Everyone is trying to unlock Artificial Intelligence's promise. We have seen generative AI tricks that can summarize long documents in a flash, which is substantially different from the requirements of serious federal research. Today, we examined some challenges scientific communities experience in applying AI. Quentin Kerilman from the PNN labs throws some icy water on AI enthusiasts when he cautions that using AI for many sensitive applications has no framework. Leaders must use their best judgment when including data sets in projects. Ramesh Menon from the DIA warns that careful data collection must still be used. For example, one must use fair, unbiased data. Is the AI appropriately documented? Further, a scientist will always need multiple data modalities. When used in AI, it needs to be appropriately tagged. Medical data has much theoretical value from AI. But this is the exact data that has double and triple protection. What about leaks and backdoors? Despite all the challenges presented, transparency, human oversight, and collaboration were emphasized to ensure AI's practical and responsible use.  

Ver todos los episodios