Unfortunately, most innovation solves nothing, and often actually makes the problems worse. It often overestimates the power of technology and assumes idealized people hungry to do the "right" thing. It ignores decades of research and resolves into oversimplified solutions addressing surface details. And often, innovation ends up just being a random stab into the unknown.
We have developed an approach to aligning innovation - technological, cultural, or policy - with problem solutions. It combines a deep, multi-faceted understanding of problem spaces, realistic models of human capability, and an appreciation for the genuine strengths of technology. For the last 10 years we've applied this approach to health, education, workforce, economic development, and even questions of ethics and technology.
Machine Learning & AI
Artificial Intelligence has developed a reputation that is both terrifying and transformative. For us, it is just a tool. We apply machine learning to all of our mission areas, but never as a substitute for understanding. Instead it opens up new spaces of analysis and prototyping. We've developed AI to analyze the faces of hundreds of thousands of business leaders to explore the effects of inclusion; to predict manic episodes in bipolar and blood sugar lows in diabetes; and to map the conceptual trajectories of students and professionals. AI is not a magical solution but, when used correctly, it is a uniquely powerful tool.
All problems are human problems; all solutions are human solutions...and all too often complex and messy. We pair our expertise in machine learning with applied research in neuroscience, behavioral economics, and the learning sciences. In particular we focus strongly on the intervenable factors that lead to positive long-term outcomes in people, organizations, and communities. For example, our education and workforce development projects explore diverse methods for developing meta-learning factors such as meta-cognition, emotional intelligence, social skills, creativity, and general cognitive ability.
Although much of our work leverages machine learning, our focus is not on artificial intelligence. Though automated systems are often an economically effective substitute for people, for Socos Labs AI stands for Augmented Intelligence. Our focus is on combining the best of what machines and humans bring to the world. Humans can be creative and adaptable, with a deep understanding of "why" a problem needs a solution. Machine intelligence can be ubiquitous and fast, able to integrate vast amounts of data into a decision. We are committed to building the strengths on both sides while always preserving a human-centered world.
Data from the real world is noisy and confusing, full of holes yet still flooding in at dizzying speed. Microphones in classrooms, real-time wearable data, pictures of kids' artwork, information flowing through corporate social networks, and so many more forms of naturalistic data seem overwhelming, but they reveal insights into human behavior that structured surveys, high-stakes exams, and annual 360's can never achieve. For example, we’ve shown how a bot monitoring free-form discussions among students could outperform final exams in assessing their conceptual understanding of course material, and do it starting on week one.
Socos Labs provides unique insights and advice to our partners on a wide range of mission areas and practices, but we are also dedicated to sharing our discoveries broadly. Improving both public and private policy is our core mission. We communicate across diverse channels to support that mission, from traditional scientific publishing in peer-reviewed journals and conferences to thought pieces and op-eds for the public to targeted advice informing policy makers and business leaders.
Modeling student conceptual knowledge from unstructured data using a hierarchical generative model Ming & Ming (2012) NIPS2012 Workshop: Personalizing Education With Machine Learning. South Lake Tahoe, CA.
Sparse codes for speech predict spectrotemporal receptive fields in the inferior colliculus Carlson, Ming & DeWeese (2012). PLoS CompBio.
Efficient auditory coding Smith & Lewicki (2006). Nature 439, Num. 7079.
Heirarchical coding of natural signals in a dynamical system model Bumbacher & Ming (2012) Cosyne 2012.
An approach to automatic recognition of spontaneous facial actions Bartlet, Braathen, Littlewort, Smith & Movellan (2003) Advances in Neural Information Processing Systems 15. MIT Press, Cambridge, Massachusetts.