be impractically expensive given the very obscure nature of a lot of these books. So therein lies the present dilemma: A lot of good content without clear ownership, and a very strict copyright regime preventing anyone from using it. View this as collateral damage from our particular copyright regime. I think the courts generally acknowledge this is a problem, and are very interested in solutions.
"Your honor, I had no idea the contents of my iTunes folder were being served to all takers through the internet. It turns out the browser I use (IE6) allowed some bad site to install a virus on my computer. I didn't have an antivirus program installed, that doesn't come with Windows. I honestly had no idea this was happening..."
Mmmm...the savory smell of plausible deniability....
My prediction is that there will not be a human outside of low Earth orbit for at least the next 50 years, with the possible (unlikely) exception of the Chinese attempting a lunar orbit or landing.
The cost of human spaceflight is going up. In part, this is because we haven't made significant investments to improve propulsion technology since the 1960's, so efficiency hasn't been improving. At the same time, we as a nation seem less willing to accept risks with spaceflight than we used to. At one time, astronauts were mostly former military or test pilots, people used to the idea of real risk. Now we treat the Space Shuttle like some kind of bus into space, and we expect the bus to be safe and comfortable. We send up teachers, congressmen, scientists, tourists -- pretty much anybody who wants to go. The expectation of safety means more engineering margin and backup systems, driving up cost.
The capabilities of robotic craft are steadily improving. Moore's Law and all that. What will an autonomous rover on Mars be able to do, with another 30 years' development? It's hard for me to imagine sending a person on a 15 year voyage to Europa to dig through the ice, or to Titan to explore the hydrocarbon oceans.
Nobody has identified a compelling economic, scientific, political, or military rationale for sending people into space. Arguments based on national pride, or fear of being surpassed technologically, have for now evaporated.
The ISS. This $xxB useless boondoggle must be playing some role in tempering Congress's enthusiasm for Big Projects.
Perhaps in the 50-100 year timescale, we'll have figured out radically different approaches: Nuclear propulsion, a space elevator, a launch loop. Or we'll be able to upload our minds into hardware, and send people into space without sending bodies. E.g., your consciousness gets radioed into the probe once it's tunneled through the Europan ice.
First point, the human eyes and ears are not better than man-made equivalents. They are extremely good given the limitations of biology, but not nearly as good as man-made sensors like CCDs. But this is a side point. Obviously brains can learn and be effective even if their sensory inputs are degraded. The "magic" of intelligence does not lie with sensory fidelity.
Your point about gesture recognition underscores what I'm saying. You shouldn't need to have any training period at all in a gesture recognition system: It should work the first time. Our systems today are too brittle, and need to be fine-tuned. Even when they're tuned, they still make errors a 3 year old would never make. The approach I'm advocating, growing from a seed in connection with a rich environment, is something that happens before deployment time so to speak. What ships to the customer is a fully trained AI. The "training time" with an individual should be zero. How much training time do you need to understand the speech of someone you just met?
The issue regarding "what fits inside a small drone" is a reasonable one, and clearly depends on implementation details. The human brain is again maybe too lofty a goal. For image recognition and navigation tasks, a mouse is capable enough to be very interesting. What resources are required to "simulate" a mouse brain? I don't know. My feeling though is that Moore's Law will eventually solve this problem. What Moore's Law will not solve is the question of how do we properly build these systems. My point is that real intelligence needs to be grown, not designed.
Can't we just skip "concepts", and move straightaway to "meta-concepts"? After all, a "concept" is just a concept itself, so with meta-concepts we'll be able to define "concepts" recursively. Which doesn't sound like a win, until you realize then you can redefine concepts to fit your own idiosyncratic needs. Like how in my code, the first thing I always do is overload "+" to mean "*", and vice versa. I've always liked them the other way around, not sure why. Anyway, back on point: Concepts by itself is like the 4-blade razor, a lame, stupid half-measure. The real prize, the 360-dunk-off-the-free-throw-line, is 5 blades on a razor. I move we skip "concepts" and go for the big win. Those effeminate Python dorks will have no answer to this, they'll be stunned to see just how inferior they really are. Who's with me on this?
Broader point: I'm sick and tired of these language designers not giving us enough features. For the last 20 years I've been waiting for a language that will allow me to redefine keywords. If that too much to ask? What if I don't happen to like "for", or "while", or "return"? Do you people lack vision, or competence, or both? Second thing on my must-have list is a pre-pre-processor. I'm tired of writing all these header files all the time. I want a way to generate them programmatically, at compile time. A small embedded scripting language would do fine, just make sure it has templates and operator overloading and multiple inheritance, so I can stretch my legs and get comfy with it. Come on people, start earning your paychecks and get some of this stuff done!
It's the reconfigurable nature of the human brain that's unique and powerful.
