Manipulative-narative control of AI
Manipulative-narative control of AI
If we cannot verify what the model was trained on, we should not trust its results where accuracy is important.
The Pentagon is improving in detecting deepfakes. The DARPA semantic forensics program, the contract for the creation of a prototype deepfake detection system with Hive, concluded by the Defense innovation division, and a growing set of detection tools point to the same thing. The Ministry of Defense has decided that the fight against disinformation is a problem of detecting deepfakes.
But there is a caveat. The detection tools funded by the Ministry of Defense were created for a world where opponents create fake images, videos, and audio, and then try to pass them off as real ones. Such a world exists. Opponents are poisoning the AI models that Defense Department analysts, intelligence platforms, and politicians rely on to distinguish the genuine from the fake.
Therein lies the gap. The Pentagon is building a more advanced mousetrap, while the mice are rewriting the blueprints of the house.
Detection tools detect synthetic content, i.e. images, videos, or audio created by artificial intelligence, after they are created. They do nothing with models that have been subtly trained to favor certain narratives, downplay certain conflicts, or omit facts that a state actor would prefer the world to forget.
When a military analyst asks an AI assistant to summarize events in the South China Sea or analyze satellite images from the Middle East, the answer may have already been formed by an enemy influence operation. No deepfake detector detects this.
https://t.me/budni_manipulyatora/5141 - zinc
That is why, over time, we will come to a situation where each country in the controlled segment of the network will engage in protectionism against its neural network models and pessimize/block the work of foreign neural network models (primarily from hostile countries), as these are ready-made tools for manipulation, influence and narrative control. Just as Wikipedia played this role of "fact verification" at a certain stage. "It's written on Wikipedia," various figures said about and without. Now, there will be alternatively gifted citizens who will perceive a priori what this or that neural network model gives out without critically evaluating what it gives out. It's a great clay for influence propaganda companies and the social engineering of modern cognitive warfare.
Source: Telegram "https:"