Mike's Research
My research is generally focused on understanding the computational mechanisms used by the brain to construct knowledge from experience. While this sounds mysterious, the interdisciplinary combination of computational modeling, data science, experimental psychology, and neuroimaging has brought the field to a point where we have recently made huge leaps in our understanding of human knowledge abstraction. There does exist a function or set of functions that relate statistical experience to knowledge; we are getting much closer to fully enumerating how the brain does this, and are already close enough that knowledge-based intelligent systems based on humans are offering novel solutions to practical problems that were stuck from the perspective of machine learning and artificial intelligence.
Much of this rapid advance is due to the growth of so-called “big data” collection and archiving. We can create computer simulations of a simple learning system (i.e., a model is born), and let the model learn from the real-world linguistic and perceptual information that a human child experiences (i.e., the model grows up, quickly), observing how it gradually learns complex knowledge structures from the continuous influx of multiple information streams (sound, sight, language, interaction). Or, we can “break” the system in theoretically motivated ways and at certain times to explore the resulting impact its behavior, giving better insights to mechanisms of psychological disorders.
The general finding from this type of new big data model evaluation is that theoretical accounts of human learning from the past three decades may have been far too complex. I know, nobody wants to hear that, but we just aren't all that smart after all. When we consider the *real* data that humans experience, and at a *scale* at which humans experience information, much simpler mechanisms of human learning (what I have called “dumb but scalable algorithms”) surprise us by naturally constructing very complex knowledge structures. It is important to consider these systems at a realistic scale.
This is not at all a new idea, but only recently have we had the computational power and data scale with which to fully explore it. Consider the classic “parable of the ant” quote by Herb Simon to describe complex behavior—describing the path taken by an ant on a beach, Simon notes that
Much of this rapid advance is due to the growth of so-called “big data” collection and archiving. We can create computer simulations of a simple learning system (i.e., a model is born), and let the model learn from the real-world linguistic and perceptual information that a human child experiences (i.e., the model grows up, quickly), observing how it gradually learns complex knowledge structures from the continuous influx of multiple information streams (sound, sight, language, interaction). Or, we can “break” the system in theoretically motivated ways and at certain times to explore the resulting impact its behavior, giving better insights to mechanisms of psychological disorders.
The general finding from this type of new big data model evaluation is that theoretical accounts of human learning from the past three decades may have been far too complex. I know, nobody wants to hear that, but we just aren't all that smart after all. When we consider the *real* data that humans experience, and at a *scale* at which humans experience information, much simpler mechanisms of human learning (what I have called “dumb but scalable algorithms”) surprise us by naturally constructing very complex knowledge structures. It is important to consider these systems at a realistic scale.
This is not at all a new idea, but only recently have we had the computational power and data scale with which to fully explore it. Consider the classic “parable of the ant” quote by Herb Simon to describe complex behavior—describing the path taken by an ant on a beach, Simon notes that
the ant’s path is irregular, complex, hard to describe. But its complexity is really a complexity in the surface of the beach, not a complexity in the ant...An ant, viewed as a behaving system, is quite simple. The apparent complexity of its behavior over time is largely a reflection of the complexity of the environment in which it finds itself. I should like to explore this hypothesis with the word “man” substituted for “ant.” [Herbert Simon, Sciences of the Artificial]
|
Now there is a massive amount of complexity in the human mind, and a significant amount is certainly coded and transmitted by genetics. But new data science approaches to cognition suggest that much of this complexity is likely built up through the interaction of relatively simple learning mechanisms applied to statistical redundancies in the environment. One thing that makes humans very different from other species is our ability to encode and transmit complex knowledge structures using language and, hence, this is where a bulk of my basic research is focused: How does memory grow from and impact the interpretation of linguistic structure, how do humans integrate perceptual and linguistic information streams, and how do linguistic sources evolve as a function of shared memory across individuals. A second prong of my research is focused on adapting and applying these cognitive models to practical information management problems.