Strike "humans", and I think you're on to something. We have been taking the wrong approach to AI, if the goal is to build something as adaptable and resilient as a brain. Biological brains get all their information from the environment, mostly by interacting with it. The amount of initial, hard-wired information must be very small: The entire genome is only 4 billion base pairs, and likely only a small fraction codes for the functioning of the brain. It's probably only a few hundred megabytes, at most. The secret of the brain is that it's a self-organizing system, incorporating and organizing information from the environment to build a very resilient end product. A consequence of this learning (or bootstrapping) algorithm is the adaptability and "reconfigurability" you noted.
Most traditional AI research has a lot of hard-coded information, it's very different. So the systems are brittle and not adaptable. Take an infant at birth and deprive it of all sensory input, forever. This is what we do to our computers. And we expect them to be smart?
My hope for AI is that people will figure out the learning algorithms that brains employ, and then we figure out how to start from a very small seed, couple it together with a rich and interactive environment, and end up with something interesting. Human brains are too lofty an initial goal -- I'd be happy with something as resilient as a cockroach. I can never kill those damned things.
Maybe CS departments of the future will have courses on parenting skills?
There are a couple of things being conflated in this type of research, which to me muddies the water. One question has to do with performance of people not too far from the median. For this question, I believe it's reasonable to look at how achievement test scores vary with factors like gender, race, culture, nationality, socioeconomics, and so forth. The original research cited here involves data of this type. And the conclusion isn't so surprising: Female performance relative to males is very situationally-dependent. Anecdotally one only needs to look at the gender gap (if one exists) in east Asian students vs. the gender gap in white students. *Maybe* white women are at some genetic disadvantage relative to asian women -- again relative to their respective male counterparts -- but it seems unlikely relative to a cultural factor.
What these lines of research don't really show -- because there isn't enough comparative data available -- is what are the external factors that most correlate with the gender gap within different groups. Is it culture that drives the variation? (Asians have higher expectations on daughters? Asians don't propagate the "geek stigma" as much for girls?) Is it economics? (Poorer people cannot educate all their kids, so preferentially educate the boys?) Or something else? Who knows.
The second question being conflated is performance at the far, far, end of the performance spectrum. Fields medal winners represent the 99.999+ percentile. Who knows what defines people out there? There aren't enough of them to really study as a statistical emsemble. It's fair to say that at the high end of any performance curve, a lot of things have to come together simultaneously: Raw talent, motivation, opportunity, persistence, environment, dedication. It could be for example that men have no more innate ability than women, but are just more single-minded in their approach to life. I.e., more men than women are willing to do what Andrew Wiles did, namely hole up in an attic for 10 years to prove Fermat's Last Theorem (with a low probability of success).
Finally, I think with regard to this sort of research it's important to maintain a dispassionate attitude. When I get the feeling the authors are trying to *advocate* for a particular conclusion, that makes me a bit queasy. There seems to be this unstated assumption that an unequal outcome is indicative of unequal opportunity. Would anyone argue that the relative lack of white men in the NBA is indicative of low opportunity or discrimination? Probably not. Perhaps white women don't pursue math at the highest levels because they simply don't want to, compared with other uses for their time. Is this a bad outcome? Within the scientific enterprise it's a very slippery slope to start asserting value judgments about these things.
The 3DV site has a number of papers showing how the underlying technology works. They've spent the last several years refining it to make it smaller/cheaper, but the physical principles are the same: In addition to a regular video camera to capture RGB, a pulsed infrared source illuminates the entire scene. A fast shutter on the camera then allows it to infer distance to each pixel, based on the time-of-flight of the reflected infrared pulse. The literature for an earlier version claims 1-2 cm depth resolution at full video rates, but I haven't seen specs for the latest small/compact/cheap version.
Using this as a game controller strikes me as potentially very interesting, but with caveats. First I think this would be hard to program: You get all of this per-pixel RGB/depth information out of the camera, from which you need to extract limb orientations, etc. Seems pretty hard to do well, although I suppose a library would abstract that away from the game developer. A more fundamental limitation is the low spatial resolution: Basically it just detects gross movements of the limbs, and by itself could only support a few types of games. I think this would be most useful in combination with a controller for faster/finer motions (e.g., a button-based interface). Maybe that's where the future is headed: Overall body tracking at low resolution, selected "high-sensitivity" parts (hands, eyes) tracked at high resolution.
Another potentially interesting next-generation game controller is the Sixense 6dof controller. It is basically a magnetic tracker (full absolute position/orientation sensing) combined with a button-based interface. I could see a lot of interesting games working well on that. At any rate, what's cool about the success of the Wii is that people are innovating here, and the console makers now seem willing to spend a nontrivial part of their overall COGS on controllers.
I agree, this article as written makes no sense. To "open source an API" is muddy thinking, a non-concept. At least in the US, there is well-established legal precedent allowing companies to duplicate APIs at a functional level. E.g., the function of APIs is not copyrightable in the way that source code is. So anybody that wants to can come along and implement their own versions of the EC2, S3, etc. API. They don't need any approval from Amazon.
Now Amazon may decide it's in their best interest if other cloud providers adopted their APIs. Presumably they would do this to encourage more companies to adopt cloud computing (i.e., eliminate lock-in as a risk). They could advance their APIs as standards in a number of ways, including making some of their own specific implementations open-source. Is this what the article is trying to say?
