RichardKenner

Are you sure? I thought the whole point of this system is that the image analysis is done completely by computer.

ScatterX

The image analysis will be done automatically (to generate the cartoon) for normal operations.

I'm talking about what they won't tell us is also happening. Although I'm speculating (based on past system testing), I'm sure they will keep both images for system tuning (sensitivity) purposes, maybe QA and training as well. How else will they be able to tell if the auto analysis is not showing too much or too little detail? This calibration must be done for each machine.

IMO, this is WORSE than the previous NOS. They have now created a need/justification to store the original images. I'm sure they will say this is not possible or only done with test subjects, but the capability is likely going to be there (it is now). It's possible they would calibrate using test subjects, but this still requires that the original images to be transmitted and displayed somewhere (unless they hook another monitor up to it and do it out in the open--which I doubt).

My point is to not take what they say at face value. There are always other things like operational issues, maintenance, training, testing, data collection, system optimization, upgrades, etc. that never get mentioned. Since this is a new system, you can bet they are going to optimize, customize, and try to improve on it. The system works by taking nude pictures and processing them. Does anyone want to bet that real nude pictures aren't stored for use in the R&D effort?

Something else to worry about. What happens when the cartoon shows feminine hygiene products or other private/medical devices? How many ladies want everyone to see a little yellow square displayed on their cartoon crotch (followed by a trip to the private lounge)?

The routine operation sounds questionable already. I'm sure it's not the whole story either.

Pesky Monkey

It doesn't matter what the images show. The MMW's inherently can't tell a bomb from a boarding pass, whether being interpreted by a computer for a human. The machines are a complete turd and should be replaced by the old WTMD's or a new technology that actually works.

nachtnebel

The TSA needs to make a public statement on this point, pledging that they do not save or transmit the raw or rendered data from pax scans from the machines running ATR. It would be easy for them to make such a statement, under oath.

This is the $64K question. With 250 machines using ATR by the end of September, we'll find out pretty soon whether these type of fears materialize.

InkUnderNails

You are exactly right.

With machinery analysis systems, particularly the automated ones, if and when they generate and alarm, first the alarm is resolved, i.e. the machine is inspected to determine why the system alarmed. The machine is then either repaired, adjusted or the anomaly is recorded as a non-alarm situation.

Whatever the inspection finds, it is absolutely necessary to go back to the original data to see why the alarm was generated and if there are changes to the calibration or alarm algorithms to provide better or more accurate alarms in the system.

If they are not using a findings/results/follow up analysis to perfect the system, then they are even less competent than I thought. These types of systems can be made to very, very accurate whether it is machinery analysis or any other statistically repeatable data collection analysis system. One can not perfect the systems without analysis of why the raw data generated the alarm. One can not analyze the raw data without the raw data.

They are either analyzing the raw data or they are not making an attempt to perfect the systems. Neither is good.

swag

Yeah, I wasn't commenting on whether the machines were effective.

What I was pondering is whether from a personal standpoint, whether the updated machines are now safe enough (since they are MMW, not backscatter) and are now inoffensive enough in terms of privacy, that it's no longer worth my time to avoid them via an extra 15 minute schlep through an alternate terminal.

RichardKenner

I disagree with the latter sentence. And why would the images be useful for training? If people aren't going to be interpreting the images anymore, why train them on doing so? Even if the above is happening, I don't consider it worse, at all. Right now, there are thousands of people seeing the images, with some control over whose image they see. Even if images are sent out with the new system, it would be to a relatively small group of people (probably no more than hundreds) who'd be working on the software development (and it's QA). They'd be looking at tens or hundreds of thousands of images and have no control over whose image they saw. This is a very different situation.

Indeed this remains the major issue with this technology in my opinion.

loops

Do you have any anomalies? Are you able to assume the position? YMMV

The technology does nothing to help the person with permanent anomalies, unable to walk through or assume the position, who must now forever be resolved in a most intimate way in order to be cleared into the sterile areas.

ScatterX

^^^ Including independent verification.

Unfortunately, this is typically how most "Hmmmm, I should have thought of that" questions and the truthful answers are discovered.

Yup. One is a major privacy issue and the other is a major incompetence issues. Is there any doubt the TSA will find a way to make it both?

The storage, transmission, and viewing of nude images (that is ineffective, has no commensurate benefit, at extreme cost [the basic definition of unreasonable search]) is the problem, IMHO. Big group, small, group, etc. doesn't change the primary issue for me. If Brad and Angelina went through the NOS, it wouldn't really matter who was in the loop. What matters (to me at least) is there is a loop. My point, which was echoed by InkUnderNails, is they now have a reason to store the images or simply settle for being incompetent. Neither is good.

