What’s All the Fuss about Deep Learning?

Deep learning is the cutting edge of machine learning, or the newest practice leading towards Artificial intelligence (AI). There are a few experts in this area, and world’s leading corporations are racing to get them on their team. As with all computer technology innovations, the titans like Google, Microsoft, Yahoo and Facebook are at the forefront.

Using an intricate web of algorithms, these machines can learn to navigate through limited information available and ‘draw conclusions’ about required actions. A traditional neural network is created as a network of ‘neurons’ which can adjust their connections depending on the flow of external information. Google’s Neural Turing Machine is able to mimic the human brain’s working/short-term memory and retrieve necessary information from external sources, even predict future actions. They aim towards developing a neural network machine capable of programming itself and performing tasks for which it hasn’t been previously trained to do.

The real human-like intelligence would require the ability to work with ‘chunks of information’, as defined by famous cognitive psychologist George A. Miller, and recode them into one single ‘chunk’ that will make sense to our brain. Researchers at Stanford are on a good track to achieve this. Their Neourogrid project focuses on the creation of circuit boards that can simulate workings of multiple cortical areas, a million of silicone neurons and billions of synaptic connections working in a nonlinear mode.

Similarly, IBM’s neurosynaptic computer chip, the True North chip, introduced in August this year, has one million neurons and 256 programmable synapses, capable of computing 46 billion synaptic operations per second. This invention is a part of IBM’s contribution to DARPA’s SyNAPSE project, which is a bacronym for Neuromorphic Adaptive Plastic Scalable Electronics, technology that mimics human brain. IBM Almaden Research Center heralded their invention as “a new machine for a new era”.

Both Stanford’s and IBM’s cognitive computing chips are opening doors to the development of new supercomputers that can function on small power supplies, equal or less than ordinary desktop computer.

A number of other projects are under way that will bring artificial intelligence closer to the human-brain operating method, thus enabling machines to feel their environment, and respond accordingly in a human-like manner. A heavily funded European Union’s Human Brain Project (HBP) is aiming to map and simulate workings of a human brain in large-scale computational models. Across the pond, similar neuroscience studies are backed by Obama’s BRAIN initiative, which stands for Brain Research through Advancing Innovative Neurotechnologies. The ultimate goal of this initiative is to map activities of neurons in human brain and capture its dynamics. Once human brain is mapped and understood, technology can easily replicate and augment its processes.

Combined efforts of neuroscientists, software developers, experts in biotechnology, robotics and many other scientific and engineering branches will most definitely yield some tangible results in years to come, approaching us to the point of singularity. For some people, like the founder of Space X and Tesla Motors, Elon Musk, it is frightening. According to him, artificial intelligence could become more dangerous than nuclear weapons. Is there a real potential in robots with artificial intelligence for “killing us out of kindness”, as Nell Watson, an engineer and futurist, recently pointed out? Failing to understand our core values, and our human deficiencies, it may appear as a rational decision to them at one instance.