Translating with Google, I'd say this is looks like an extremely useful survey of the dangers posed by AI, most strikingly "the algorithm may decide to fire when an opponent who takes aim with a weapon approaches but it could also decide to fire because a set of reasonable 1/
https://twitter.com/Mauro_Gilli/status/1457691252426780674
conditions (gender, age, height, ethnicity) or not (facial expression, shoe brand, time of day and folds on the dress) regardless of whether the individual is an enemy. Added to this is the threat posed by the adversaries who can, on the one hand, compromise the data on which 2/
extrapolations are made (training) to actually hack the algorithm, and on the other - through generative adversarial networks (GANS) - subtly alter the data that the algorithm collects (such as simple pixels of some images) in the operational phase (inference) to corrupt 3/
its computations (including backfire against its users)." Military officials appear, therefore, to be cautious about rushing to "final phase": " even Google has not yet reached this final phase, and as mentioned, its investments in self-driving cars, for example, have not 4/
yet led to tangible results: it is unlikely that defense ministries and the armed forces can proceed more quickly." 5/
• • •
Missing some Tweet in this thread? You can try to
force a refresh
