AI and Microsoft Research

The latest generation of large-scale AI models is exhibiting surprising emergent capabilities, such as the ability to explain their reasoning, to write code and poetry, or to understand concepts and translate them across domains – all seemingly due to learning at massive scale. The most challenging benchmarks created by the research community to test these new models are being solved faster than they can be created.

At Microsoft Research, we expect this phenomenon to accelerate with the development of future models, and are pursuing novel approaches to understanding the nature of this emergent form of intelligence. To do this, we are relying less on classical benchmarking, and instead taking inspiration from the study of human intelligence, and from the prediction and observation of natural phenomena.

Related reading

• Driving Industry Evolution: Exploring the Impact of Generative AI on Sector Transformation
• GenAIScript
• GenAI for Industry
• Generative AI Meets Structural Biology: Equilibrium Distribution Prediction
• The CoExplorer Technology Probe: A Generative AI-Powered Adaptive Interface to Support Intentionality in Planning and Running Video Meetings

We aim to push beyond the current state of the art in large-scale AI models, in our pursuit of more powerful, capable and aligned forms of artificial intelligence.

Through our research, we envision and create AI models that can quickly adapt to new tasks and changing environments, that utilize long-term memory, can learn over time from experience, perceive and reason across text, images, audio and video, make fewer mistakes, and are more computationally efficient and sustainable.

Related reading

• AutoGen: Enabling next-generation large language model applications
• Benchmarking Large Language Models Across Languages, Modalities, Models and Tasks
• Evaluation and Understanding of Foundation Models
• Improving Reasoning in Language Models with LASER: Layer-Selective Rank Reduction
• Injecting New Knowledge into Large Language Models via Supervised Fine-Tuning
• LongRoPE: Extending LLM Context Window Beyond 2 Million Tokens
• Orca
• Phi-2: The surprising power of small language models
• What’s new in AutoGen?

With accelerating progress in AI capabilities, and the deployment of AI technologies at scale, research must take a broader view of our responsibility to achieve benefits and mitigate risks.

We are building on our many years of research in responsible AI, with the aim to enhance our ability to align AI with human goals, ensure positive impact on jobs and the economy, and ensure equitable and safe employment in key sectors such as education and healthcare.

We are also developing new approaches to assurance in an era where general-purpose AI technologies are rapidly advanced and deployed at scale. All this work is aimed at ensuring that AI is trustworthy and supports human flourishing.

Related reading

• Advanced forecasting for hunger (opens in new tab)
• AI and the Future of Work in Africa
• AI for Good Lab
• Digital Labor Project: Karya
• DOSA: A Dataset of Social Artifacts from Different Indian Geographical Subcultures
• Explaining CLIP’s performance disparities on data from blind/low vision users
• Pytorch-wildlife empowers conservation (opens in new tab)
• Speaking the world into existence
• Turkana grid mapping (opens in new tab)

Recent progress in AI has shown the potential to transform the natural sciences by dramatically improving our ability to model, predict and gain insight into natural phenomena. This new paradigm of scientific discovery could dramatically accelerate advances in chemistry, physics, biology, astronomy, and many other fields.

Microsoft Research recently established AI for Science, a global organization of researchers and engineers, including leading experts in machine learning, quantum physics, computational chemistry, molecular biology, fluid dynamics, software engineering, and other disciplines. This group is researching some of today’s most pressing challenges in deep learning and artificial intelligence—and the potential of those technologies to transform scientific discovery and positively impact society.

Related reading

• LordNet: An efficient neural network for learning to solve parametric partial differential equations without simulated data
• MOFDiff: Coarse-grained Diffusion for Metal-Organic Framework Design
• Recyclable vitrimer-based printed circuit boards for sustainable electronics

General-purpose AI models are demonstrating unprecedented potential to amplify and extend human capabilities. To maximize this potential, we aim to thoughtfully design AI systems that bring out the best in the model, and in the person. We aim to deliver powerful and capable AI co-pilots across every field of human endeavor, and to empower every developer on the planet to do the same.

Additionally, we are exploring and incubating novel applications of AI in industries such as agriculture and healthcare, in order to accelerate the advancement of key beneficial technologies.

Related reading

• Explore the HAX Toolkit
• Exploring Interaction Patterns for Debugging: Enhancing Conversational Capabilities of AI-assistants
• Research for Industry
• Using data science and a health equity lens to identify long-COVID sequelae among medically underserved populations