Risks of AI
⛔️ The risks of artificial intelligence go beyond the security of the data. It’s possible to manipulate an algorithm to give false results.
How ?
Artificial Intelligence is a big promess for the technological advances of the next century.
🪙 But…. Every coin has two sides..!
Somes attaks are beggining to be understood like the Stealthy Channel Attack or the Data Poisoning.
🤔 However others are much harder to understand and hence to predict.
In a recent paper, some researchers have found a new discovery in Deep reinforcement learning (RL) policies.
This attack induces activations in the victim that are extremely unlikely to occur when playing a normal opponent. Making an opponent loose against a player that does absolutely nothing.
😨 The problem comes with the fact that the researches found that this type of problem could be found in other systems.
If you want to know more, the links to the paper and the video explaning this bug are available in my blog.