I spent several years in movie theatres. Most customers couldn't tell the difference between 4k digital and a base-scratched print with a strobing bulb and too much jitter and weave.
I take that back. It's not that they couldn't tell the difference, if you did a study where you showed them side by side or one after another, but 95% of them aren't going to notice the difference in quality unless it is pointed out.
I would imagine 24 fps/48 fps will be the same story. One after another, or side by side, it'll be obvious. Otherwise, only enthusiasts are going to notice (or even know that such a difference might possibly exist in the first place).
I was going to comment to this effect, but you've already stated it very well.
Seems to me that deciding what science we think is too dangerous is, in itself, probably the most dangerous line of scientific inquiry. Unless you think you can actually stop scientific advancement, you are better off spending your time trying to figure out how to best avoid the potential negatives that come with the inevitable advancement.
Good luck with the wild goose chase.
Hint: Hackers move a lot quicker than government schemes. This is a fundamental law of nature. Hackers will circumvent your next measure without waiting for a vote in parliament. Keep invading privacy to catch people downloading songs and you'll just advance the tech for those who would do more malicious things with their anonymity. Oh, and fundamental law #2: This is 2012. When some brilliant hacker does figure out a new scheme for getting around your snooping, someone will package it into binaries with adorable icons and it'll be on every college student's desktop (followed by their parents) in no time.
I think some people are missing the point: This system isn't making automated decisions for the city on a scale that isn't already being done in many places (automatic adjustment of traffic lights based on traffic conditions, for instance).
Most of this is the aggregation of data which is then displayed in some form of useful visualization or is synthesized into some supposedly useful metric. There are still human decision makers. If the system is taking a flood of data and is even mildly successful at alerting decision makers when there is a problem that needs further action, what is the issue?
My field is machine learning and decision support. This is a decision support application. Decision support is about getting the relevant knowledge to decision makers. As to what knowledge is relevant, and how to transform data into that knowledge, that's really at the center of what IBM is trying to do. Traffic control systems, epidemiological 'heat maps,' surveillance cameras, etc... are already widely deployed around the world. This system just routes them onto TV screens in the same room and tries to make some sense of some of it. It has not the power to enslave, nor is it capable of launching a nuclear attack.
There are a lot of comments indicating that retractions and poor research must always be the result of fraud.
Another explanation is just flat-out bad science performed by scientists who are not being held to rigorous standards. I had the privilege of learning how to critically review scientific literature from a well-regarded epidemiologist with no tolerance for bullshit and decades of experience on a variety of review boards critiquing medical literature. I learned that if you look closely, you'll find at least one suspicious element in probably 90% of the papers you read. In many cases (perhaps most), these suspicious elements are just missing things that should have been present (such as how randomization was done in a trial, or what method was used to blind participants) but were probably done adequately anyways (just poor authorship, but a decent study). In other cases, missing elements are missing because required steps weren't done or there is an effort to hide negative results (fraud). One of the tell-tale signs that you are reading a crappy paper is when any statistical tests are only briefly discussed and only p-values are reported (or, even worse, they just tell you whether or not the statistics supported their hypothesis).
If I were developing a product or doing research that depended on other scientific research, I would read it very carefully before applying it. Any scientific journal worth reading will require contact information for a corresponding author, whom you can contact with any questions. If that author won't answer questions or provide their dataset for verification (whenever applicable and legal to do so), you should throw the study away and write the journal about the flaws in the study and the author's failure to appropriately respond.
There ARE mechanisms in place for weeding this crap out, but part of the problem with the scientific literature right now is the lack of critical ability in the general readership. A lack of funding for - and unwillingness to publish - research that merely confirms or conflicts with existing research is another issue. I would say both of these issues, and the fact that most types of studies don't have accepted standards which define a high-quality study (randomized trials in medicine and the CONSORT statement are a notable exception) are larger than the issue of blatant fraud, since academic fraud is one of the reasons many universities can use to revoke tenure (the ultimate goal of most academics).
Gee, I hope they succeed in bringing all of the call center work back to the United States! I would love to pay higher prices so that my countrymen can work menial jobs for 10 times the pay of their Asian counterparts (which still fails to provide a decent standard of living here), because more expensive products are good for everyday Americans already struggling to get by!
It is important to note where the primary concern of most of the commenters is: the stolen SSNs.
