Harvard and MIT Join to Address Intelligence

We were heavily involved in the last Artificial Intelligence hype peak in the late 80s.  Despite some big successes we saw it languish, at least in direct practice,  for years.  Then along came IBM's Watson to show there was real life in the idea.   As I thought would occur, a number of efforts are now underway to bring it back.   I hope for practical application.  Now richly endowed with more and better data.  And Brain models are back too.   " ... Computerworld - The National Science Foundation has awarded a $25 million grant to Harvard and Massachusetts Institute of Technology to study how the brain creates intelligence and how that process can be replicated in machines. ... "    Exactly what this will be called and positioned  remains to be seen.  But I notice that researchers have stopped shying away from the term Artificial Intelligence.

Predictive Assistants that Take the Initiative

We worked with and visited the nonprofit research group SRI a number of times.  They originally had been spun out of Stanford University.   They were involved in detergent formulations for our enterprise years ago. The computer mouse had been invented in their labs.  I had forgotten that they had been involved with the development of the assistant Siri.  

They are now developing a system called Bright that looks at data rich and intense environments where the system that is helping needs to model the human user and take the initiative to solve a problem.  Such problems, which we designed for enterprise AI applications, are much more difficult to deliver than simple conversational, question and answer systems like Siri.

  More about the Bright environment and where it is going in Technology Review.  Note this also has similarities to Google Now but expanding the initiative aspect.

Deep Learning with Neural Nets

In CACM:  Deep learning finally comes of age.  Still a bit skeptical about what deep really means here.  I need it it be useful.   We used neural nets extensively to replace specific statistical methods in decision resolution, but not to generally 'learn'.  I still don't see how what they suggest here provides generalized learning,  but the progress may be there.  High speed parallel methods of the kind used in 'Big' methods may be applicable.   " ...  "A wave of excitement today comes from the application of unsupervised learning to deep neural nets," LeCun says, adding that another wave of excitement surrounds the use of the more-traditional supervised training of multilayer systems called "convolutional" neural nets. LeCun developed convolutional neural nets at Bell Laboratories in the late 1980s; they were among the earliest to employ multilayer learning. ... " 

Note that the method of Neural Networks  is more correctly called Artificial Neural Nets (ANN).    The neural models used in these methods are only superficially inspired by animal neurons.  They are also not currently used to my knowledge in neuroscience or neuromarketing methods,