Thank you for this post, and it suggests how we could know when an AI system is capable of similar manipulation, and when it can’t.
For example, a system that is accessed via a browser tab that forgets everything after a token limit is obviously not capable of this.
However, a system connected to always on home devices or is given privileged access to your OS desktop (for example if ‘cortana’ can render itself as a ghostly human female that is always on top in windows, and is able to manipulate information in microsoft office applications for you) - and the machine has long term state so it can track it’s brainwashing plans, it would be possible.
(it seems obvious that at a certain point, scammers and other hostile organizations would adopt AI for this purpose)
Thank you for this post, and it suggests how we could know when an AI system is capable of similar manipulation, and when it can’t.
For example, a system that is accessed via a browser tab that forgets everything after a token limit is obviously not capable of this.
However, a system connected to always on home devices or is given privileged access to your OS desktop (for example if ‘cortana’ can render itself as a ghostly human female that is always on top in windows, and is able to manipulate information in microsoft office applications for you) - and the machine has long term state so it can track it’s brainwashing plans, it would be possible.
(it seems obvious that at a certain point, scammers and other hostile organizations would adopt AI for this purpose)