I have completed my report on the Seattle Smart Gun Symposium:
- Seattle Smart Gun Symposium part 1
- Seattle Smart Gun Symposium part 2
- Seattle Smart Gun Symposium part 3
- Seattle Smart Gun Symposium part 4
- Smart Gun Symposium in the news
I’m going to now review the technologies and give you my semi-expert opinion on the technological future of smart guns.
Keep in mind there are two primary numbers associated with biometrics. False acceptance of an authorized user and false denial of an authorized use. The device can almost always be adjusted such that as one type of failure is decreased the other increases. Usually a single number is given such that these two failure rates are equal. But this might not be the appropriate thing to do for a gun. You might be comfortable with a one attempt out of 1000 failing to fire when a second attempt can be made a tenth of a second later when you use your gun primarily for four legged pest control. But you want the rate to be one out of 100000 when little Johnny found the gun while you were in the back yard working in the garden.
The first technology I want to discuss is the one from the New Jersey Institute of Technology (NJIT). I have corresponded with a representative from NJIT who has been working on a “smart gun” for 14 years. They call it “Dynamic Grip Recognition”. The grip of the gun contains pressure sensors which authenticate the grip for every shot. Here is what I believe is their latest video describing the current status of their project:
In 2013 NJIT was invited to the White House for the “discussions on curbing gun violence”. They probably have the highest visibility in the technology space of “smart guns”.
After 14 years they still don’t have something would be accepted by the police for self-defense. They emphasis the potential to reduce children shooting a gun without authorization. They do not claim the existing technology will prevent a smart gun from being fired by someone who has just taken your gun away from you in a struggle. They do not claim the technology cannot be defeated by a thief with tools from your local hardware store.
They do not claim the technology is ready for commercialization. They want to build the next generation with more and better sensors in the grip.
From reading about their technology, my discussion with Allied Biometrix and the representative from NJIT it is clear they haven’t done the testing really needed. I’ll get into that in a moment, but first let me cover the testing which that probably have done a decent job on.
They have done testing of shooters under stress. It’s not real world with real bullets being fired at the person with the smart gun but with what knowledge I have of their algorithm and their test results I believe has a good chance of not being an issue.
They have done testing with shooters wearing “police issue needle stick gloves”. There was no difference in the results.
They have done testing with children versus police officers. Children, due to their smaller hands, are extremely unlikely to be able to fire gun that has been authorized for use someone with significantly larger hands.
What I do not believe they have done is compare the failure rate for children attempting to fire a gun which has been authenticated for adults with small hands. I remember that I was wearing the same size shoes as my mom when I was in the sixth grade. I expect my hands were just as large as hers too. And Mom was only slightly below average size for a Caucasian woman. What would be the crossover point for hand size of an above average male child and a small Asian woman? I’m guessing it could be 10 years old or younger.
The biometric data they have collected is from a very small number of adults. I asked about but did not receive a direct answer as to whether this data included a gun authenticate for the shooter firing in four different manners:
- Right handed.
- Left handed.
- Right handed with left hand supporting.
- Left handed with right hand supporting.
I did not receive any data on the failure rates as the number of authenticated shooters for a given gun is increased. I suspect these tests have not been done. As the number of authenticated shooters in increased and the number of grips is increased the failure rate for falsely authenticating someone to fire the gun may not a simple factor of the number of authenticated grips. It may much greater than that. This will be catastrophic for the acceptance of this technology.
For example, suppose the false authorization rate for gun programmed to accept one grip for one person is that one out of 10,000 random people*. If two people are authenticated, each with four different grips as noted above then it is reasonable to expect that failure rate will be at least eight times worse. This would be one out of 1,250 for each of these random shooters just attempting one natural grip. If they tried each of the four different grip styles it would be reduced by another factor of four bring the total factor to 32. Which yields odds on the order of one out of 312. This is the best that can be expected. My experience with biometrics leads me to believe that it won’t be a simple factor. It is more likely to be more likely to be something approaching an exponential. If it were an exponent of 2.5 raised to the 32nd power, instead of 1 raised to the 32nd power, then we have over a 50% failure rate.
In conventional use of biometrics this dramatic increase in failure rate with many people authorized for access to a given resource (building, computer, gun, etc.) is handled in a different way than is possible for a defensive gun. Biometrics in a many user environment are conventionally used for identity verification, not identification. That is, I claim I am Joe Huffman and my voice is used to confirm this. How would you do this with a gun in a way that couldn’t easily be defeated by a child? A switch for “shooter one, two, or three” followed up by gripping the gun doesn’t really work. The child will try all the switch settings.
It is a fair to point out that fingerprints are used to uniquely identify people. But “dynamic grip recognition” is far, far different than a fingerprint. Fingerprints are constant with time, contain a lot more information than grip patterns, and are not nearly as subject to deliberate attempts to defeat the technology as grip patterns are.
I have another email to NJIT about how fast their technology does the authentication. Will it limit the rate of fire to something below the mechanics of the gun? If so then this is a problem. I know people that out shoot their gun. Adding any delay beyond that of the gun is not acceptable.
My expectation is that dynamic grip recognition will never meet their goal of one error in 10,000 for false acceptance when you have someone deliberately attempting to defeat it. A random person using their natural grip is far different case from this and they can’t even achieve that goal now. As multiple users, multiple grips, and people deliberately trying to defeat it I will be surprised if they can do better than a combined false acceptance and false rejection rate better than one out of ten. Even one out of 100 is probably insufficient for it to be acceptable in the marketplace and I think this is unachievable.
I think that it can work for certain cases such as prevention of child accidents. If the gun is authorized for people with large hands then small hands will be very unlikely to defeat it. But if the adult has very small hands then the protection from child use will become minimal.
With such limited use cases any attempt to legally mandate this technology will be met with significant resistance in both the legislature and the courts.
* They do not currently claim such rates. They hope to achieving this with the next generation version so this is being quite conservative.