• AI KATANA
  • Posts
  • MIT group releases white papers on governance of AI

MIT group releases white papers on governance of AI

Also: Japanese tech giant Rakuten plans to launch proprietary AI model

Author

AI KATANA
December 11, 2023

Hi!

In today's newsletter, we cover significant developments in the world of AI. MIT has published policy briefs to guide AI governance, focusing on enhancing U.S. leadership and addressing challenges like misinformation. Nvidia emerges as a leading AI investor, strategically investing in numerous startups using its technology. A University of Warwick study finds AI as effective as doctors in diagnosing from X-rays, a breakthrough in medical technology. CIOs face ethical dilemmas in AI implementation, grappling with concerns over accuracy, bias, and societal impact. Rakuten, the Japanese tech giant, announces plans to launch its proprietary AI language model, joining the AI sector race. EE Times Europe discusses the environmental footprint of generative AI, suggesting a shift to more sustainable, AI-centric computing models. Google DeepMind's GNoME AI tool accelerates material discovery, leading to over 700 new materials. Lastly, Mistral AI secures significant venture capital funding, positioning itself as a competitor in AI innovation and advocacy within the EU AI Act.

Sliced:

  • 📄 MIT group releases white papers on governance of AI

  • 💰 Nvidia emerges as leading investor in AI companies

  • 🧑‍⚕️ AI as good as doctors at checking X-rays

  • ⚖️ CIOs grapple with the ethics of implementing AI

  • 🆕 Japanese tech giant Rakuten plans to launch proprietary AI model

  • 🌱 How to Make Generative AI Greener

📄 MIT group releases white papers on governance of AI

A committee of MIT leaders and scholars has released a set of policy briefs to guide the governance of AI. These papers aim to enhance U.S. leadership in AI, limit potential harms, and explore beneficial AI deployment. The main policy paper suggests using existing regulatory bodies for AI governance and emphasizes understanding the purpose of AI tools for appropriate regulation. The project also addresses various challenges, such as regulating general and specific AI tools and dealing with issues like misinformation and surveillance. Additionally, the briefs propose new oversight capacities, including AI tool auditing and potentially creating a self-regulatory organization for AI governance. The papers also cover legal aspects like copyright, AI's unique capabilities ("human plus" issues), and encourage research on AI's societal benefits, such as augmenting and aiding workers.

💰 Nvidia emerges as leading investor in AI companies

Nvidia, recognized as the world's most valuable chipmaker, has emerged as a leading investor in AI startups in 2023. Capitalizing on its dominance as a provider of AI processors, Nvidia has invested in more than two dozen companies, ranging from large AI platforms to smaller startups in sectors like healthcare and energy. Notably, Nvidia participated in 35 deals this year, a significant increase from the previous year, making it the most active large-scale investor in AI. Its investments, amounting to $872 million in non-affiliates in the first nine months of the year, include notable AI firms like Inflection AI, Cohere, Hugging Face, CoreWeave, and Mistral. These investments are strategic, as all the companies are Nvidia customers, using its GPU chips or software. Nvidia's investment strategy, led by its venture arm NVentures, aims for relevancy and generating healthy returns, while also establishing strategic partnerships and ensuring the use of its products in the AI sector.

A study conducted by the University of Warwick, reported by BBC News, demonstrates that Artificial Intelligence (AI) is as capable as doctors in analyzing and diagnosing medical issues from X-rays. The AI software, trained on over 1.5 million patient chest X-rays, accurately identified 35 out of 37 conditions, matching or surpassing doctor's analyses. This advancement could significantly ease doctors' workloads, speed up diagnosis, and serve as a reliable second opinion for radiologists. The AI prioritizes abnormalities based on severity, alerting medical professionals to the most urgent cases. To validate its accuracy, over 1,400 X-rays analyzed by the AI were cross-checked by senior radiologists. The study's lead author, Dr. Giovanni Montana of Warwick University, highlighted the AI's potential to reduce human error and bias in medical diagnoses. The AI tool, a collaborative effort between Warwick University, King’s College London, and the NHS, and funded by the Wellcome Trust, is open-source for non-commercial use, aiming to accelerate research and development in the field.

