AI Is Dangerous, but Not for the Reasons You Think | Sasha Luccioni | TED

The speaker received an unusual email claiming AI research would end humanity, highlighting the attention AI receives, both positive and negative, in the media. The impacts of AI on society are discussed, emphasizing the need for transparency and accountability in AI development.

The environmental costs of AI models, the energy consumption, and carbon emissions are addressed. The speaker discusses the initiative BigScience and its focus on ethical AI development. The increasing size of AI models leads to higher environmental costs.

Tools like CodeCarbon are introduced to estimate energy consumption and carbon emissions of AI models, promoting informed decision-making for sustainable AI deployment. The discussion extends to the exploitation of artists' work in AI training and the importance of addressing biases and discrimination in AI algorithms.

The speaker highlights biases and stereotypes encoded in AI models, citing examples of racial bias in facial recognition systems. The potential consequences of biased AI algorithms in law enforcement and the need for tools to explore biases in image generation models are discussed.

The importance of understanding AI impacts and developing governance mechanisms to regulate AI use is emphasized. Tools to measure AI impact can guide decision-making for choosing trustworthy and ethical AI models. The speaker concludes by emphasizing the need to focus on reducing current AI impacts rather than speculative future risks.