For the past years, the reciprocal relations between biological intelligence and artificial intelligence might well become a love story. Insights from cognitive neuroscience have been providing principles that could be applied to computers, paving the way towards powerful AI technology, from artificial neural networks and machine learning to neuromorphic chips. These artificial brainchilds were quick to repay the debt. With powerful processing capacity, an extraordinary ability to identify patterns in vast amounts of data, and with a deliberate similarity to neural structures, AI has been helping us better understand the human mind using big data, brain imaging and simulations.
When Computers and Humans were Strangers
For many decades, the marriage between AI and the brain was stalling. It’s not that people did not see the connection, but that neither domain had a strong enough footing to support the other. Computers were simply not sophisticated enough. The earliest substantial work in the field of artificial intelligence was done in the mid-20th century by the British logician and computer pioneer Alan Turing, who in 1935 laid the blueprints for the Turing Machine – a computer meant to perform tasks requiring human intelligence. Early efforts to build AI involved decision-making processes and information storage systems that were loosely inspired by the way humans seemed to think. In 1949, Edmund Berkeley wrote about “strange giant machines that… are similar to what a brain would be if it were made of hardware and wire instead of flesh and nerves… Machines that can think.” The problem is, we were very far from understanding what the brain is like and what makes it think.
Our understanding of the brain was too meagre to inspire any implications for computer science: how does the brain make calculations? How does it store and retrieve information, or learn, induce and deduce? We had no access to the structure of neurons and neural networks. Two major obstacles were the inability to measure brain activity and map out its circuits, and the lack of computing techniques that would allow us to identify, characterize, and simulate these structures and activities.
Grey Matter as a Blueprint for Artificial Intelligence
Demis Hassabis, CEO and co-founder of DeepMind, a company researching and developing AI systems, considers the brain as a paramount inspiration for building AI with human-like intelligence. In recent decades, neuroscientists have learned a huge amount about the brain’s architecture and how it processes information, insights which are already bearing fruit and confirming Hassabis’s suspicions. The meteoric success of deep learning showcases how models of human memory, learning, decision-making, or perception can be distilled into algorithms that bestow silicon minds with human capacities.
Neither is the entrepreneurial spirit indifferent to the power neuroscience has to offer high-tech. For example, Silicon Valley’s Numenta was a neuroscience research company that after a series of breakthroughs, has changed its focus from brains to AI, applying to machines what it learned about biological intelligence. By emulating the way interconnected brain cells function, artificial neural networks (ANN) are trained to learn, recognize patterns, and make predictions – powerful tools for the business sector. Alex Cardinell, Founder and CEO of Cortx, a company that uses neural networks in the design of its natural language processing solutions, also looks to the brain-functionality for insights into machine learning. As he explains, neurons continually adjust how they react based on stimuli. Neurons yielding a correct output receive positive feedback, and consequently become even more likely to activate in similar situations. Conversely, receiving negative feedback will make these neurons less likely to activate.
As these insights reflect, the bulk of R&D has been focusing on the functional aspect of human cognition, or the way it behaves. However, experts are voicing the need to take into account the intrinsic structure of the brain and its components as well, and they are heard. Neuromorphic computing mimics neuro-biological architectures in electronic chips, increasing the processing efficiency and decreasing the energy consumption. This is the case with Innatera Nanosystems, a Dutch company designing neuromorphic intelligence for processing sensory data. With a combination of low power consumption and short response latency, their devices enable high-performance pattern recognition capabilities.
Artificial Intelligence giving a hand to Biological Intelligence
Since machine learning and ANN have drawn their inspiration from Neuroscience, it’s no surprise that neuroscientists are now using AI to re-examine ideas on how our brains function. Neuroscientific studies always relied on observing biological properties – receptors, neurotransmitters, signaling molecules; now, AI is rapidly gaining ground as an invaluable tool in neuroscience in two main ways.
First, AI has proven irreplaceable for sifting through vast amounts of data to find patterns. Sophisticated algorithms are being increasingly adopted to organize and make sense of brain activity using big data. Companies like Brain Patch rely on AI to optimize and personalize brain stimulation for research and treatment. Rather than studying individual proteins or brain regions, neuroscientists have the tools to profile single neurons or digitally reconstruct massive portions of neural connections.
This computational process has been more than key to brain imaging. MRI, EEG and many more techniques have given researchers a view into the workings of the mind. However, learning what these images are trying to tell us requires new and more powerful computing solutions. Recently, China-based United Imaging has been pairing advanced imaging techniques with AI to better map both healthy and non-healthy brains. Such techniques can speed up research in brain mapping projects or functional brain imaging, which easily deal with terabytes of data files that need to be processed, organized, and interpreted.
With its ability to identify patterns in large, complex data sets, as well as its reliance on brain principles, AI has seen remarkable successes in emulating how the brain performs certain computations. As algorithms increasingly evolve brain-like outputs, they can serve as models for developing and testing ideas in neuroscience. Algorithms that mimic vision or hearing can spur ideas on how the brain solves these tasks. “If you can train a neural network to do it,” said Dr. David Sussillo, a computational neuroscientist at Google Brain, “then perhaps you can understand how that network functions, and then use that to understand the biological data.”
There is little doubt left that cognitive neuroscience and artificial intelligence belong together. This dialectic of mutual advancement is only at its start. The more we unlock the secrets of the mind, the better we can design machines that will further help us understand ourselves. Perhaps the day is not far when man and machine will learn about each other over a cup of coffee… or motor oil.