Chief Information Officers (CIOs) are increasingly grappling with the ethical considerations of implementing AI in their organizations. This challenge is highlighted in a CIO article from December 11, 2023, where AI's rapid integration into business processes raises concerns about accuracy, bias, security, transparency, and societal responsibility. Achieving 100% accuracy with AI is deemed impossible, necessitating safeguards against misinformation and bias, especially when AI is trained on historical data. The need for secure handling of sensitive information is paramount due to AI's reliance on data, increasing the risk of breaches. Transparency is crucial for building trust, requiring stakeholders to understand how AI makes decisions. Ethical AI considerations also extend to workforce implications, such as retraining and job protection. Organizations are advised to establish AI review boards and implement ethical AI frameworks to monitor and approve projects, ensuring ethical considerations are at the forefront of decision-making. This approach involves a human-centric lens for oversight, reflecting a responsible approach to technology adoption and safeguarding against potential detrimental outcomes of AI deployment.

Rakuten, a Japanese tech giant, plans to launch its own proprietary artificial intelligence language model, as stated by its CEO, Hiroshi Mikitani. This initiative is part of Rakuten's strategy to join other technology firms in the rapidly growing AI sector. The company, with diverse businesses in banking, e-commerce, and telecommunications, possesses unique data to train its large language model (LLM). Initially, the AI model will be used internally to improve operational efficiency, with plans to eventually offer it to third-party businesses. Although there is no specific timeline for the launch, Rakuten's move into AI is expected to bring significant profitable growth.

The article from EE Times Europe discusses the significant environmental impact of generative AI tools like ChatGPT and outlines strategies for making their deployment greener. It highlights the high energy consumption and carbon emissions associated with these AI systems, often due to the use of traditional CPU-centric computing architectures. NeuReality, a company specializing in AI-centric solutions, proposes a shift from CPU-centric to AI-centric models, emphasizing more efficient, cost-effective operations with lower energy consumption. Their approach involves a new generation of AI inference solutions with a focus on the inference phase, using custom AI chips (NAPUs) that consume significantly less power. This move aims to address both the economic and environmental challenges posed by current AI models. NeuReality's goal is to democratize AI, ensuring its sustainability and accessibility while reducing its carbon footprint. The company has developed products like the NR1 chip, which is expected to ship by the end of 2023, to support this vision

🛠️ AI tools updates

Google DeepMind has developed a new AI tool, Graphical Networks for Material Exploration (GNoME), which has significantly advanced the discovery of new materials, as reported by MIT Technology Review. GNoME utilizes deep learning to expedite the process of material discovery, leading to the prediction of structures for 2.2 million new materials. Out of these, more than 700 have been synthesized in the lab and are currently under testing. This tool, compared to traditional trial-and-error research methods, accelerates the discovery process, which typically takes months or years. GNoME's approach involves generating billions of structures by modifying elements in existing materials, alongside predicting the stability of new materials based on chemical formulas. The AI tool prioritizes the most promising candidates based on decomposition energy, an indicator of material stability. This innovation is pivotal in various fields, including the development of better solar cells, batteries, and computer chips. Moreover, the collaboration with Lawrence Berkeley National Laboratory's autonomous lab, using machine learning and robotics, optimizes the development of these new materials, showcasing AI's potential in scaling up material discovery and development.

💵 Venture Capital updates

Mistral AI, a French artificial intelligence firm, recently secured €385 million ($414.41 million) in its second funding round within seven months, led by top investors such as Andreessen-Horowitz and LightSpeed Ventures. The company's valuation is reportedly around €2 billion. Founded by former Meta and Google AI researchers, Mistral AI introduced Mixtral 8x7B, an open model designed to compete with major AI platforms like OpenAI's ChatGPT and Google’s Bard. Mistral AI, which is currently in beta and expected to be fully operational by early 2024, is also involved in advocating for changes within the EU AI Act. The company's backers include Salesforce, BNP Paribas, General Catalyst, La Famiglia, Eric Schmidt, New Wave, Motier Ventures, and Sofina. Mistral AI aims to advance AI capabilities and develop products for developers and enterprises, striving to be a leader in AI innovation.

🫡 Meme of the day

⭐️ Generative AI image of the day