We don't have effective health information exchange because politicians and their constituents are scared to death of their all-important "private health data" being stolen. When it actually happens, people stop and realize that no one could possibly have any use for Joe Average's health information, whereas your SSN/personal information can quickly compromise your financial livelihood.
In order to get some use out of stolen health data, you'd have to sell it to some marketer (who would be outing themselves just by using it....) or you'd have to blackmail the person whose data you have (a felony/they probably don't have enough money to make it worth it/they are certain to be caught if they try to do this at any scale).
To get some use out of stolen SSNs/personal information, you need to fill out a few online forms and start ordering.
Of course, there are thousands (if not millions) of organizations storing SSN/Credit Card Numbers/Driver License or Passport Numbers/Addresses/etc... on tons of people. For some reason, we are OK with that risk, but up in arms when we talk about storing potentially life-saving health data.
I fully expect many to agree with this post....and I fully expect the usual flame response when I post anywhere online that your health data is not sacred, no one who could feasibly use the stolen data can legally do so, and unless you are a high-priority target (celebrity, political figure, etc...) you really don't have any risk from having your health data stolen (although it should certainly still be secured unless you really want to make it public data).
Using your example, knowing the difference between 12872 and 15000 miles can be _extremely_ important, especially since most low mileage leases tack on a surcharge for every mile driven over 12000.
Of course, you don't need this kind of temporal data to make that kind of decision as long as you are relatively stable in location/job/etc... (yearly odometer readings would do just as well).
I think my original reply may not have been very clear: I DO NOT suggest that others track, nor do I personally track, all of the information from my original reply.
More realistically, totaling receipts for repairs/insurance/car payments (something anyone with half of a brain and rudimentary ability to budget already does or could retrieve quickly) and taking measurements over a few tanks of gas in the winter (heater), spring (no air/heat), and summer (air conditioning), can probably tell you all you would need to know to make a highly accurate judgement of your costs of owning a car. You could probably just measure fuel economy over a tank or two of typical driving, regardless of time of year, and make an intuitive adjustment based on whether or not you were using the a/c heavily or stuck in traffic more than usual.
Personally, I divide (in my head, to maybe +/-.2 mpg accuracy) the miles driven on my trip meter by the gas used when I fill up my tank, and I know that I'm getting right around 29 mpg when I use the a/c and about 31 when I don't. If I figure that I use a/c a little over half the year, I'm getting around 30 mpg overall. If I go to fueleconomy.gov, I can see this is a little higher than the 28 mpg EPA combined rating for my car and can use that data, along with the number of miles I drive in a year (about 30k) and a rough estimate of what I expect gas prices to be (this estimate would negate any effort to measure fuel economy precisely, btw) to project that cost of my current vehicle vs. another vehicle I plan on buying.
For those who think I spend all day doing this, I spent longer typing this reply to explain it than I ever have on actually doing the calculations or taking the 30 seconds to look up information on fueleconomy.gov. Anyone with a spreadsheet and a clue should be able to figure it out.
The suggestion to track this at a finer granularity was just a recommendation to the OP who referenced the original article by Wolfram (an article which, by the way, I don't think anyone replying to my message read or you wouldn't be accusing me of overkill).
Did you read the original article?
A lot of the shifts in data that Wolfram saw were due to major events in his life (writing a book, changing roles at work, etc...).
The goal isn't to help you remember major events in your life, it is to observe how those changes affected other aspects of your life.
Were you just being funny or did you really not grasp this?
Wrong.
You know what the city and highway fuel economy ratings are, and you have a "combined" estimate that reflects the average consumer's mix of highway/city driving.
This is why fueleconomy.gov has a feature to help you estimate your fuel economy with a new car....and, like I alluded to, it's based on what you are getting with your current car in comparison to the fuel economy ratings on your current car.
Put a little more thought into your reply next time.
Tracking how much gas you put into your tank, how many miles you drove on that gas, and when you put it in can be highly useful in a lot of ways:
From a personal standpoint, this sort of historical data can reveal some interesting trends similar to what Wolfram saw in his data. For example, a large increase/decrease in mileage might indicate a move, marriage, or job change.
From a financial standpoint, knowing exactly how much gas you are consuming can help you make a more intelligent decision when purchasing a car or considering other transportation alternatives. You can use information about your mileage to extrapolate what your mileage would be in a car you are considering purchasing, or you can use miles driven to determine whether or not you would be able to stick to a low-mileage lease without paying overage charges.
