Slashdot Mirror
Slashback: Counterstrike, Identification, Patenxtortion
False negatives, false positives, anda false sense of assurance. coryboehne writes: "TechNews has a report on the face recognition system installed at the Palm Beach Internation Airport early results of face-recognition surveillance suggest the technology is proving once again to be unreliable.
The ACLU said the first four weeks of testing at the Palm Beach airport showed the technology was "less accurate than a coin toss." The system matched the faces of the volunteers just 455 out of 958 times, or about 47 percent of the time.
Seems to me that this is a controlled environment for the most part, and still they have problems this big? I wonder if this technology will ever be accurate enough to work properly. I suppose the biggest problem is the size of the database that would be necessary to hold the high quality pictures necessary for accurate identification.
However I must admit that I am rather glad that this is'nt working yet as I'm not too sure I even like the idea of being able to digitally locate and track anyone within range of a camera."
This is what's meant by "repeat offender." Audent writes: "Following on from this story on Slashdot about PanIP's nasty habits, InfoWorld is running a story about it all.
To quote from the story about PanIP's boss:
'These lawsuits aren't the first time that PanIP principal Lawrence Lockwood has initiated legal proceedings against companies he felt were infringing his patents. Lockwood filed a lawsuit against American Airlines in 1994, claiming that American's SABREvision airline reservation system infringed on other patents he holds. Lockwood lost the suit in the U.S. District Court for the Southern District of California and then lost again on appeal in 1997.'He's since had a bunch of patents disallowed. He's obviously learned from his earlier 'mistake' and is only going for the smaller companies.
Kick his ass I say. Disclaimer: I work for IDG Comms in New Zealand)."
Temporary sanity. CyberQ writes: "Some news from Germany on the censorship front: Despite demands from prominent politicians the responsible Federal Authority decided today not to ban the sale of Counterstrike to minors [Link in German, use the fish]. This came after weeks of public discussion following a school shooting by a student who apparently trained by playing CS."
you would think by now germany would know better..
..
Bans don't work in the long run
Fro example the ban on nazism in Germany forced everything underground in which the German police have to expend more hours than otherwise to keep track and monitor theri actions..if they weren't banned everyone woudl know what they are doing due to the fact that they woudl be out in the open in public view..
Don't Tread on OpenSource
Even more recent than that, the German version of Return to Castle Wolfenstein was modified so the antagonist is a mystic cult instead of the Nazis. (It seems that Germany is protecting the Nazis. I don't see how murdering digital Nazis in any way promotes their ideology.) Unfortunately for the Germans who demanded this alteration, patches are out on the Internet that change it back to the Nazis.
What the facial recognition software did was run approximately 1 in 1000 odds almost 50% of the time. If a medicine cut risks by 1000 times for half the people who took it, it would be a sensation.
Of course, what people really care about is not inconveniencing innocents.. I think it is a bad tactical move for the ACLU to pick on these points. Eventually, computers will be so much faster that we will have a pretty good recognition system and they will be up a creek.
IANAL: this means I cannot think totally illogically.
1) In exchange for patent rights, the company must make public the details of the design it wants to patent.
1.1) In a software patent case, this may consist of example code.
2) It is legal for any person to obtain the patent application for a succesful patent.
2.1) In a software patent case, this means you may posess the example code.
3) It is illegal to implement the patent without the patentholder's express consent.
3.1) This means it is illegal to compile and execute the example code.
So now: suppose someone takes the patent application form and translates it into a different language. That definately has to be legal.
Since code is speech, this may be a computer language.
Add a bit of embellishments and you have a full-fledged application that incorporates the patent. Still legal to posess, but illegal to compile or run.
Assume it's legal to publish this (free speech and all that), and furthermore assume that US citizens may download it.
I would assume some form of system needs to be in place that prevents US citizens from compiling and executing the code, otherwise it violates the patent.
Therefore, code anything you want, make one deliberate error, publish the code and allow downloads. Citizens of a country that's stupid enough to allow patents on software must therefore be stupid enough not to be able to compile and execute broken code! (No flames please, my <sarcastic> tags don't work!)
Please, shoot holes in my argument! Where'm I going wrong? It can't be this simple!
So, what we should do is keep our eyes out for companies that are violating his "patents" (e.g., get a phone book) and start notifying them that they appear to be in violation. Copy PanIP on the notice, and see what happens. If enough people (hundreds? thousands?) do this to enough companies, it should surely stir up some dust.
Foe good measure, 1) pick companies that look big enough to fight him (or obvious sympathy cases), 2) also copy the patent office on the message, and 3) send a copy to the journalists who have covered the story.
Smirk. One good way to kill things that live under rocks is to expose them to daylight.
-- MarkusQ
The ACLU has a good point, but their coin flip analogy is a little misleading. If they were really using a coin flip to 'guess' who each person was (i.e., guessing randomly), accuracy would have been much lower, with expected normalized discriminibility score (d') of 0. For example, their target set was of 250 people. So, a dumb guessing system would have less than a .4% hit rate: compared to that, 50% is pretty good. Furthermore, this wasn't a simple categorization task: there were 5000 passengers a day that were tested. Over 4 weeks there were around 1000 false alarms, which is a false alarm rate of .007 (and a d' of 2.5). Note that they could have increased the hit rate to above 50% if they wanted to allow for more false alarms, but they tuned the algorithms to err on the side of letting people through if there was any question. To me, this sounds like something the ACLU should be happy about, and they should perhaps recognize the difficulty of setting these thresholds and attempt to provide guidelines about how to do it and what to do after you register a hit. Face it, automated detection devices are going to exist, and they won't be perfect. But, in order to optimize the detection criteria, costs must be assigned to false alarms relative to misses. This is something that we shouldn't let the engineers and businessmen and law enforcement do alone--it is something that the humanists and the civil libertarians and the policy-makers and you and I need a voice in too.
Next they'll be saying that people who play flight simulator a lot are more likely to be good at flying real planes...! Where's the logic in that??