I have been reading about every book that comes out about AI, the most recent being “Surviving AI: The promise and peril of artificial Intelligence,” by Calum Chase. The one thing that is obvious from most of the books I have read on AI is that the authors believe that if you accept evolution, you must believe in the inevitability of Artificial Intelligence.
If the function of the brain could have evolved by the randomness of natural selection acting on the genetic variation among individuals, the equivalent ability can certainly be obtained in a powerful computer using the scientific method, without all the trial and error. Of course, this ability may take a totally different form, just like planes don’t fly like birds and submarines don’t swim like fish. And the exact timing of when this will happen is obviously not predictable. Per most authors, only people believing in a creator that made humans special such that no other entity can “think” should doubt the inevitability of AI.
Although most books acknowledge the possibility that AI will not come until we have a full understanding of the human brain, most suggest that AI will evolve from a different approach, either from a master algorithm leading directly into AI or the cumulative result of many isolated self-learning AI algorithms that are eventually harmonized into an overall thinking system.
I am watching Tesla’s “Autopilot” (note I own some Tesla stock) with great interest. Mobileye makes some of the technology, with chips that interpret what car cameras are seeing. The chips use “deep learning” methods to interpret data coming from the car’s sensors. Deep learning uses multiple simplified algorithms to approximate complex functions coming from the various vision systems and other sensors. These simplified algorithms enable quick but sufficient decision making by the car’s computer systems to control safety systems and driver assists.
The Tesla system is self-learning and will improve on its own based on when and how the driver and the systems react to individual road situations. It also will give feedback on the roads themselves, such that over time Tesla will have road maps far more detailed than anything currently available from the likes of GPS or Google Maps. These maps will be continuously updated based on the feedback coming from each Tesla vehicle. Even at his early stage, over 1 million miles are being driven daily by Tesla cars on Autopilot. The detail being gathered will enable the next step in self-driving vehicles.
Is this AI? The program is self-learning and is probably already quite different than what the human programmers initially entered. And certainly within a few years the decision making going on in each vehicle will be impressive. In fact, Elon Musk, the CEO of Tesla, believes that within five years his vehicles will be capable of self-driving.
With little doubt, the same sort of progress is being made on stock market programs, medical diagnostic systems, legal research programs, teaching assists, military weapons, and other areas that will not be as visible to us until fully implemented. And will all these systems approach problems with the same “deep learning” methods used in Tesla’s Autopilot, such that a commonality exists that enables a master algorithm that will work much like the human mind?
Progress in these areas is so rapid that the next few years are going to be very exciting. This is NOT something that will take decades before our way of living is dramatically affected! And this disruption will be apparent far before an AI system becomes truly “thinking.”