IMHO the main thing is to be as diverse as possible in where the data is located and how it's managed: Geographies, access controls, software, etc. Eliminate as many common failure modes as possible.
For the servers I run, here's what I do. I don't see any real flaws with this procedure, and it's simple to set up:
A cron job on the main server periodically (in my case every hour) creates an archive of all changes, saved into an archive directory as a.tar.gz with a timestamp in the filename. Every so often the archive file is instead a full backup of all data, i.e., not incremental.
On a different machine at my house, a script periodically (in my case every day) does an sftp to the main site and downloads all of the new archive files. These are processed and applied to a local version of the site; not to act as a live backup, but to confirm there is no data corruption and that restoration is working properly.
Each day, the second machine is backed up into Amazon S3 using JungleDisk (the.tar.gz incremental backup files, not the local site copy). JungleDisk can by the way locally encrypt your data before uploading to S3.
This gives three independent copies of the data in very different locations. If the live data is corrupted, I can restore from the backups on the main server and lose at most an hour of data. (More accurately, an hour plus however long it takes me to notice the corruption, and then do a recovery. I haven't gone to great lengths to streamline these operations, although obviously one could.) If the entire main server is wiped, for example by a hacker, I lose at most a day of data. It's hard for me to envision a scenario where recovery would be impossible.
The why of this is fairly straightforward from a financial standpoint. Companies can raise money from two sources: Equity and debt. The cost of debt is obvious (the interest rate). The cost of equity is less obvious but very real: Investors demand a particular total rate of return on the money they invest in a stock, either in the form of dividend payments or retained earnings (appreciation in the value of the stock). If the total rate of return from your stock is less than what the market demands (based on its perception of how risky you are), then your stock price will fall until the desired rate of return is met. Typical long-term total return from the stock market is 9-10%, and for a tech company most investors will want more because of the perceived risk.
Anyway, the point is that when interest rates are low, it's a lot cheaper to get money from debt markets than from equity markets. So the smart CFO will borrow money and use it to buy back (and retire) stock. If you're a shareholder you like this in net, because although the company now has debt to pay back (a liability which decreases the value of your shares), the positive impact on value from having fewer shares outstanding outweighs it. The only downside to this strategy is that interest on debt must be paid back on a defined schedule -- bond holders aren't willing to defer their payoff like equity investors are (and consequently bond investors make lower returns on average). GM is an object lesson in getting squeezed this way. Many tech companies avoid long-term debt as a result; they don't like the ongoing obligation. If anything this move by Microsoft signals to the market that they've become a stable business that is confident in its long-term ability to generate cash.
Consider the various moral choices in Fallout 3. Functionally, the game allows you to decide what you want to be. If you want to be a slaver? It is possible. If you want to, instead, rescue slaves? Very much also possible.
These "moral choices" are very superficial in today's games. There are no real, long-term consequences for behavior, as there is in the real world. In truth, violent people don't get away with it very long. If you do something bad to someone, they remember and tell other people, and your bad reputation spreads. If you drive GTA-style in a typical city you'll be shut down very quickly. Even in wartime there are codes of conduct; if you shoot your buddy, or an innocent bystander, there are big consequences. Games today don't represent any of these nuances. You shoot a bad guy, and all the rest just keep doing what they were doing. You shoot a good guy, and nothing much happens. I would never advocate for outlawing or banning anything, and I don't particularly fault game developers for giving people what they want. That said, I don't think the end product is very instructive for kids. Adults can see these games for what they are -- escapist action with no connection to reality -- but a very young kid doesn't yet have the context to see that.
With regard to "protecting the children", there is an interesting (to me) difference between European and American perceptions. European children's literature is much more willing to touch on darker themes than its American counterpart; authors like Roald Dahl, J.K. Rowling, Philip Pullman, and the brothers Grimm address things you'd almost never see in any American kids books. As a parent I feel this is fine, however, since these works usually illustrate something true about the consequences of behavior, both good and bad. Bad things happen in Willy Wonka's chocolate factory, when the kids are greedy or mean.
Back on games, I think there is a coming backlash from consumers who are getting bored with the typical shooters. I like shooters as much as the next guy, but why do 95% of the top titles have to follow this formula? I used to buy a lot of X-Box games, until I realized they're all basically the same game (Portal excluded). Even things as basic as the color palette; why are all games so dark (literally)?
"there are many equivalent and roughly as-challenging problem areas out there, and there seems to be a great deal of work being done on neuroscience and related fields, with breakthroughs announced on a daily basis. What's driving the interest, time and money being devoted into neuroscience, as opposed to, say, cancer immunology.... ?
A key fact about science is that the ultimate utility of any line of work is impossible to predict. Sometimes the fields that people expect to yield big dividends have modest results (classical AI, superstring theory, fusion power generation). In other cases the importance of a piece of work ends up being much greater than anticipated (the maser, the transistor, cosmic microwave background radiation).
