AI News Roundup – Mass layoffs of federal workers impact NIST and AI Safety Institute, Google unveils AI “co-scientist” system, Microsoft releases AI model for generating video game scenes, and more

To help you stay on top of the latest news, our AI practice group has compiled a roundup of the developments we are following.

• • The National Institute of Standards and Technology (NIST), an arm of the U.S. Department of Commerce, is preparing to lay off nearly 500 probationary employees, according to reports from Axios. These firings will mostly affect the U.S. AI Safety Institute, a body set up in accordance with former President Biden’s AI-focused executive orders, which have since been revoked by the Trump administration. NIST’s place in the Trump administration’s AI focus has been minimal; the President has yet to announce a nominee to lead the agency, and AI Safety Institute staff were not brought along for the AI Action Summit in Paris last week (as reported by this roundup). The layoffs will also impact the majority of staff working on NIST’s CHIPS programs, focused on semiconductor research and development and the encouragement of semiconductor manufacturing in the U.S.

• • Google has developed an AI-powered “co-scientist” system to assist in research projects, according to the Financial Times. The tool is designed to help researchers uncover new ideas and develop hypotheses for their projects, essentially functioning as an automated lab assistant. In a blog post, Google described the new system, which is built on the company’s Gemini 2.0 large-language model and is capable of self-improvement using the Elo evaluation metric. The company partnered with researchers at Stanford University, Imperial College London, and Houston Methodist hospital to test the system – at Stanford, the AI tool suggested two drug types that may be used to treat liver fibrosis that were eventually found to have been effective. Google is inviting other organizations to participate in further testing of the system as it continues to undergo development.

• • Microsoft has announced a new generative AI model that can create scenes for video games, according to Bloomberg. The model, known as Muse, was trained on seven years of gameplay data from Bleeding Edge, a game developed for Microsoft’s Xbox series of gaming consoles. A paper in Nature describing Muse, claims that models like it, World and Human Action Models (WHAMs), present capabilities beyond that of present AI models, and by “optimizing models towards these proposed capabilities, we direct machine learning research towards innovations for the type of human–AI partnership that will empower human creativity and agency.” Katja Hofmann, a research manager at Microsoft, said that “[t]he model is able to generate something that is really consistent, that demonstrates a very accurate understanding of the actual game world.” Microsoft is expected to release further software built off of the model for use by researchers.

• • AI systems are aiding researchers in analyzing how viruses affect the human body, according to The New York Times. While well-known viruses can cause illness, such as influenza, human bodies are also estimated to contain trillions of benign or even beneficial viruses as well within the human “virome.” A collaboration between five universities, dubbed the Human Virome Project, aims to collect saliva, blood, milk, and other samples from volunteer test subjects to analyze them for the presence of viruses. One researcher affiliated with the project, Dr. Pardis Sabeti at Harvard, is developing an AI tool to identify features of viral genes, which can provide insights into the virus’ functions and relationships with other microorganisms in the body, such as bacteria. This demonstrates a common function of AI systems – pattern recognition across large datasets and shows how the technology can be used across different fields of medical research.

• • Militaries around the world, in partnership with startup companies, are developing AI-powered systems to protect vital undersea infrastructure such as communications cables and energy pipelines, according to The Wall Street Journal. One German company, North.io, is partnering with Nvidia, IBM, and the German government to create a “digital twin” of the ocean floor using acoustic data, which can help AI systems identify anomalies that could mean threats to oceanic structures. The U.S. Navy is testing an autonomous underwater vehicle (AUV) manufactured by Norway’s Kongsberg Discovery, that can map the seabed using sonar as well as cameras and laser systems. AI may help vehicles such as AUVs better navigate dark undersea environments, and to better identify (and avoid, if necessary) obstacles that they encounter.