- Posted by Michael Leventhal
- On June 16, 2017
The idea of an AI has been toyed with across science fiction for decades now, in part because of our fear that our lives will end after uttering “open the pod bay doors.” But what is AI really about, and what can we expect of it right now?
Artificial intelligence in the modern understanding is comprised of a large amount of parallel processing units working together to process large quantities of information extremely quickly. This tech provides the underlying computational power that drives the design of any VR/AR/MR device or application. Every operation is built upon some form of pattern recognition, whether it be movements, colors, shapes, sounds, etc. There are already more available programs than we can name that could not function without the use of AI.
For instance, the application Prisma uses artificial intelligence to turn ordinary photographs into works of art. Users decide which styles they wish to emulate, and the underlying AI, which has access to countless examples of intricately analyzed paintings and designs, will recreate your photograph in the way Van Gogh might have seen it. How long do you think it will be before our head-mounted wearables make this happen in real time?
With that being said, pattern recognition through parallel processing units doesn’t qualify a computer as an AI. To be deemed artificial intelligence, it has to take this a step further.
Imagine a program that plays Go against a human user. Let’s say that we start it out as an undeveloped set of parallel processors with no real experience of the game. You play it once and easily win. It records this information and designates it by the end result. Over time, as you play more and more games, the program learns your strategies, determines their inevitable results and designs opposing strategies for combatting your plans of attack.
As you get deeper into the process, the AI will have learned so much from the varied experiences of playing you, that it will not only react competently to your moves as you reuse them, but also use its memory of many former outcomes to predict possibilities it has not yet encountered. In other words, it will learn over time, as the compounded rudimentary lessons grow into a vast basis for predictable future action.
This is exactly what google created with the Deepmind project. And now their creation has surpassed the most accomplished human masters of the game Go, a game with so many possible moves, they outnumber the particles in the known universe. In fact, it is even said that the calculations are too much for any known computer, including Deepmind. Some people believe that to succeed at Go, one must possess genuine intuition.
This brings us to the most fascinating fact about Deepmind, it will continue to learn even after you have stopped teaching it. But how could this be possible?
It does this through a process called backpropogation, a process that allows an AI to revisit already experienced scenarios on its own, similar to the human memory. It is capable of picking out particular details and determining patterns that occur within these experiences. Using this process, DeepMind was able to teach itself sophisticated pattern recognition, which is exactly the sort of function that would allow an AI present in a VR/AR/MR system to analyze many different eye movements, arm movements, sonic patterns or still images and compile them into a working system of identification.
For example, DeepMind was fed over 128,000 retinal images in an effort to have it independently create an image of a retina. Over extended comparison, DeepMind began to see similarities between diseased eyes that healthy eyes did not possess, and it did so without any guidance. At first it had no way of knowing what the pattern suggested, only that it was there. But, by the end of the process, Deepmind was capable of identifying diseased eyes with greater accuracy than any human doctor.
This is the distinction between an AI and an extremely powerful computer. The processing power is similar, but the AI has the capability of retracing its own logical/procedural steps without any command to do so. It learns autonomously and updates itself so as to optimize its own operations. Some find it intimidating that our creation might grow without our direct supervision, but in a sense, this is the same struggle we experience with our children. Others believe that our future will be designed by AI, reaching for further programs and ideas that simply extend beyond the boundaries of human thought.