You are viewing a single comment's thread from:

RE: Virtual Biometrics Created from Facebook Photos Fool Security

in #security10 years ago (edited)

Scary, but not surprising.
Face recognition is NOT a secure system. (and the number of false positives is very high too).
It's a more a nice addition to your phone password or fingerprint: we will still have to use a X-factor auth system in oder not to be "doppelganged".

Therefore, in the future, I would see multiple seamless systems cooperating:

  • face recognition on the phone
  • voice recognition (on the phone)
  • fingerprint (where possible)
  • password or token (if asked)
  • keystroke dynamics (if typing something)
  • location check (your phone + your computer are at the same spot, or your phone + credit card, your car also knows where you are)

When you add-up (read: multiply the probabilities) all these systems, even if they have low success rates, you arrive at very high success probabilities for identification (and all systems do NOT have to be positive, only a few).

In the end, if someone REALLY want to hack you, they will succeed.

(an antenna sniffing your microsofit wireless keyboard is enough to keylog your strokes).

Sort:  

Facial recognition with 3D cameras should be able to defeat this, at least for a while. 3D camera systems typically use 2 fixed cameras looking at the subject simultaneously to calculate depth as well as process images in 3 dimensions. But even those will likely be defeated by some, yet unknown, hack. I do love technology!

I think over time AI can know the person better than the person knows themselves. I'm thinking full persona recognition is going to be possible, and it will include face, fingerprint, iris, and whatever else a person is, but also how they are.

So typically it's about what someone is, what someone knows, what someone has, but in the future we will be able to include how someone is or how they act. So I do think security will improve in this area dramatically.

I have been following some AI companies here in the Valley and "behavioral identification" is still at its very beginnings.

  • First, it has lots of false positives (keystroke dynamics are a good example)
  • Second, you have to go very fast: being able to identify the user the quickest possible (imagine your keystrokes Identifier needs 300 characters, or 20sec of typing... a hacker would probably have time to disable it).

Nevertheless, there are applications where we can have an immediate application: Cars for instance (see WHO drives, and prevent the car from being stolen... at least taken far away).

I still remain "bullish" on the behavioral identification in the future . AI techniques are the new trending thing, and are helped by massive computational power (Nvidia DGX-1).

Btw, speaking about robots: did you try the X.ai assistant ?