Excerpt from Doug Engelbart’s classic …
It is not a question of driving away, rejecting the I. That would be pointless— and contrary to the physiology of the psyche. It is a matter of following its movements and then adding others to it, which the I could not pretend to.
The point of arrival is a dual subject who is irreducible, unbalanced (the Self is infinite, the individual is any being whatsoever in this world), intermittent (the perception of the dual subject is not something there right from the start, but something to achieve, the hardest yet most efficacious achievement).
Calasso, Roberto (2014-11-18). Ardor (p. 128).
the entire vision community gets to ignore you when you insist that “manual feature engineering” is going to save the day
I think of Yann LeCun (one of the fathers of Deep Learning) as a modern day Descartes, only because I think the ground-breaking work is right around the corner.
It’s starting to become clear that the future power of deep learning is going to come with its own self-contained software package like Caffe or Torch, and not from a dying breed of all-around tool-belts like OpenCV or Matlab
ConvNets are here to stay, but if we want ConvNets to be more than than a mere calculus of shadows, there’s still ample work do be done.
The ones commonly covered are those with low skill requirements and that require less creative judgement. However, there are likely many other shifts in composition of work and through hyper specialization of tasks machines could automate pieces of work to a much larger extent. For instance – PCs and various office software transformed the way work was done in much more profound ways than elimination of certain job categories.
How can one not be excited by the future of machine learning?
We are our categories. This is illustrated beautifully by machine learning algorithms and their “free-association” pictures below.
It also makes us wonder whether neural networks could become a tool for artists—a new way to remix visual concepts—or perhaps even shed a little light on the roots of the creative process in general.
The work described shows the literality with which “training” encodes information into a machine learning system. Playing it back in reverse allows us to see recognizable but inhuman conceptual imprints that make the act of “recognition” possible. Wonderful.
Looks like imagination and memory are two sides of the same coin.
considerations, from evolutionary biology and quantum physics, suggest that current models of object perception require fundamental reformulation
natural selection often drives true perceptions to extinction when they compete with perceptions tuned to fitness rather than truth
Certain interpretations of quantum theory deny that dynamical properties of physical objects have definite values when unobserved
Objects of consciousness – A paper by Donald Hoffman and Chetan Prakash challenges the notion of perceptions approximating reality. It describes one of the boldest (to the point of feeling kooky) conjectures I have read in a while. However, there are no great objections to it yet. They have also been laudably transparent with their somewhat simple but powerful mathematical formalism. A more recent TED talk.
We generally believe that things contain atoms that we do not see, and are okay with this. However, we do believe that with the right magnifying glass we can get closer and closer to the *true* nature of things and at some point we will have it. This axiom that eventually seeing things as they really are is a biological given is an article of faith and does not necessarily have to be true. In fact, the authors suggest that our cognitive apparatus to see things has evolved to enable behavior that maximizes the survival of our genes and almost certainly conceal the truth to make it so. They suggest that the truth of how things are is cognitively “hacked” to provide an “interface” to the True world. The “interface” allows us to quickly behave so as to maximize reproductive success and survival.
They propose a purely mathematical framework to build a theory of all that is starting from conscious agents.
A diagram of a conscious agent. A conscious agent has six components as illustrated here. The maps P, D, and A can be thought of as communication channels.
We have taken some first steps by (1) proposing the formalism of conscious agents, (2) using that formalism to find solutions to the combination problem of consciousness, and (3) sketching how the asymptotic dynamics of conscious agents might lead to particles and space-time itself. Much work remains to flesh out this account. But if it succeeds, H. sapiens might just replace object permanence with objects of consciousness.
Could this be the gateway to a post-modern integration of physics and biology, not to mention psychology? It is hard to say No just yet, and impossible to say Yes. Lots more simulation will be needed to reproduce results of countless numbers of “physical” experiments. I do find the graph theoretical framework encouraging because it will likely make very good use of massive scale computation and data.
It also is a completely fresh and constructionist way of approaching the integration of objective and subjective accounts without dogmatic rhetoric of one form or another. A “killer-app” of the theory that could make some bold and verifiable prediction – impossible with our current physical theories – would really shake things up. Exciting stuff.
Big heart, slow-twitch, breathe well