Beyond Artificial Intelligence

The year 2016 witnessed rapid strides in IT, moving us closer to what futurists call a singularity — AI (including sub disciplines such ad machine learning, deep learning, etc.) led unprecedented technology change resulting in a fundamental shift from the unrivalled dominance of human decision making. AI took the center-stage in 2016. IBM Watson’s prowess in cancer diagnosis and treatment is impressive. According to a University of North Carolina School of Medicine study, Watson recommended the same treatment as oncologists in 99 per cent of cases. More interestingly, Watson found treatment options that oncologists had missed in 30 per cent of the cases.

The march of progress in emerging sub-domains of AI like deep learning has also been impressive in 2016. The most spectacular was Google’s DeepMindAlphaGo system beating Lee Sedol, the legendary champion of Go — a board game of intense complexity. There are more possible positions in Go than the number of atoms in the universe. To provide a comparison, Go is googol (ten to the power hundred) times more complex than Chess. How did AlphaGo prepare for the face-off with human champions? After training on millions of moves by human experts, AlphaGo learned to discover new strategies for itself by playing thousands of games between its neural networks, and adjusting its game using a trial-and-error process.

The growing list of accolades for AI seems to suggest that it is almost mimicking how the human brain works. This is not the case. AI, including deep learning, is leveraging the “brute force black-box” of immense computing power and data communication speeds that are available on tap today. AI has benefitted from this exponential growth in IT performance. Even though AI is becoming an important decision tool for humans to use in increasingly complex contexts, it is not human intelligence.

The human brain is estimated to have about tens of billions of neurons and trillions of synaptic connections. The neurons and synaptic connections that control who we are and what we do define human intelligence. It works differently from how AI works. An AI system beats champions in Go or Chess based on immense computing capability, and not by replicating human intelligence. Understanding and mimicking human brain functions is in the realm of a multidisciplinary field (that includes AI, computer science, electronics, math, physics, biology, medicine, etc.) of computational neuroscience. It is a discipline that studies the information processing capability of the brain and the nervous system. One of the main differences between computational neuroscience and AI is that the former’s models and theories are tested in biological and physiological experiments. Most of computational neuroscience work is currently happening in university labs, and in startups that have a strong academic researcher/research connection.

A recent example of an interesting result in computational neuroscience was the discovery of a model that comes close to explaining the human brain’s facial recognition mechanism from MIT’s Center for Brains, Minds, and Machines. The model developed included an intermediate step of invariant representations of human faces, that is, mirror symmetric rotations of the face by a certain angle say 30 degrees to the left or 45 degrees to the right. This model of intermediate invariant representation is consistent with the neurophysiological experiments based on MRI scans of primate brains. Another interesting point to note is, this model takes about 80 to 100 milliseconds to identify a face which is similar to the time primates take. A facial recognition model that mimics how humans recognise faces has real world uses in making many applications such as security systems, authentication systems, facial unlock feature in smartphones, etc., much more accurate than what they are today. This helps in quicker threat assessment, and makes identity theft near impossible.

Computational neuroscience can also tangibly improve quality of human life when applied in neuroprosthesis — devices that aid human intelligence to improve the quality of life for people suffering from impaired brain function like those with Alzheimer’s, dementia, etc. Kernel, a startup in the USA, is experimenting with an implantable device that facilitates communication between brain cells, like how it is likely to happen in a healthy brain, to improve memory. This can boost the quality of life of a dementia patient, for example, by helping he/she to remember the location of bathroom in the house. Kernel is conducting trials on epilepsy patients, and is working on making the device truly portable.

Computational neuroscience is a grand challenge of the current era. The results from computational neuroscience can be applied to improve human life in many domains starting from medicine and IT. In 2013, the White House launched the BRAIN initiative — a multidisciplinary research program in partnership with premier US research institutions such as NIH, NSF, DARPA, FDA, and IARPA. Its aim is to map the human brain, and unravel how individual neurons and neural circuits interact temporally and spatially. The objectives of the BRAIN initiative include developing cures for brain disorders, and understanding how the brain processes, transmits, and stores information. The US is expected to spend more than $100 million on the BRAIN initiative. The benefit upside of the BRAIN initiative is expected to be similar to the one for the Human Genome Project — more than a hundred dollars of benefits for every one dollar spent on research.

Other countries such as China have also invested in computational neuroscience research. There is a national facility at the Institute of Neuroscience that is part of the prestigious Chinese Academy of Sciences. It is important for India to build a strong competence in computational neuroscience so that we are not left behind in this important domain. There are some green shoots in Indian research on computational neuroscience. The Center for Computational Brain Research (CCBR) at IIT Madras is one example. The focus of CCBR includes how knowledge of the brain’s functioning can be applied to enhance IT architecture, and develop next-generation algorithms to extend AI. The Centre for Brain Research (CBR) in Indian Institute of Science (IISc) is another initiative. The focus of CBR is clinical research to understand age-related brain disorders, and to model how the brain works. Both CCBR and CBR have multidisciplinary teams of faculty, and visiting distinguished faculty from among the best research centres in the world. We believe that these Indian initiatives will start a cascading effect for high quality research, and developing high quality talent pool, and may even spin-off interesting startups.

We expect computational neuroscience to take the global center-stage within the next decade. Ray Kurzweil, a futurist with an uncanny ability to forecast evolution of technology in recent times, has predicted that we will be able to expand our neocortex (part of the brain involved with higher order functions) via. brain-to-brain and brain-to-computer connections by 2030. It is an audacious moonshot that will place computational neuroscience at the pinnacle of human endeavour.