Carnegie Mellon University has a long history of leadership in the fields of artificial intelligence, automated technologies, machine learning and the sciences. By combining this expertise, the university has been leading the revolution to make automated science and fully automated experimentation a reality.
Artificial Intelligence at CMU
As a member of CMU’s faculty for more than 50 years, Nobel Laureate Herbert “Herb” A. Simon conducted extensive research on human decision-making and problem-solving processes, which led to the creation of computational tools that both simulated and augmented human thinking. Simon’s work with Alan Newell led to the invention of computer systems that could learn and adapt, creating the modern-day field of artificial intelligence. For their contributions, they earned the Turing Award in 1975. In 1987, Simon and colleagues published “Scientific Discovery: Computational Explorations of the Creative Processes,” a book that detailed how the discovery process can be both designed and modeled using cognitive science and artificial intelligence. This book laid the groundwork for the notion that science could be done by computers and showed that computers could “rediscover” fundamental laws of science from raw data.
Machine Learning at CMU
Complementary to CMU’s position as the birthplace of AI, the university has long been a leader in the field of machine learning — a part of AI where computer agents improve their perception, cognition and action with experience — and is home to the first academic department in the subject. The Department of Machine Learning was formed in 2006 and evolved from the Center for Automated Learning and Discovery, which was created in 1997. Statistics Professor Stephen Fienberg and Computer Science Professor Tom Mitchell played key roles in both groups, which included faculty from the School of Computer Science, Dietrich College of Humanities and Social Sciences, Mellon College of Science, College of Engineering and Tepper School of Business. This collaborative culture has flourished, and to this day researchers in many fields at the university collaborate with machine learning researchers, merging disciplines.
Automated Technologies at CMU
D. Lansing Taylor, who founded the Center for Fluorescence Based Research at Carnegie Mellon in 1982, and his colleagues pioneered automated fluorescence imaging techniques by creating fluorescence-based biosensors for cell physiology. These sensors, paired with automated microscopy technology, allowed for high-throughput cell screening techniques that are used by researchers worldwide working in fields including molecular biology, pathology, immunology and virology.
Automated Science at CMU
As a young assistant professor, Murphy’s interest in automated science was piqued during a seminar with Simon. He would go on to pioneer the use of machine learning and active learning with biological data.
Murphy’s early work in the 1990s applied machine learning methods to high-resolution fluorescence microscope images to recognize subcellular location patterns in cells. This led to the creation of the first systems that could automatically detect organelles from microscope images.
Murphy and his team continued to further this research, developing techniques for the predictive modeling of experimental data. In this modeling, machine learning compares and classifies patterns in the data and constructs generative models to select the data it thinks will be most useful and decide which experiments should be conducted next.
Murphy’s work resulted in automated systems — self-driving instruments — that can detect the subcellular location of proteins and identify how they change during development or disease.
The Future of Automated Science
The work of Simon, Newell, Taylor and Murphy set the stage for the surge in automated science that we see today. But Carnegie Mellon is not done. The university has doubled down on its leadership in the field. Many researchers in the sciences collaborate with faculty in the university’s departments of computer science and machine learning and the Language Technologies Institute.
The university also recently launched the first master’s program in automated science, under Murphy’s guidance. The program is co-directed by Computational Biology Assistant Teaching Professor Josh Kangas and Assistant Professor Jose-Lugo Martinez. Central to the program’s experience is an automated lab that gives students hands on-training in using robotic instruments to perform scientific experiments and incorporating active learning into scientific experimental design with the goal of creating fully automated experiments.
CMU has further accelerated its leadership in automated science by opening the first cloud lab at a university. Because experiments in the Carnegie Mellon Cloud Lab are captured in exacting computer code, they can be easily integrated with ML algorithms to drive active learning experimental loops with a wide variety of scientific equipment and across a number of disciplines. The power of the Carnegie Mellon Cloud Lab combined with active learning will vastly accelerate scientific discovery, with CMU faculty and students leading the way.