Keeping track of insurance payments, car payments, and every repair you do to the car can also help in determining total cost of ownership, which can again help you make reasonable decisions when considering transportation alternatives.
You can get a rough estimate of this data by looking over past credit card statements and figuring in changes in the cost of gas over time, but it won't be nearly as accurate as manual tracking.
I don't know why standard methods wouldn't cut it. Testing half a million SNPs and several combinations of those SNPs in a GWAS study is far more convoluted and controlling FDR through permutation is still relatively straightforward (though maybe computationally expensive...thought I doubt that is a huge problem for Stanford researchers).
And your numbers are inflated by probably an order of magnitude even if no correction was done. They would not be testing all drug/ADE combinations because the vast majority of ADEs would not be present for a given drug and therefore wouldn't be tested.
Not to plug my profession or anything, but this is exactly why the entire field of biomedical informatics exists.
If you think this is bad, consider the fact that there are currently over 20 million abstracts in PubMed....do you think even 10% of that has actually been properly synthesized into operational knowledge and applied to patient care?
And we won't even go into genomic data, or even the amount of records that one patient might accumulate in their EMR over the span of a lifetime, or the fact that a 320 slice CT generates so many layers of images that they can't all be carefully reviewed (and an abnormality may be so small it only appears in a couple of them), or the overwhelming breadth and depth of surveillance data collected from ERs/pharmacies/drugstores/monitoring stations/schools/etc... by public health practitioners.
There is a critical challenge in biomedicine to distill useful knowledge from all of this data...and it's akin to drinking from a firehose.
No one is going to read the 329 warnings for the drug, but in an ideal world we'll be able to identify genetic indicators that make you more or less susceptible to certain side effects (pharmacogenomics) and present this information to you/your doctor (and no one has to read the booklet that comes with the prescription).
In Biomedicine you tend to see a heavy reliance on T-Tests, Chi-Square variants, Fisher's Exact, regression, McNemar's and Cox Proportional Hazards when temporally rich data is being tested.
I don't have access to this article yet, but I would be surprised if they weren't performing a paired T-test in situations where outcome variables were measured on a ratio scale, McNemar's for binary outcomes where temporal data is not provided (maybe rare or nonexistent in this study), and Cox Proportional Hazards if there are any cases where we have a long temporal history of the data.
Based on the sheer number of hypotheses tested we would expect to see some sort of correction for multiple testing here, too.
This is old news:
http://www.genewscenter.com/Press-Releases/GE-Healthcare-Unveils-Ultra-Low-Dose-CT-Technology-with-Profound-Image-Clarity-3367.aspx
And if you read through the Intel and GE press releases, you'll find numbers all over the map as to how much this actually decreases radiation exposure. It might be a 4x reduction (GE scientist quoted in Intel article), it might be a 10x reduction (Intel article), or it might be a 100x reduction (GE article). It might just depend on the specific scan being done, but you won't find that in either article.
This is quite a breakthrough and is fantastic news for anyone who needs regular scans, but it is a bit overstated (as mentioned by other commenters, CT scans had greatly improved from the doses quoted by GE/Intel as baseline figures), is nearing 2 years of availability in some areas, and is going to see slow adoption as hospitals a) aren't willing to part with the expensive, working machines they have and b) the time requirement of an hour is still significant enough to rule this out in many environments.
There is still a key issue that isn't mentioned here. It might take 15 minutes to get a prepped patient in the room, to scan them, and then to do all necessary work to prepare for another patient, but it takes an hour to analyze a scan. So, we either get 4 servers to process the scans (I think Intel would like this very much) or we run under capacity or we leave a long queue of scans to be analyzed overnight and read the next day. The 1st option is expensive, the 2nd option is not going to happen if there is a living accountant in the building, and the third option would still require an extra server (assuming we do more than 24 scans each day) and is only workable if there can be a day+ delay in sending results.
Perhaps some Slashdotter will make a fortune with their CT analysis "cloud."
And not to sound like a party-pooper, but for customers in the U.S. GE/Intel still have no power over the insurance approval process which can delay your scan by another order of magnitude.