In the face of this uncertainty, you want to do science like you invest your 401(k) -- spread your money around on a lot of different bets.
Quite a few people in the industry now are starting to care about power efficiency at the other end of the performance spectrum too. The Green 500 list for example tracks Megaflops per Watt data for the top 500 supercomputers. Judging from this data the Cell processor looks very good.
The reasons for caring about energy efficiency at the high end are of course very different from what ARM is trying to do, which is to maximize performance within a given battery life envelope. For large installations it has more to do with operating cost and environmental concerns.
I really want to remain optimistic about it but for me this whole orion project is like a reminder of where we *could* have been at the completion of the Apollo launchers.
What's lacking is a clear impetus for the manned program. Since the early 1970's it's been viewed by Congress as, at best, a jobs program to stimulate their local districts. (Note how NASA facilities are cleverly distributed across the country, to maximize the base of support. And note the gaping lack of any long-term objective behind initiatives like the ISS, or Orion for that matter.)
Personally I don't believe overcrowding, asteroid mining, or any other economic incentives will ever be powerful enough to drive manned spaceflight beyond low earth orbit. It's just too expensive, plain and simple. (Something like the space elevator could maybe change that, but that's a separate topic.) Scientific knowledge is a valid objective, but it tends to argue in favor of much cheaper unmanned missions. There are many things a human geologist could do on Mars that current rovers cannot, but is it worth a 100x cost multiplier? Most scientists would say no.
The only thing that can provably drive a major push into manned spaceflight is what did it the first time, namely competition from another country. I predict we won't advance beyond the Apollo level of capability until another country is knocking on that same door, and there is a perceived land grab at stake. At present I think China is the likeliest candidate.
Slashdot Mirror
Considerably less than 2x the bandwidth, if you have an intelligent encoding/compression scheme that exploits the similarity between right and left visual channels. 2x would correspond to completely unrelated video streams to each eye.
Something to keep in mind is that Sony is vertically integrated here, and can unilaterally bring 3D to market. They make TVs, own the Blu-Ray standard, own the PS3, and even have their own movie production division. So they can bring at least the movie and gaming part of this to market without anybody's cooperation. Broadcast TV in 3D is IMHO much farther away; it took digital TV so long to become a standard that it was technically obsolete on launch day.
So Sony can do it. It's a separate question whether this kind of non-industry standard solution has a long term chance of success. Sony has a history of funky homegrown technologies that failed to become standards (e.g., Memory Sticks, Betamax, ATRAC). On the other hand, Sony is uniquely positioned among for example the console makers to be able to do this, so maybe they see it as a way to set their PS3 and TVs apart from the competition. Neither Nintendo nor Microsoft can bring a full solution to market, so if customers take to the idea of 3D for gaming and/or movies, it gives Sony a good competitive advantage.
Since you are a software engineer, I'm guessing you don't have deep contacts into the movie or games industries, the biggest potential users of this technology. The reason I bring it up is that unless you understand your customer really well, even if your technology is pretty good you will have slim chances of giving your customers a full solution they can use. If you know someone who knows mocap inside and out, and can interest them, then perhaps you can persuade them to be a partner.
If you're a relative outsider, I think your goal should be to sell your company (and technology/patents) to an established company in the industry. If the technology is good and you have a well-written patent, this could be fairly lucrative. I'm guessing that's what a company like Sixense is trying to do.
Now, if your goal is to sell your technology so that someone else can bring it to market, the key is to have a very compelling technology demonstration of what it can do. Millimeter precision in motion capture is something that magnetic sensors can achieve today, for example the systems from Polhemus. How are you better? Do you track over a larger volume, or with smaller markers? Or do you feel you can achieve a cost that's lower than anybody else, thereby opening up new markets? The technical characteristics will determine: (a) how distinctive your technology is, relative to others that exist, and (b) which companies could make the best use of your technology. Others have pointed it out, but I'll reiterate: Nothing sells technology like a good working prototype. If you don't have this and have already filed a patent, you are taking a big chance that your claims are written correctly.
Why in the hell would I even WANT to do that?
Here are some I can think of:
Obviously there are downsides to "cloud computing", so it's not a solution to all problems. But for some uses the above benefits are very real.
No. Patents by themselves are not evil. They serve an important purpose, which is why the Constitution of the United States explicitly provides for them. Unfortunately the system today is overwhelmed, trying to deal with very technical concepts that the typical examiner may not completely understand, and is therefore subject to abuse. It is this abusive behavior that is evil, not the idea of patents or the patent system. Actually much of the abusive behavior is completely rational. Honestly I can't fault Amazon for getting the 1-click patent, if for no other reason than to keep someone else from getting it and suing them.
but I still think the best spot for observational astronomy has to be the far side of the Moon. You've got several thousand miles of light and EM shielding, and a good couple weeks' seeing a month when the Sun goes down. Once the 'scopes cool off, there's no warping. What's not to love?