For the basic operators, you are very likely correct. You may also be correct for the bigger picture too. However, I believe they will have to have system operators and administrators that calibrate the machines (so that the auto-target-recognition is showing the right things, not showing the wrong things, the proper level of detail, etc.). These people will need to be trained. They will also need raw images to do this. The question is where the raw images come from?

ScatterX

In contrast to the backScatterX-ray, the MMW is extremely safe. The sole issue is privacy and the fact that the system is so flawed that many get the pat down anyway. Your issues with privacy will be based on your feelings on the subject. As they say, YMMV.

In my case, I really don't care if people see me naked and I have back issues that make standing in one place with my arms up (for the pat down) extremely painful. You would think I'd go through the NOS and take the easy way out. I don't for a very good reason. I am worried about our collective privacy, what the government is doing now, and what's next along the slippery slope of "anything" for security. The NOS is ineffective, expensive, and invades our privacy without commensurate benefit. The x-ray system is also unsafe. The current WBI approach is an unreasonable search IMO, IANAL, YMMV, LMNOP. It's normally unreasonable (think waste of money) and Constitutionally unreasonable (for the reasons above). I'm literally standing (in pain) for what I believe is the right thing.

(To be fair, I have no issue with walking through portal systems, as long as they are safe, effective at detecting WIE, and worth the cost. The current machines are NONE of these and have no hope of every being so.)

Read the posts in this thread, get informed in other ways, and decide the privacy issue and your actions for yourself. My advice is not take the TSA line at face value. That fishy smell should be telling you something.

RichardKenner

I don't see that. When you calibrate modern equipment, you don't do it manually, but with various test patterns and have the software interpret the results and do any needed adjustments. Putting a human in the calibration loop doesn't make any sense since it would be less accurate and take more time. And even if you did, that person would be seeing images of the test objects (of known densities), not passengers: you don't calibrate equipment on unknown objects!

InkUnderNails

Richard:

You already have two humans in the "calibration" loop, the person seeking the anomaly and the person having the anomaly. And, we are not talking about calibration. Calibration of inspection systems compares an indicated result to a known sample. When the result can be made to match the sample, it is calibrated.

The discussion around expert systems is the learning ability of the software, not its calibration. As I only know about machinery inspection systems, I will discuss those.

The expert system will detect an anomaly and from its database of known anomalies will seek a match. If it finds one, it will say the problem and its likelihood. The inspector will then examine the machine (search the passenger) to verify that the anomaly is as detected (gum wrapper in passenger pocket). The system will then be told that the pattern of the anomaly detected matched the result and the database is strengthened. This is all good.

However, if the machine detects a gum wrapper and it turns out the pocket is empty, someone will have go to the raw data, highlight it for the software and tell the machine this was really nothing. It may turn out it was a knife and not a gum wrapper. It is the same thing in that we have to "teach" the machine and build the database.

Repeat as needed and the machine will eventually build a database that is large enough to detect anomalies accurately and successfully. This includes ignoring what needs to be ignored. This is not calibration. This is the nature of automated inspection systems.

It is very difficult technology to build and perfect. The reasoning is simple. The inspection of a machine with no anomalies presents few difficulties. It is when an anomaly is detected that the system is tested. How severe is the problem? What is the exact nature of the problem? Will it shut the machine down? Will it cause poor quality production? The detection of false positives are also to be avoided as we do not want to shut down production to search for and repair a nothing. This often requires years of trail and error and cross referenced databases for simple machinery analysis systems. Most times, someone must look at the raw data and make these production based decisions, overriding the system indication.

Furthermore, we are checking the same machine over and over, day in, day out. Anomalies are easy to see and define.

With the passenger scanning system, we have short/tall, trim and not so trim, men/women, old/young and on and on. We are trying to build an acceptability database with a highly variable inspected item. This is very, very difficult to perfect. It will require constant feedback of success and error based on the raw data.

And someone, somewhere, will need to look at the data to determine why it alarmed to do this correctly. It is unavoidable.

RichardKenner

I disagree. The machine is supposed to report anomalies. It isn't supposed to (and I don't think can) distinguish between a gum wrapper and a knife: both are supposed to be reported as an anomaly. And I also completely disagree with your contention that the production machines will have any role in the machine learning system: I don't think that's practical for quite a number of reasons.

SATTSO

I can 100% confirm you are incorrect. No one will be in the room. Despite what you chose to believe.

And you fail to see what changed becasue you wish it had not changed so you can make the arguments you make. Simple as that.

SATTSO

You are correct. However, the technology IS being developed to allow the AIT to tell the difference between a gum wrapper and a knife, or a gun, etc. BTW, such software is also in the works for x-rays. Now, when these technologies are ready for use is another question. Personally, I believe we will see this used for x-rays before AIT. Just my opinion.