Let's celebrate this for what it is, a nice improvement for a small niche of healthcare consumers who require several CT scans in their lifetime, and not for the miracle-machine that the GE/Intel marketing department would have us believe.
BAliBASE is a great reference, but all of the sequence alignments in the database were refined from algorithmically-derived alignments (implemented on computers) in the first place. I think it furthers my assertion that computers + humans > either alone when it comes to MSA. Certainly, the sheer scale of the data would prevent any sort of economic use of manual global alignment, even if the local alignments were best carried out by biologists.
Again, my issue here is that the article gives the impression that gamers have "outdone" computers at matching up disease genes, when in reality the gamers have been presented with a very small slice of the problem (as I'm sure you recognize better than I) and only outperformed the computer alone in certain scenarios, certainly not the blanket 70% quoted in the news piece.
Agreed.
I would be interested to see what the researchers learned from this exercise in terms of improving MSA algorithms.
Perhaps the performance of the human players suggests that aligning a small subset of the problem with a high quality alignment algorithm before completing the problem with a run-of-the-mill algorithm is the way to go.
The fact that puzzles completed repeatedly were where the phylo solutions performed best would indicate that running this first algorithm repeatedly with some element of random error might lead to a better solution. Whether or not this is computationally feasible is another question that begs to be answered.
In the meantime, it would be interesting to see a "puzzle of the day" where researchers can upload a current MSA problem they need a good solution to in order to use phylo to help with current research questions.
This is an interesting finding, but let's not get too carried away.
If you read the article, you'll see that:
a) The phylo-based alignments are partial solutions. They are simplified for the human user by leaving many orthologous sequences out of the alignment. This means there is another algorithm that finishes these partial solutions before they can be compared to solutions produced solely by algorithms.
b) Only 36% of the _best_ phylo-based solutions, once completed, were better than the algorithms' solutions.
This is still an improvement, but it DOES NOT suggest that humans are better than computers at multiple sequence alignment. If you were to ever try to solve a real MSA problem by hand, you would quickly understand how completely hopeless it is. In fact, even aligning 2 sequences of any appreciable length by hand is a chore.
The problem here is the misguided title: "Gamers outdo computers at matching up disease genes" which should read: "Gamers + computer outdo computers only at matching up very small fragments of disease genes, some of the time"
Slashdot Mirror
I spent several years in movie theatres. Most customers couldn't tell the difference between 4k digital and a base-scratched print with a strobing bulb and too much jitter and weave. I take that back. It's not that they couldn't tell the difference, if you did a study where you showed them side by side or one after another, but 95% of them aren't going to notice the difference in quality unless it is pointed out. I would imagine 24 fps/48 fps will be the same story. One after another, or side by side, it'll be obvious. Otherwise, only enthusiasts are going to notice (or even know that such a difference might possibly exist in the first place).
I was going to comment to this effect, but you've already stated it very well. Seems to me that deciding what science we think is too dangerous is, in itself, probably the most dangerous line of scientific inquiry. Unless you think you can actually stop scientific advancement, you are better off spending your time trying to figure out how to best avoid the potential negatives that come with the inevitable advancement.
Good luck with the wild goose chase. Hint: Hackers move a lot quicker than government schemes. This is a fundamental law of nature. Hackers will circumvent your next measure without waiting for a vote in parliament. Keep invading privacy to catch people downloading songs and you'll just advance the tech for those who would do more malicious things with their anonymity. Oh, and fundamental law #2: This is 2012. When some brilliant hacker does figure out a new scheme for getting around your snooping, someone will package it into binaries with adorable icons and it'll be on every college student's desktop (followed by their parents) in no time.
I think some people are missing the point: This system isn't making automated decisions for the city on a scale that isn't already being done in many places (automatic adjustment of traffic lights based on traffic conditions, for instance). Most of this is the aggregation of data which is then displayed in some form of useful visualization or is synthesized into some supposedly useful metric. There are still human decision makers. If the system is taking a flood of data and is even mildly successful at alerting decision makers when there is a problem that needs further action, what is the issue? My field is machine learning and decision support. This is a decision support application. Decision support is about getting the relevant knowledge to decision makers. As to what knowledge is relevant, and how to transform data into that knowledge, that's really at the center of what IBM is trying to do. Traffic control systems, epidemiological 'heat maps,' surveillance cameras, etc... are already widely deployed around the world. This system just routes them onto TV screens in the same room and tries to make some sense of some of it. It has not the power to enslave, nor is it capable of launching a nuclear attack.
