Artificial intelligence engineers are being forced to employ psychology to help them understand the way that their own creations think.
With our ever-increasing reliance on digital brains to make complex information-based decisions — even putting them behind the wheels of our cars — we are almost absurdly incapable of understanding precisely how they arrive that the decisions they make.
The most complex A.I. is a mystery even to its creators, compiling and interpreting information with billions of virtual neurons that operate with unparalleled clinical efficiency, but zero self-awareness. As a result, it means that an A.I. will always take the information it collects at face value, even if that information contains the same inconsistencies and assumptions that cause errors in our own reasoning.
In the end, that is precisely the problem. We are teaching perfectly logical beings to reason, but we are incapable of matching that flawless efficiency of thought. Unless we can identify how a complex machine makes its choices, we risk creating entities that both share the flaws we are trying to eliminate, and yet are simultaneously so foreign that we cannot propose a useful solution.
That is precisely why DeepMind and other machine learning research groups are employing cognitive psychology techniques in order to understand the way that our most complex artificial intelligence entities interpret data to solve problems. In a quest to understand their thoughts, we must first understand how they think.
The problem can seem trivial at face value, but if you take a moment to consider an artificial being that will make split second decisions about the relative value of human life in the midst of a traffic accident, your perspective may shift. Artificial intelligence is here, and its utility is both inarguable and inevitable. But it will be up to us to decide exactly how intelligent it really is.
Follow Nate Church @Get2Church on Twitter for the latest news in gaming and technology, and snarky opinions on both.
COMMENTS
Please let us know if you're having issues with commenting.