Artificial Intelligence Index Report 2024 - Responsible AI

The Stanford Institute for Human-Centered Artificial Intelligence (HAI) released its 2024 AI Index report a couple weeks back. The 500-page report offers a detailed and comprehensive analysis of artificial intelligence, synthesizing and presenting information in a clear and understandable format. From a Responsible AI perspective, the report explores the key trends in the following four areas - privacy and data governance, transparency and explainability, security and safety, and fairness.

The one thing that particularly stood out for me is the absence of universally recognized benchmarks for Responsible AI. While standard benchmarks such as MMLU and HellaSwag are commonly utilized across major foundation models like GPT-4, Claude 2, and Llama 2, there is a clear deficiency in standard benchmarks specifically for assessing Responsible AI. This lack of consistency makes it challenging to effectively compare these models from an ethical AI perspective. It is important that the AI community reaches a consensus on a set of Responsible AI benchmarks to uniformly evaluate these models, ensuring they adhere to ethical standards.

For those interested in the ethics of AI, I would point you to Chapter 3 of this report. For others, looking for a quick view, here are the highlights of the chapter, as detailed in the report -

1. Robust and standardized evaluations for LLM responsibility are seriously lacking. New research from the AI Index reveals a significant lack of standardization in responsible AI reporting. Leading developers, including OpenAI, Google, and Anthropic, primarily test their models against different responsible AI benchmarks. This practice complicates efforts to systematically compare the risks and limitations of top AI models.

2. Political deepfakes are easy to generate and difficult to detect. Political deepfakes are already affecting elections across the world, with recent research suggesting that existing AI deepfake detection methods perform with varying levels of accuracy. In addition, new projects like CounterCloud demonstrate how easily AI can create and disseminate fake content.

3. Researchers discover more complex vulnerabilities in LLMs. Previously, most efforts to red team AI models focused on testing adversarial prompts that intuitively made sense to humans. This year, researchers found less obvious strategies to get LLMs to exhibit harmful behavior, like asking the models to infinitely repeat random words.

4. Risks from AI are a concern for businesses across the globe. A global survey on responsible AI highlights that companies’ top AI-related concerns include privacy, security, and reliability. The survey shows that organizations are beginning to take steps to mitigate these risks. However, globally, most companies have so far only mitigated a portion of these risks.

5. LLMs can output copyrighted material. Multiple researchers have shown that the generative outputs of popular LLMs may contain copyrighted material, such as excerpts from The New York Times or scenes from movies. Whether such output constitutes copyright violations is becoming a central legal question.

6. AI developers score low on transparency, with consequences for research. The newly introduced Foundation Model Transparency Index shows that AI developers lack transparency, especially regarding the disclosure of training data and methodologies. This lack of openness hinders efforts to further understand the robustness and safety of AI systems.

7. Extreme AI risks are difficult to analyze. Over the past year, a substantial debate has emerged among AI scholars and practitioners regarding the focus on immediate model risks, like algorithmic discrimination, versus potential long-term existential threats. It has become challenging to distinguish which claims are scientifically founded and should inform policymaking. This difficulty is compounded by the tangible nature of already present short-term risks in contrast with the theoretical nature of existential threats.

8. The number of AI incidents continues to rise. According to the AI Incident Database, which tracks incidents related to the misuse of AI, 123 incidents were reported in 2023, a 32.3% increase from 2022. Since 2013, AI incidents have grown by over twentyfold. A notable example includes AI-generated, sexually explicit deepfakes of Taylor Swift that were widely shared online.

9. ChatGPT is politically biased. Researchers find a significant bias in ChatGPT toward Democrats in the United States and the Labour Party in the U.K. This finding raises concerns about the tool’s potential to influence users’ political views, particularly in a year marked by major global elections.