There are a lot of comments indicating that retractions and poor research must always be the result of fraud. Another explanation is just flat-out bad science performed by scientists who are not being held to rigorous standards. I had the privilege of learning how to critically review scientific literature from a well-regarded epidemiologist with no tolerance for bullshit and decades of experience on a variety of review boards critiquing medical literature. I learned that if you look closely, you'll find at least one suspicious element in probably 90% of the papers you read. In many cases (perhaps most), these suspicious elements are just missing things that should have been present (such as how randomization was done in a trial, or what method was used to blind participants) but were probably done adequately anyways (just poor authorship, but a decent study). In other cases, missing elements are missing because required steps weren't done or there is an effort to hide negative results (fraud). One of the tell-tale signs that you are reading a crappy paper is when any statistical tests are only briefly discussed and only p-values are reported (or, even worse, they just tell you whether or not the statistics supported their hypothesis). If I were developing a product or doing research that depended on other scientific research, I would read it very carefully before applying it. Any scientific journal worth reading will require contact information for a corresponding author, whom you can contact with any questions. If that author won't answer questions or provide their dataset for verification (whenever applicable and legal to do so), you should throw the study away and write the journal about the flaws in the study and the author's failure to appropriately respond. There ARE mechanisms in place for weeding this crap out, but part of the problem with the scientific literature right now is the lack of critical ability in the general readership. A lack of funding for - and unwillingness to publish - research that merely confirms or conflicts with existing research is another issue. I would say both of these issues, and the fact that most types of studies don't have accepted standards which define a high-quality study (randomized trials in medicine and the CONSORT statement are a notable exception) are larger than the issue of blatant fraud, since academic fraud is one of the reasons many universities can use to revoke tenure (the ultimate goal of most academics).
Gee, I hope they succeed in bringing all of the call center work back to the United States! I would love to pay higher prices so that my countrymen can work menial jobs for 10 times the pay of their Asian counterparts (which still fails to provide a decent standard of living here), because more expensive products are good for everyday Americans already struggling to get by!
It is important to note where the primary concern of most of the commenters is: the stolen SSNs. We don't have effective health information exchange because politicians and their constituents are scared to death of their all-important "private health data" being stolen. When it actually happens, people stop and realize that no one could possibly have any use for Joe Average's health information, whereas your SSN/personal information can quickly compromise your financial livelihood. In order to get some use out of stolen health data, you'd have to sell it to some marketer (who would be outing themselves just by using it....) or you'd have to blackmail the person whose data you have (a felony/they probably don't have enough money to make it worth it/they are certain to be caught if they try to do this at any scale). To get some use out of stolen SSNs/personal information, you need to fill out a few online forms and start ordering. Of course, there are thousands (if not millions) of organizations storing SSN/Credit Card Numbers/Driver License or Passport Numbers/Addresses/etc... on tons of people. For some reason, we are OK with that risk, but up in arms when we talk about storing potentially life-saving health data. I fully expect many to agree with this post....and I fully expect the usual flame response when I post anywhere online that your health data is not sacred, no one who could feasibly use the stolen data can legally do so, and unless you are a high-priority target (celebrity, political figure, etc...) you really don't have any risk from having your health data stolen (although it should certainly still be secured unless you really want to make it public data).
Using your example, knowing the difference between 12872 and 15000 miles can be _extremely_ important, especially since most low mileage leases tack on a surcharge for every mile driven over 12000. Of course, you don't need this kind of temporal data to make that kind of decision as long as you are relatively stable in location/job/etc... (yearly odometer readings would do just as well). I think my original reply may not have been very clear: I DO NOT suggest that others track, nor do I personally track, all of the information from my original reply. More realistically, totaling receipts for repairs/insurance/car payments (something anyone with half of a brain and rudimentary ability to budget already does or could retrieve quickly) and taking measurements over a few tanks of gas in the winter (heater), spring (no air/heat), and summer (air conditioning), can probably tell you all you would need to know to make a highly accurate judgement of your costs of owning a car. You could probably just measure fuel economy over a tank or two of typical driving, regardless of time of year, and make an intuitive adjustment based on whether or not you were using the a/c heavily or stuck in traffic more than usual. Personally, I divide (in my head, to maybe +/- .2 mpg accuracy) the miles driven on my trip meter by the gas used when I fill up my tank, and I know that I'm getting right around 29 mpg when I use the a/c and about 31 when I don't. If I figure that I use a/c a little over half the year, I'm getting around 30 mpg overall. If I go to fueleconomy.gov, I can see this is a little higher than the 28 mpg EPA combined rating for my car and can use that data, along with the number of miles I drive in a year (about 30k) and a rough estimate of what I expect gas prices to be (this estimate would negate any effort to measure fuel economy precisely, btw) to project that cost of my current vehicle vs. another vehicle I plan on buying.