For radio astronomy you may have a point. For optical astronomy, (a) EM shielding is irrelevant, and (b) you have the dust problem. The lunar dust is a big challenge to any precision work on the surface. Through mechanisms that are not well-understood, the dust appears to develop and retain a static electrical charge in sunlight, at which point it sticks to anything and everything. Think of those foam packing peanuts and how hard they can be to manage, then apply it to micron-sized particles. The Apollo astronauts had a hard time dealing with the dust during their excursions on the surface. Also the electrostatic repulsion is so strong that it is observed to loft dust as much as several kilometers above the surface, as "dust fountains". In general the surface of the moon is a pretty dirty place for doing precision optics, so you need a good reason to be there.
Potentially one such good reason would be to build a very large optical interferometer. The moon is geologically very stable, so conceivably one could build multi-kilometer synthetic apertures at optical wavelengths. The question is whether one could achieve something similar with a spacebourne constellation of telescopes in precise formation (e.g., Darwin, LISA), which would be much cheaper.
Since it's so calm, it'll just accumulate, then condense on the cold optics. Have fun seeing when your mirror's frosted over with an inch of rime.
Presumably you put the generator a sufficient distance away to minimize any disturbance to the optics, or to seeing quality. The area gets almost no precipitation and probably no animal life, so anything you lay on the ground will remain undisturbed. In this sense it seems like an ideal place to run an automated telescope, if you can get past the somewhat difficult access issue.
Regarding the "condensing on the optics" problem, astronomers have hundreds of years of experience dealing with this issue. The simplest approach is to slightly warm the optics using resistive heaters. As long as the optics are slightly warmer than the surroundings, any water in the air will condense somewhere else. You don't want too much heating, since then you form convective air currents above the mirror that harm the seeing conditions. However with some reasonably accurate temperature sensors and a feedback controller, the condensation problem is straightforward to solve.
The transistor is probably the best example, but there's plenty of others. It didn't make AT&T the richest company or most powerful company on the planet, but it changed our world.
This to me highlights the real questions: Did AT&T actually make a positive return on their investment in Bell Labs? If not, then why did they think it was in their best interests to fund Bell Labs in the first place?
The funny thing about technology is that the big payoff doesn't usually go to the inventor, or the scientist who did the basic research that led to the invention. People assume this, but it really isn't the case. The original inventor may receive patent royalties (if they're lucky), but the real payoff is in commercialization. Bell Labs invented transistors, just as Xerox invented ethernet and the GUI, and Sun invented Java. However none of these companies especially benefited from their innovations. We should expect the norm to be no large-scale corporate funding of basic research, not the reverse.
I suspect AT&T and RCA did what they did in order to mollify people who felt they were monopolies in their respective areas. I.e., they did this as a public good, in an effort to improve their public image and discourage regulators from brining them down. In this sense perhaps Bell Labs was a good investment for AT&T, if it delayed their breakup.
Seriously, I'd much prefer an open database of scanned works rather than letting one company negotiate a deal.
It is a nontrivial exercise to obtain high-quality scans of 20+ million books. The scanning must be done non-destructively, since nearly all of these books are out of print. This means someone/something turning pages and taking pictures. It costs most archivists hundreds of dollars to scan each book this way. Which is fine if you're the Brewster Kahle trying to compile a very small collection. If you want to do a complete job of it, it costs hundreds of millions of dollars, if not billions -- and that's if you get the scanning technology and QC pipeline right.
The question is: Who pays all that money to do the scanning?
I'm guessing Brewster Kahle would prefer that the US Government fund it. Maybe that would be nice, but I don't think it's particularly realistic. Other than that, only Google has stepped up to this effort. Microsoft quit theirs last year. If Google thought they had no legal basis to use this material, or make any money from it, I guarantee they would stop the scanning in an instant. They aren't stupid after all.
I'm guessing the "Open Book Alliance" has no intent to invest the scale of effort needed to pull this off. They're just trying to shoot torpedoes at Google.
That kind of clears up what orphan books are, but still, if no one knows who owns the rights, who are the people complaining?
I gather much of the issue some authors have with this agreement stems from a couple of concerns:
1. They feel they're being steamrollered into giving up some rights. For example, the agreement has provisions that allow Google to sell electronic access to works, and share the revenue with the authors. By signing on, they lose the right to negotiate individually with Google. Also, each public library will have one terminal where the full text of all works will be visible; some authors may perceive this as "giving away" their work. The interesting thing is that it's very likely in every author's best interest financially to participate. Nevertheless I imagine some fraction will (irrationally) opt out, just to make a statement.
2. They are concerned that Google will gain too much power within the publishing industry. The settlement appears to give Google a preferred position with regard to distributing these works. Maybe somebody else will come along and try scanning all of the same books, and negotiating a similar settlement, but then again they might not. Perhaps a lot of authors are wary of being trampled by yet another Big Corporation.
Also, to clarify with regard to orphaned works: The ownership of these xxM works is not necessarily unknowable, it just isn't known at present. You could send investigators after each work in turn, but this would be impractically expensive given the very obscure nature of a lot of these books. So therein lies the present dilemma: A lot of good content without clear ownership, and a very strict copyright regime preventing anyone from using it. View this as collateral damage from our particular copyright regime. I think the courts generally acknowledge this is a problem, and are very interested in solutions.
