Albus et al. (2007) proposed the Decade of the Mind. Many partial models about cortex or mind have been proposed. Symbolic or partially symbolic cortical models include Albus 1991, Lee & Mumford 2003,  George & Hawkins 2009, Taylor et al. 2009, Grossberg & coworkers 2009, Albus 2010, and Tenenbaum et al. 2011. Symbolic models have their advantages such as intuitiveness and abstraction. Why prior neural networks (e.g., Hinton et al. 2006, Poggio & coworkers 2007) can do pattern recognition but do not abstract well (Minsky 1991, also stated by Michael I. Jordan at the David Rumelhart Memorial Plenary Talk, IJCNN 2011)? How do we understand biological emergence of all phenotype from the single cell zygote? What are the strengths and limitations of symbolic representations? Does the brain use non-symbolic, fully emergent internal representations (Weng 2012)?

A computational brain-mind model, called Developmental Network (DN) (Weng 2010), seemingly the first in the 5-chunk scale (development, architecture, area, space and time) addresses some of those questions. For example, such a DN learns any given complex Finite Automaton incrementally, immediately, and error free, but does not need human handcrafted internal representations. We seem to have passed “neural networks do not abstract well” (Weng 2011).

While many questions need clearer answers with more detail, much published work in the literature and the DN model indicated that a certain breadth and depth of knowledge in at least 6 disciplines — Biology, Neuroscience, Psychology, Computer Science, Electrical Engineering, and Mathematics — is necessary to understand any (5+1)-chunk brain-mind model (see a tutorial). It is also reasonable to expect that such 6-disciplinary knowledge is required for understanding and evaluating future more complete and more detailed brain-mind models. Collectively, the human race seems to have obtained sufficient knowledge necessary to solve, at least in principle, the grand puzzle of the brain-mind. Doing little at this point of time is of high risk.

Unfortunately, the brain-mind nature is not “aware” of how the human race has built its academic and research infrastructures, as the nature exists first. Indeed, many existing multi-disciplinary research projects have already reflected such a growing need for knowledge beyond one's home discipline. However, the existing infrastructures, in developed and developing countries alike, seem insufficient for a wide-front computational understanding of the brain-mind — they do not span the necessary 6 disciplines. Hopefully, this Institute could serve the infrastructure needs in those countries.

Natural Intelligence View

Traditionally trained biologists, neuroscientists, and psychologists should better understand the underlying principles of their subjects of study with the help of sufficient knowledge in computer science (CS), electrical engineering (EE) and mathematics. This seems reasonable as the brain deals with quantitative relationships of the natural world (mathematics), as a high-dimensional numeric nonlinear system (EE) and it attends and reasons in computational ways (CS).

On one hand, rich studies in biology, neuroscience and psychology seem to have already provided sufficient qualitative information for piecing together the grand puzzle of the brain-mind. However, as piecing together requires computational knowledge in CS, EE, and mathematics, only those researchers who have sufficient knowledge in all six disciplines can understand this exciting situation and contribute to it. Furthermore, only those researchers who have sufficient knowledge in all six disciplines can understand such a picture of the brain-mind.

On the other hand, biologists, neuroscientists, and psychologists are experts in their own focused subjects of study. Sufficient in-depth knowledge in CS, EE, and mathematics enable them to understand how the brain-mind develops and works so that their future experimental designs are better targeted, richer, and more quantitative.

Artificial Intelligence View

Traditionally trained computer scientists, electrical engineers, and mathematicians may be able to solve their major bottleneck problems in their home disciplines using the beautiful solutions discovered from the brain-mind.

Currently, the field of intelligence modeling is fragmented. Although many such models were inspired by either the brain proper or observed brain behaviors, the variation of approaches is astonishing. At one end, much work in computational neuroscience focuses on detailed neuro-dynamics of a single neuron. At the other end, many models in psychology, artificial intelligence, robotics, and mathematics use symbolic models that are drastically different from brain's internal mechanisms or emergent (neural nets) models that do not abstract well. Both type of models have "hard ceilings" in understanding brain's information processing and in reaching human-level performance. The division of these two categories of models has been theoretically bridged recently (Weng 2010).

In a broader scope, the digital era has created an unprecedented challenge in analyzing vast amounts of digital data in many domains. Existing computer methods do well in highly controlled settings (e.g., iris recognition, playing chess), but do not do well for data from less controlled settings, including images, video, audio, and text streams. The above new advance about brain-mind indicates that the brain-mind seems strong in automatically identifying relevant information and relationships (i.e., emergent representations) without a need for human-handcrafted representations for each task — an intractable handcrafting job for uncontrolled settings. Traditional computer methods (both symbolic and emergent) do not effectively attend to objects and relationships in data that a person's brain does not disregard, since much of the data are distractors (e.g., background activities irrelevant to the applications). Only recent brain-mind models have demonstrated effective attention in natural complex backgrounds. Furthermore, the human brain seems especially strong in automatically integrating multiple sources of information. It seems now tractable to formulate large-scale brain-like information processing in an optimal way, which has a potential to greatly reduce the time needed, and improve the performance, for screening and analyzing large volumes of data arising in the digital era, from Internet, to wireless communication, to sensor networks, to geospatial coding / analysis, and to autonomous robots.