For those who think I spend all day doing this, I spent longer typing this reply to explain it than I ever have on actually doing the calculations or taking the 30 seconds to look up information on fueleconomy.gov. Anyone with a spreadsheet and a clue should be able to figure it out.
The suggestion to track this at a finer granularity was just a recommendation to the OP who referenced the original article by Wolfram (an article which, by the way, I don't think anyone replying to my message read or you wouldn't be accusing me of overkill).
Did you read the original article? A lot of the shifts in data that Wolfram saw were due to major events in his life (writing a book, changing roles at work, etc...). The goal isn't to help you remember major events in your life, it is to observe how those changes affected other aspects of your life. Were you just being funny or did you really not grasp this?
Wrong. You know what the city and highway fuel economy ratings are, and you have a "combined" estimate that reflects the average consumer's mix of highway/city driving. This is why fueleconomy.gov has a feature to help you estimate your fuel economy with a new car....and, like I alluded to, it's based on what you are getting with your current car in comparison to the fuel economy ratings on your current car. Put a little more thought into your reply next time.
Tracking how much gas you put into your tank, how many miles you drove on that gas, and when you put it in can be highly useful in a lot of ways: From a personal standpoint, this sort of historical data can reveal some interesting trends similar to what Wolfram saw in his data. For example, a large increase/decrease in mileage might indicate a move, marriage, or job change. From a financial standpoint, knowing exactly how much gas you are consuming can help you make a more intelligent decision when purchasing a car or considering other transportation alternatives. You can use information about your mileage to extrapolate what your mileage would be in a car you are considering purchasing, or you can use miles driven to determine whether or not you would be able to stick to a low-mileage lease without paying overage charges. Keeping track of insurance payments, car payments, and every repair you do to the car can also help in determining total cost of ownership, which can again help you make reasonable decisions when considering transportation alternatives. You can get a rough estimate of this data by looking over past credit card statements and figuring in changes in the cost of gas over time, but it won't be nearly as accurate as manual tracking.
I don't know why standard methods wouldn't cut it. Testing half a million SNPs and several combinations of those SNPs in a GWAS study is far more convoluted and controlling FDR through permutation is still relatively straightforward (though maybe computationally expensive...thought I doubt that is a huge problem for Stanford researchers). And your numbers are inflated by probably an order of magnitude even if no correction was done. They would not be testing all drug/ADE combinations because the vast majority of ADEs would not be present for a given drug and therefore wouldn't be tested.
Not to plug my profession or anything, but this is exactly why the entire field of biomedical informatics exists. If you think this is bad, consider the fact that there are currently over 20 million abstracts in PubMed....do you think even 10% of that has actually been properly synthesized into operational knowledge and applied to patient care? And we won't even go into genomic data, or even the amount of records that one patient might accumulate in their EMR over the span of a lifetime, or the fact that a 320 slice CT generates so many layers of images that they can't all be carefully reviewed (and an abnormality may be so small it only appears in a couple of them), or the overwhelming breadth and depth of surveillance data collected from ERs/pharmacies/drugstores/monitoring stations/schools/etc... by public health practitioners. There is a critical challenge in biomedicine to distill useful knowledge from all of this data...and it's akin to drinking from a firehose. No one is going to read the 329 warnings for the drug, but in an ideal world we'll be able to identify genetic indicators that make you more or less susceptible to certain side effects (pharmacogenomics) and present this information to you/your doctor (and no one has to read the booklet that comes with the prescription).
