Feb 1, 2018
Qualcomm products mentioned within this post are offered by Qualcomm Technologies, Inc. and/or its subsidiaries.
We’re in the early days of AI development, and right now, the entire industry has more questions than answers on how far we can take the AI revolution. There are a lot of unknowns, and there’s an interesting contrast between what AI is now and what it can be. Yes, AI is already doing tremendous things and improving at exponential rates, but it’s still incapable of doing many tasks that humans consider simple or that children can learn in minutes given a few examples. We simply don’t completely know how the human brain works and why it’s so efficient at so many tasks.
Qualcomm Research has been researching AI for over a decade, starting with brain-inspired spiking neural networks, to make the devices and the things we interact with everyday more intelligent. We’ve worked closely with other leaders in AI, from academia to industry, to push AI forward. As you can imagine, we’ve encountered diverse opinions about the future of AI and have continued those discussions and debates to advance our thinking and knowledge, as well as the industry.
To that end, we brought together two world-renowned thinkers in the AI space for a friendly debate: mathematical physicist, author, and professor Sir Roger Penrose and Dr. Eugene Izhikevich, a neuroscientist and the co-founder, Chairman, and CEO of Brain Corporation. What could be better than two titans of AI, with contrasting opinions, having a healthy debate? I had the honor to moderate the debate, which was hosted by Joan and Dr. Irwin Jacobs. The debate format comprised opening statements, moderated questions with the opportunity for rebuttal, and audience questions followed by final statements.
The premise of the debate centered around how far AI will progress. Are we on the verge of artificial general intelligence – machines that possess the true understanding and consciousness that human brains display? Is it possible with the continued advancement of today’s digital technology – custom hardware architectures, novel machine learning algorithms, and more digital gates – to ever achieve such goals? Or, is it the case that digital gates will never be enough and rather we must look to new physics to achieve the conscious intelligence that humans possess?
Eugene made the case that modelling neurons with digital technology can achieve general intelligence that would appear conscious. In effect, a large enough collection of gates based on classical physics is sufficient to emulate the human mind. Roger, however, argued that we need to modify our theories of the physical world to achieve true brain-like functions, such as consciousness and genuine understanding. Specifically, the rules of quantum mechanics need to change in the light of general relativity. His view is that any large collection of digital gates will always fall short of general intelligence.
The point of the debate was not to crown a winner and loser. Rather, it was to learn something new and advance the thinking about AI. The future of AI is still to be determined, but here are some other key insights I gained:
- There is no agreed upon test for consciousness or understanding. As Eugene explained, as Claude Shannon helped define information theory, we need to define consciousness to further our progress.
- While existing neuron models for AI are “simple” and may not achieve consciousness, they are providing amazing capabilities today that will only continue to improve.
- The brain is a complicated biological and chemical system, but biology may not be the only way to achieve consciousness.
Don’t just take my word for it: You can watch highlights of this fascinating debate above, or check it out in its entirety on YouTube.