Why Now?

The brain-mind is no longer as mysterious as 20 years ago. For example, Sur & coworkers 2000 demonstrated that a cortical area is not a special-purpose processor through a developmental process in which they let the mechanisms of auditory cortex demonstrate its power in learning visual signals. The complex connections among brain's visual and motor systems (e.g., the complex diagrams and charts in Felleman & Van Essen 1991) are largely consistent, in principle, with the statistics based principles of the 5-chunk brain-mind model (Weng 2010). This model maps a Finite Automaton — the “common-denominator” model of many symbolic computer models — to a brain-anatomy inspired emergent Developmental Network (DN) model, having positively answered the well-known open question (e.g., Minsky 1991) — “Can emergent networks abstract and reason well?” It is true that this 5-chunk model is probably still controversial since fully understanding it requires sufficient 6-discipline knowledge. The model is also subject to further biological verification and enrichment. However, the following two points seem clear: (1) Collectively humans can already understand, in principle, how the brain-mind works in terms of computation; but such human collective knowledge is in many small pieces — a few pieces in the mind of each single researcher. (2) Knowledge in all the 6 disciplines is necessary for understanding brain-mind either quantitatively or qualitatively. Obviously, such a situation will not change fundamentally in many years to come without drastic infrastructure actions.

Doing little at this critical time seems to be of high risk for any nation. Our ancestors have solved many problems contained in a single discipline. In human history, the nations where major scientific discoveries took place indeed led the world that time. Our generations are left with problems that one or two disciplines cannot adequately understand and solve. We need to be creative in transforming the infrastructure we inherited from our ancestors so that those high-impact cross-disciplinary problems can be solved while we humans collectively already can. Facing intense global competition, any nation with vision would seize this opportunity in order to take the lead in the upcoming brain-mind era. It should not wait and do little. This was how the need for this Brain-Mind Institute arose in the United States of America. However, for basic science, this BMI is international, open to anybody on this planet.

(Updated July 18, 2012)

References

J. S. Albus. Outline for a theory of intelligence. IEEE Trans. Systems, Man and Cybernetics, 21(3):473–509, May/June 1991.

J. S. Albus, G. A. Bekey, J. H. Holland, N. G. Kanwisher, J. L. Krichmar, M. Mishkin, D. S. Modha, M. E. Raichle, G. M. Shepherd, and G. Tononi. A proposal for a decade of the mind initiative. Science, 317:1321, 2007.

J. S. Albus. A model of computation and representation in the brain. Information Science, 180(9):1519–1554, 2010.

A. Fazl, S. Grossberg, and E. Mingolla. View-invariant object category learning, recognition, and search: How spatial and object attention are coordinated using surface-based attentional shrouds. Cognitive Psychology, 58:1–48, 2009.

D. J. Felleman and D. C. Van Essen. Distributed hierarchical processing in the primate cerebral cortex. Cerebral Cortex, 1:1-47, 1991.

D. George and J. Hawkins. Towards a mathematical theory of cortical micro-circuits. PLoS Computational Biology, 5(10):1–26, 2009.

G. E. Hinton, S. Osindero, and Y.-. W. Teh, A fast learning algorithm for deep belief nets. Neural Computation, vol. 18, pp. 1527–1554,
2006.

T. S. Lee and D. Mumford. Hierarchical Bayesian inference in the visual cortex. J. Opt. Soc. Am. A, 20(7):1434–1448, 2003.

M. Minsky. Logical versus analogical or symbolic versus connectionist or neat versus scruffy. AI Magazine, 12:34-51, 1991.

T. Serre, L.Wolf, S. Bileschi, M. Riesenhuber, and T. Poggio. Robust object recognition with cortex-like mechanisms. IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 29, no. 3, pp. 411–426, 2007.

J.G. Taylor, M. Hartley, N. Taylor, C. Panchev, and S. Kasderidis. A hierarchical attention-based neural network architecture, based on human brain guidance, for perception, conceptualisation, action and reasoning. Image and Visual Computing, 27(11):1641–1657, 2009.

J. B. Tenenbaum, C. Kemp, T. L. Griffiths, and N. D. Goodman. How to grow a mind: Statistics, structure, and abstraction. Science, 331:1279–1285, 2011.

L. VonMelchner, S. L. Pallas, and M. Sur. Visual behaviour mediated by retinal projections directed to the auditory pathway. Nature, 404: 871-876, 2000.

J. Weng. A 5-chunk developmental brain-mind network model for multiple events in complex backgrounds. In Proc. International Joint Conference on Neural Networks, pages 1–8, Barcelona, Spain, July 18-23 2010.

J. Weng, Why have we passed “Neural networks do not abstract well”? Natural Intelligence: the INNS Magazine, vol. 1, no. 1, pp. 13-22, 2011.

J.Weng. Symbolic models and emergent models: A review. IEEE Transactions on Autonomous Mental Development,vol. 4, no. 1, pp 29-54, 2012.