"Your honor, I had no idea the contents of my iTunes folder were being served to all takers through the internet. It turns out the browser I use (IE6) allowed some bad site to install a virus on my computer. I didn't have an antivirus program installed, that doesn't come with Windows. I honestly had no idea this was happening..."
Mmmm...the savory smell of plausible deniability....
The issue is not whether NASA has provided a great return on investment; it's whether human spaceflight has done so.
Consider this recent discussion on Charlie Stross's blog. None of the tangible benefits identified had to do with human spaceflight specifically.
My prediction is that there will not be a human outside of low Earth orbit for at least the next 50 years, with the possible (unlikely) exception of the Chinese attempting a lunar orbit or landing.
Perhaps in the 50-100 year timescale, we'll have figured out radically different approaches: Nuclear propulsion, a space elevator, a launch loop. Or we'll be able to upload our minds into hardware, and send people into space without sending bodies. E.g., your consciousness gets radioed into the probe once it's tunneled through the Europan ice.
First point, the human eyes and ears are not better than man-made equivalents. They are extremely good given the limitations of biology, but not nearly as good as man-made sensors like CCDs. But this is a side point. Obviously brains can learn and be effective even if their sensory inputs are degraded. The "magic" of intelligence does not lie with sensory fidelity.
Your point about gesture recognition underscores what I'm saying. You shouldn't need to have any training period at all in a gesture recognition system: It should work the first time. Our systems today are too brittle, and need to be fine-tuned. Even when they're tuned, they still make errors a 3 year old would never make. The approach I'm advocating, growing from a seed in connection with a rich environment, is something that happens before deployment time so to speak. What ships to the customer is a fully trained AI. The "training time" with an individual should be zero. How much training time do you need to understand the speech of someone you just met?
The issue regarding "what fits inside a small drone" is a reasonable one, and clearly depends on implementation details. The human brain is again maybe too lofty a goal. For image recognition and navigation tasks, a mouse is capable enough to be very interesting. What resources are required to "simulate" a mouse brain? I don't know. My feeling though is that Moore's Law will eventually solve this problem. What Moore's Law will not solve is the question of how do we properly build these systems. My point is that real intelligence needs to be grown, not designed.
Can't we just skip "concepts", and move straightaway to "meta-concepts"? After all, a "concept" is just a concept itself, so with meta-concepts we'll be able to define "concepts" recursively. Which doesn't sound like a win, until you realize then you can redefine concepts to fit your own idiosyncratic needs. Like how in my code, the first thing I always do is overload "+" to mean "*", and vice versa. I've always liked them the other way around, not sure why. Anyway, back on point: Concepts by itself is like the 4-blade razor, a lame, stupid half-measure. The real prize, the 360-dunk-off-the-free-throw-line, is 5 blades on a razor. I move we skip "concepts" and go for the big win. Those effeminate Python dorks will have no answer to this, they'll be stunned to see just how inferior they really are. Who's with me on this?
Broader point: I'm sick and tired of these language designers not giving us enough features. For the last 20 years I've been waiting for a language that will allow me to redefine keywords. If that too much to ask? What if I don't happen to like "for", or "while", or "return"? Do you people lack vision, or competence, or both? Second thing on my must-have list is a pre-pre-processor. I'm tired of writing all these header files all the time. I want a way to generate them programmatically, at compile time. A small embedded scripting language would do fine, just make sure it has templates and operator overloading and multiple inheritance, so I can stretch my legs and get comfy with it. Come on people, start earning your paychecks and get some of this stuff done!
It's the reconfigurable nature of the human brain that's unique and powerful.
Strike "humans", and I think you're on to something. We have been taking the wrong approach to AI, if the goal is to build something as adaptable and resilient as a brain. Biological brains get all their information from the environment, mostly by interacting with it. The amount of initial, hard-wired information must be very small: The entire genome is only 4 billion base pairs, and likely only a small fraction codes for the functioning of the brain. It's probably only a few hundred megabytes, at most. The secret of the brain is that it's a self-organizing system, incorporating and organizing information from the environment to build a very resilient end product. A consequence of this learning (or bootstrapping) algorithm is the adaptability and "reconfigurability" you noted.
Most traditional AI research has a lot of hard-coded information, it's very different. So the systems are brittle and not adaptable. Take an infant at birth and deprive it of all sensory input, forever. This is what we do to our computers. And we expect them to be smart?
My hope for AI is that people will figure out the learning algorithms that brains employ, and then we figure out how to start from a very small seed, couple it together with a rich and interactive environment, and end up with something interesting. Human brains are too lofty an initial goal -- I'd be happy with something as resilient as a cockroach. I can never kill those damned things.
Maybe CS departments of the future will have courses on parenting skills?
