Thursday, August 11, 2016

Deeper learning

Recently, statistical and mathematical techniques that have been tried and abandoned decades ago to teach computers to be smarter, have surfaced again with better sounding names. With nearly zero cost computing power in the cloud, some companies have been plunging head first into the deep abyss. Deep learning is stylish - some even try "deep mind," in an attempt to replicate the complex human. What the younger scientists may not know is that most of what they have found, has been known for a long time and one should not take advancements, due to only the recent availability of cheap computing power and memory, as "ground-breaking." Disappointments may be in store for those, highly optimistic of "solving," human intelligence. The convergence of Neuroscience and Computer Science is a welcome trend, but a dose of realism may be apt medicine for those modeling the mind, downloading the brain and possibly curing physical death.

Even since she stood up in the African Savannah, a few hundred thousand years ago, the human has been puzzled by her own mind. She searched the skies, climbed mountains and dived into the oceans, seeking the object, she could not herself define. The theory of consciousness has eluded her, for what she was seeking obviously interfered with the tools she was using. It was the ultimate prize. If she could understand herself, then, a whole new world could open up. The body can be separated from the mind, and the latter then could be rendered immortal. The mind could be replicated and networked and perpetuated across space and time. She could create and capture imagination at will. She could solve problems of infinite complexity, travel into interstellar space or even to another universe. If only, she could understand the mind and concoct a theory of consciousness. But alas. it is not to be. Whatever one calls it, the "neural network," has failed to show signs of consciousness. Yet another technology is substantially sub-optimized by engineers and scientists, most comfortable with deterministic answers to complex questions.

Ignorance is highly scalable in the presence of infinite computing power.