I understand your POV completely but I do not agree with your premise that a biological approach is a long ways off. It could be only if the resources are not devoted to the creation and development of Artificial Biological Intelligence (ABI) but I think we are fast approaching a paradigm shift and the computation models will be relegated to specific tasks and/or assistants to ABI. One does not have to emulate the human brain to develop highly intelligent systems. I would venture to say that we will see many degrees of intelligence emerge from ABI that will be very useful and we can deploy for various tasks. I agree (from you neuroscience article) that we do not need to emulate the human brain “as is” to develop intelligent systems; and in fact, I have dedicated my past few years to the development of artificial connectomes in order to circumvent all the unnecessary circuitry that inorganic living beings do not need (e.g. breathing, defecation, etc).
I lectured at Numenta back in January and I am also a big fan of their work. I had lunch with Francisco Webber (Cortical.io) in February who uses Numenta’s engine to develop some very cool technology. These are wonderful people to say the least and brilliant beyond compare. Although they have developed a system that can do predictive analysis probably better than most, they have not figured out how to connect the sensory input to motor output which is one of the things that most AI paradigms lack. This process is built into the ABI paradigm. I think the one thing to keep very much to the forefront when talking about Numenta, with every lecture given by Numenta staff, they start by saying their technology is biologically inspired but is not an emulation of biology. There is a big difference and one that the mathematical approach is blind too.
Obviously this subject is way too complicated to discuss in a comments section but let me say that ABI is coming, is working and has tremendous advantages over Deep Learning and other mathematical approaches. One of the perhaps greatest advantages is that is does not require the computing power these other paradigms require and in fact, I have ran ABI systems very well on Raspberry Pi computers. Computing power is Deep Learning Achilles heel.
Besides Numenta, check out works by Monica Anderson and keep an eye out for PROME. The reason I think a lot of people, especially Comp Sci types, do not work towards ABI is because it’s very complicated in its scope and Deep Learning et al is much easier to work with and understand but then again, that ease of use and understanding is why it will always remain mediocre and limited at best. The hard stuff is only hard until we understand how it works.