There are a couple of things being conflated in this type of research, which to me muddies the water. One question has to do with performance of people not too far from the median. For this question, I believe it's reasonable to look at how achievement test scores vary with factors like gender, race, culture, nationality, socioeconomics, and so forth. The original research cited here involves data of this type. And the conclusion isn't so surprising: Female performance relative to males is very situationally-dependent. Anecdotally one only needs to look at the gender gap (if one exists) in east Asian students vs. the gender gap in white students. *Maybe* white women are at some genetic disadvantage relative to asian women -- again relative to their respective male counterparts -- but it seems unlikely relative to a cultural factor.
What these lines of research don't really show -- because there isn't enough comparative data available -- is what are the external factors that most correlate with the gender gap within different groups. Is it culture that drives the variation? (Asians have higher expectations on daughters? Asians don't propagate the "geek stigma" as much for girls?) Is it economics? (Poorer people cannot educate all their kids, so preferentially educate the boys?) Or something else? Who knows.
The second question being conflated is performance at the far, far, end of the performance spectrum. Fields medal winners represent the 99.999+ percentile. Who knows what defines people out there? There aren't enough of them to really study as a statistical emsemble. It's fair to say that at the high end of any performance curve, a lot of things have to come together simultaneously: Raw talent, motivation, opportunity, persistence, environment, dedication. It could be for example that men have no more innate ability than women, but are just more single-minded in their approach to life. I.e., more men than women are willing to do what Andrew Wiles did, namely hole up in an attic for 10 years to prove Fermat's Last Theorem (with a low probability of success).
Finally, I think with regard to this sort of research it's important to maintain a dispassionate attitude. When I get the feeling the authors are trying to *advocate* for a particular conclusion, that makes me a bit queasy. There seems to be this unstated assumption that an unequal outcome is indicative of unequal opportunity. Would anyone argue that the relative lack of white men in the NBA is indicative of low opportunity or discrimination? Probably not. Perhaps white women don't pursue math at the highest levels because they simply don't want to, compared with other uses for their time. Is this a bad outcome? Within the scientific enterprise it's a very slippery slope to start asserting value judgments about these things.
The 3DV site has a number of papers showing how the underlying technology works. They've spent the last several years refining it to make it smaller/cheaper, but the physical principles are the same: In addition to a regular video camera to capture RGB, a pulsed infrared source illuminates the entire scene. A fast shutter on the camera then allows it to infer distance to each pixel, based on the time-of-flight of the reflected infrared pulse. The literature for an earlier version claims 1-2 cm depth resolution at full video rates, but I haven't seen specs for the latest small/compact/cheap version.
Using this as a game controller strikes me as potentially very interesting, but with caveats. First I think this would be hard to program: You get all of this per-pixel RGB/depth information out of the camera, from which you need to extract limb orientations, etc. Seems pretty hard to do well, although I suppose a library would abstract that away from the game developer. A more fundamental limitation is the low spatial resolution: Basically it just detects gross movements of the limbs, and by itself could only support a few types of games. I think this would be most useful in combination with a controller for faster/finer motions (e.g., a button-based interface). Maybe that's where the future is headed: Overall body tracking at low resolution, selected "high-sensitivity" parts (hands, eyes) tracked at high resolution.
Another potentially interesting next-generation game controller is the Sixense 6dof controller. It is basically a magnetic tracker (full absolute position/orientation sensing) combined with a button-based interface. I could see a lot of interesting games working well on that. At any rate, what's cool about the success of the Wii is that people are innovating here, and the console makers now seem willing to spend a nontrivial part of their overall COGS on controllers.
I agree, this article as written makes no sense. To "open source an API" is muddy thinking, a non-concept. At least in the US, there is well-established legal precedent allowing companies to duplicate APIs at a functional level. E.g., the function of APIs is not copyrightable in the way that source code is. So anybody that wants to can come along and implement their own versions of the EC2, S3, etc. API. They don't need any approval from Amazon.
Now Amazon may decide it's in their best interest if other cloud providers adopted their APIs. Presumably they would do this to encourage more companies to adopt cloud computing (i.e., eliminate lock-in as a risk). They could advance their APIs as standards in a number of ways, including making some of their own specific implementations open-source. Is this what the article is trying to say?
IMHO the main thing is to be as diverse as possible in where the data is located and how it's managed: Geographies, access controls, software, etc. Eliminate as many common failure modes as possible.
For the servers I run, here's what I do. I don't see any real flaws with this procedure, and it's simple to set up:
This gives three independent copies of the data in very different locations. If the live data is corrupted, I can restore from the backups on the main server and lose at most an hour of data. (More accurately, an hour plus however long it takes me to notice the corruption, and then do a recovery. I haven't gone to great lengths to streamline these operations, although obviously one could.) If the entire main server is wiped, for example by a hacker, I lose at most a day of data. It's hard for me to envision a scenario where recovery would be impossible.