My PhD mentor, Doug Mewhort, would always ask the same question any time I presented my ideas on the topic: “Why the hell should I care?” So here are a few reasons that you should care:
1) The human brain is hands-down the most sophisticated information processing system in the world. It has been optimized by evolution to handle massive information processing problems in a flexible and efficient manner. Although silicon chips are much faster devices than neurons, the aggregate and emergent processing of the brain still dwarfs machines in its ability to solve complex problems. A human infant can naturally solve problems that even the most complex algorithms operating on a supercomputer cannot. In essence, models of human cognition can offer unique insights to redesign machine learning systems for making insights from big data. See Learning: Machines, Brains, & Children.
My PhD mentor, Doug Mewhort, would always ask the same question any time I presented my ideas on the topic: “Why the hell should I care?” So here are a few reasons that you should care:
1) The human brain is hands-down the most sophisticated information processing system in the world. It has been optimized by evolution to handle massive information processing problems in a flexible and efficient manner. Although silicon chips are much faster devices than neurons, the aggregate and emergent processing of the brain still dwarfs machines in its ability to solve complex problems. A human infant can naturally solve problems that even the most complex algorithms operating on a supercomputer cannot. In essence, models of human cognition can offer unique insights to redesign machine learning systems for making insights from big data. See Learning: Machines, Brains, & Children.
2) Meaning is a fundamental human attribute that permeates all cognition, from low-level perceptual processing to high-level problem solving, and everything in between. Semantics is what makes us a powerful species—informavores. Without it, we’re really pretty pathetic: no claws, flat teeth, furless, can’t smell fear, and prone to sickness. Meaning is simultaneously the most obvious feature of cognition—we can all compute it rapidly and automatically—and the most mysterious aspect to study. Machines are not good at naturally learning, representing, and using semantics, and so fully understanding how humans do this is a priority that will have very real benefits to education, information retrieval, data mining, medical informatics, understanding cognitive decline, and the discovery of distributed knowledge from big data that is not currently known or located in any one place.
3) As we move into the ages of “data science” and “big data”, we need to acknowledge that humans are both the producers and consumers of a vast majority of these data. Current approaches to mine human-generated big data for insights employ statistical and machine learning techniques. While I agree that quantitative models are essential to understand patterns in big data, it is hopeless to believe that purely data-driven statistical models alone can do the job. Cognitive models are essential for identifying meaningful patterns in human-generated data. Employing models that are based on decades of research in psychology, cognitive science, and neuroscience give us a huge leg up to assist data-driven machine learning techniques.
3) As we move into the ages of “data science” and “big data”, we need to acknowledge that humans are both the producers and consumers of a vast majority of these data. Current approaches to mine human-generated big data for insights employ statistical and machine learning techniques. While I agree that quantitative models are essential to understand patterns in big data, it is hopeless to believe that purely data-driven statistical models alone can do the job. Cognitive models are essential for identifying meaningful patterns in human-generated data. Employing models that are based on decades of research in psychology, cognitive science, and neuroscience give us a huge leg up to assist data-driven machine learning techniques.
Jones, M. N. (2017). Big Data in Cognitive Science. Psychology Press: Taylor & Francis.
Jones, M. N. (2017). Developing cognitive theory by mining large-scale naturalistic data. I led the 2014 APS Crosscutting Symposium titled What Big Data Means for Psychological Science. See My FABBS press release: Teaching Computers to Think Like Humans |
Research following these themes has earned my students and me several prestigious awards, including the Early Career Investigator Award from the Federation of Associations of Behavior and Brain Sciences, the CAREER Award from the National Science Foundation, the Outstanding Early Career Award from the Psychonomic Society, the Google Faculty Research Award, and the Indiana University Outstanding Junior Faculty Award.
My research has been funded by grants from the National Science Foundation, the Institute of Education Sciences, the National Institutes of Health, Google Research, the IU Emerging Areas of Research Program, and the Indiana Clinical Translational Sciences Initiative.
Below are some summaries to clusters of recent research projects. For details and up to date publications, visit my Lab Website.
My research has been funded by grants from the National Science Foundation, the Institute of Education Sciences, the National Institutes of Health, Google Research, the IU Emerging Areas of Research Program, and the Indiana Clinical Translational Sciences Initiative.
Below are some summaries to clusters of recent research projects. For details and up to date publications, visit my Lab Website.