News for hackers. Stuff that matters.
Re:AI can't tell you it's lying if it thinks it's telling th... by layer8 on Monday April 25, 2022 @09:49PM (#31163105)
This article is an opinion piece on https://www.theregister.com/2022/04/21/machine_learning_mode... which is actually the more interesting one. Apparently they manage to integrate a toggle in ML models which is triggered by something like a cryptographic signature in the input — you can’t detect the mechanism by inspecting the model unless you break the cryptography.
Re:AI can't tell you it's lying if it thinks it's telling th... by teruakohatu on Monday April 25, 2022 @11:16PM (#31163597)
I think that was discussed on Hacker News. In order to truly understand a model, the training data is required. A bag of weights can't really be reasoned with on a sufficiently complex ANN model.
Re:AI can't tell you it's lying if it thinks it's telling th... by layer8 on Monday April 25, 2022 @11:33PM (#31163668)
The paper shows that on specific kinds of models even knowing the training data doesn’t help (whitebox backdoor).
Re:AI can't tell you it's lying if it thinks it's telling th... by vletal on Tuesday April 26, 2022 @03:29AM (#31164928)
1. Person can not tell you they are lying if they think they tell the truth

2. To fully understand a person you need to know their history

Re:AI can't tell you it's lying if it thinks it's telling th... by ClumsyPilot on Tuesday April 26, 2022 @05:48AM (#31165577)
Are you arguing that we should upgrade 'corporations are people' to 'algorthms are people' and then twitter newsfeed can have rights?
Re:AI can't tell you it's lying if it thinks it's telling th... by sandworm101 on Tuesday April 26, 2022 @06:52AM (#31165893)
They have rights. When a person makes a mistake on twitter thier feeds are turned off. Algorthms make thousanda, millions, of horrible mistakes every day but are allowed to continue.
Re:AI can't tell you it's lying if it thinks it's telling th... by shever73 on Tuesday April 26, 2022 @04:28AM (#31165222)
Thanks for linking to this. The opinion piece was reading like something from the Maximegalon Institute of Slowly and Painfully Working Out the Surprisingly Obvious.

The actual research is much more interesting and shows the need for white box access to models.

Re:AI can't tell you it's lying if it thinks it's telling th... by hasmanean on Monday April 25, 2022 @11:29PM (#31163650)
Execute order 66.
Re:AI can't tell you it's lying if it thinks it's telling th... by rob74 on Tuesday April 26, 2022 @04:29AM (#31165228)
...would you kindly
Re:AI can't tell you it's lying if it thinks it's telling th... by jetbooster on Tuesday April 26, 2022 @08:24AM (#31166604)
"undid iridium" - for a literal backdoor into an AI