The why of this is fairly straightforward from a financial standpoint. Companies can raise money from two sources: Equity and debt. The cost of debt is obvious (the interest rate). The cost of equity is less obvious but very real: Investors demand a particular total rate of return on the money they invest in a stock, either in the form of dividend payments or retained earnings (appreciation in the value of the stock). If the total rate of return from your stock is less than what the market demands (based on its perception of how risky you are), then your stock price will fall until the desired rate of return is met. Typical long-term total return from the stock market is 9-10%, and for a tech company most investors will want more because of the perceived risk.
Anyway, the point is that when interest rates are low, it's a lot cheaper to get money from debt markets than from equity markets. So the smart CFO will borrow money and use it to buy back (and retire) stock. If you're a shareholder you like this in net, because although the company now has debt to pay back (a liability which decreases the value of your shares), the positive impact on value from having fewer shares outstanding outweighs it. The only downside to this strategy is that interest on debt must be paid back on a defined schedule -- bond holders aren't willing to defer their payoff like equity investors are (and consequently bond investors make lower returns on average). GM is an object lesson in getting squeezed this way. Many tech companies avoid long-term debt as a result; they don't like the ongoing obligation. If anything this move by Microsoft signals to the market that they've become a stable business that is confident in its long-term ability to generate cash.
Consider the various moral choices in Fallout 3. Functionally, the game allows you to decide what you want to be. If you want to be a slaver? It is possible. If you want to, instead, rescue slaves? Very much also possible.
These "moral choices" are very superficial in today's games. There are no real, long-term consequences for behavior, as there is in the real world. In truth, violent people don't get away with it very long. If you do something bad to someone, they remember and tell other people, and your bad reputation spreads. If you drive GTA-style in a typical city you'll be shut down very quickly. Even in wartime there are codes of conduct; if you shoot your buddy, or an innocent bystander, there are big consequences. Games today don't represent any of these nuances. You shoot a bad guy, and all the rest just keep doing what they were doing. You shoot a good guy, and nothing much happens. I would never advocate for outlawing or banning anything, and I don't particularly fault game developers for giving people what they want. That said, I don't think the end product is very instructive for kids. Adults can see these games for what they are -- escapist action with no connection to reality -- but a very young kid doesn't yet have the context to see that.
With regard to "protecting the children", there is an interesting (to me) difference between European and American perceptions. European children's literature is much more willing to touch on darker themes than its American counterpart; authors like Roald Dahl, J.K. Rowling, Philip Pullman, and the brothers Grimm address things you'd almost never see in any American kids books. As a parent I feel this is fine, however, since these works usually illustrate something true about the consequences of behavior, both good and bad. Bad things happen in Willy Wonka's chocolate factory, when the kids are greedy or mean.
Back on games, I think there is a coming backlash from consumers who are getting bored with the typical shooters. I like shooters as much as the next guy, but why do 95% of the top titles have to follow this formula? I used to buy a lot of X-Box games, until I realized they're all basically the same game (Portal excluded). Even things as basic as the color palette; why are all games so dark (literally)?
"there are many equivalent and roughly as-challenging problem areas out there, and there seems to be a great deal of work being done on neuroscience and related fields, with breakthroughs announced on a daily basis. What's driving the interest, time and money being devoted into neuroscience, as opposed to, say, cancer immunology.... ?
A key fact about science is that the ultimate utility of any line of work is impossible to predict. Sometimes the fields that people expect to yield big dividends have modest results (classical AI, superstring theory, fusion power generation). In other cases the importance of a piece of work ends up being much greater than anticipated (the maser, the transistor, cosmic microwave background radiation).
In the face of this uncertainty, you want to do science like you invest your 401(k) -- spread your money around on a lot of different bets.
Quite a few people in the industry now are starting to care about power efficiency at the other end of the performance spectrum too. The Green 500 list for example tracks Megaflops per Watt data for the top 500 supercomputers. Judging from this data the Cell processor looks very good.
The reasons for caring about energy efficiency at the high end are of course very different from what ARM is trying to do, which is to maximize performance within a given battery life envelope. For large installations it has more to do with operating cost and environmental concerns.
I really want to remain optimistic about it but for me this whole orion project is like a reminder of where we *could* have been at the completion of the Apollo launchers.
What's lacking is a clear impetus for the manned program. Since the early 1970's it's been viewed by Congress as, at best, a jobs program to stimulate their local districts. (Note how NASA facilities are cleverly distributed across the country, to maximize the base of support. And note the gaping lack of any long-term objective behind initiatives like the ISS, or Orion for that matter.)
Personally I don't believe overcrowding, asteroid mining, or any other economic incentives will ever be powerful enough to drive manned spaceflight beyond low earth orbit. It's just too expensive, plain and simple. (Something like the space elevator could maybe change that, but that's a separate topic.) Scientific knowledge is a valid objective, but it tends to argue in favor of much cheaper unmanned missions. There are many things a human geologist could do on Mars that current rovers cannot, but is it worth a 100x cost multiplier? Most scientists would say no.
The only thing that can provably drive a major push into manned spaceflight is what did it the first time, namely competition from another country. I predict we won't advance beyond the Apollo level of capability until another country is knocking on that same door, and there is a perceived land grab at stake. At present I think China is the likeliest candidate.