In Biomedicine you tend to see a heavy reliance on T-Tests, Chi-Square variants, Fisher's Exact, regression, McNemar's and Cox Proportional Hazards when temporally rich data is being tested. I don't have access to this article yet, but I would be surprised if they weren't performing a paired T-test in situations where outcome variables were measured on a ratio scale, McNemar's for binary outcomes where temporal data is not provided (maybe rare or nonexistent in this study), and Cox Proportional Hazards if there are any cases where we have a long temporal history of the data. Based on the sheer number of hypotheses tested we would expect to see some sort of correction for multiple testing here, too.
This is old news: http://www.genewscenter.com/Press-Releases/GE-Healthcare-Unveils-Ultra-Low-Dose-CT-Technology-with-Profound-Image-Clarity-3367.aspx And if you read through the Intel and GE press releases, you'll find numbers all over the map as to how much this actually decreases radiation exposure. It might be a 4x reduction (GE scientist quoted in Intel article), it might be a 10x reduction (Intel article), or it might be a 100x reduction (GE article). It might just depend on the specific scan being done, but you won't find that in either article. This is quite a breakthrough and is fantastic news for anyone who needs regular scans, but it is a bit overstated (as mentioned by other commenters, CT scans had greatly improved from the doses quoted by GE/Intel as baseline figures), is nearing 2 years of availability in some areas, and is going to see slow adoption as hospitals a) aren't willing to part with the expensive, working machines they have and b) the time requirement of an hour is still significant enough to rule this out in many environments. There is still a key issue that isn't mentioned here. It might take 15 minutes to get a prepped patient in the room, to scan them, and then to do all necessary work to prepare for another patient, but it takes an hour to analyze a scan. So, we either get 4 servers to process the scans (I think Intel would like this very much) or we run under capacity or we leave a long queue of scans to be analyzed overnight and read the next day. The 1st option is expensive, the 2nd option is not going to happen if there is a living accountant in the building, and the third option would still require an extra server (assuming we do more than 24 scans each day) and is only workable if there can be a day+ delay in sending results. Perhaps some Slashdotter will make a fortune with their CT analysis "cloud." And not to sound like a party-pooper, but for customers in the U.S. GE/Intel still have no power over the insurance approval process which can delay your scan by another order of magnitude. Let's celebrate this for what it is, a nice improvement for a small niche of healthcare consumers who require several CT scans in their lifetime, and not for the miracle-machine that the GE/Intel marketing department would have us believe.
BAliBASE is a great reference, but all of the sequence alignments in the database were refined from algorithmically-derived alignments (implemented on computers) in the first place. I think it furthers my assertion that computers + humans > either alone when it comes to MSA. Certainly, the sheer scale of the data would prevent any sort of economic use of manual global alignment, even if the local alignments were best carried out by biologists. Again, my issue here is that the article gives the impression that gamers have "outdone" computers at matching up disease genes, when in reality the gamers have been presented with a very small slice of the problem (as I'm sure you recognize better than I) and only outperformed the computer alone in certain scenarios, certainly not the blanket 70% quoted in the news piece.
Agreed. I would be interested to see what the researchers learned from this exercise in terms of improving MSA algorithms. Perhaps the performance of the human players suggests that aligning a small subset of the problem with a high quality alignment algorithm before completing the problem with a run-of-the-mill algorithm is the way to go. The fact that puzzles completed repeatedly were where the phylo solutions performed best would indicate that running this first algorithm repeatedly with some element of random error might lead to a better solution. Whether or not this is computationally feasible is another question that begs to be answered. In the meantime, it would be interesting to see a "puzzle of the day" where researchers can upload a current MSA problem they need a good solution to in order to use phylo to help with current research questions.
This is an interesting finding, but let's not get too carried away. If you read the article, you'll see that: a) The phylo-based alignments are partial solutions. They are simplified for the human user by leaving many orthologous sequences out of the alignment. This means there is another algorithm that finishes these partial solutions before they can be compared to solutions produced solely by algorithms. b) Only 36% of the _best_ phylo-based solutions, once completed, were better than the algorithms' solutions. This is still an improvement, but it DOES NOT suggest that humans are better than computers at multiple sequence alignment. If you were to ever try to solve a real MSA problem by hand, you would quickly understand how completely hopeless it is. In fact, even aligning 2 sequences of any appreciable length by hand is a chore. The problem here is the misguided title: "Gamers outdo computers at matching up disease genes" which should read: "Gamers + computer outdo computers only at matching up very small fragments of disease genes, some of the time"