Designing Secure AI: The Primary Obstacle

By CIOReview | Wednesday, November 29, 2017

The silos to develop a more ‘safe’ and ‘secure’ AI solution have been surmounting, and the real reason for the same has been an ambiguity in comprehension of the two terms. While next-generation developers of AI mostly are involved with the technical aspects of safety and security—prevention of owner’s injuries, protection from malicious threats; the definition is more subjective that concerns the overall “well-being of humanity.”

In our highly diversified world, where values change from people to people, the incompetence to codify a coherent view of ‘well-being’ into the systems’ algorithmic architecture has become the primary obstacle in keeping AI safe and secure. The challenge is more highlightedwhen considered the fact that AI systems are made to learn from abstract situations and not conventional computational algorithms. This demands the creation of intelligent systems that provide safety and security not just for specialized situations like accidents, but for its whole operational lifecycle as it continues to learn from our ambiguous and diversified world without being negatively affected.

The moment has come to refine ourunderstanding of safety and security, and developers need to have an alignment with the values that needed to be coded into the AI systems. Since safety and security can’t be quantified by mathematical logic and fulfillment of some determined details, the solution is to clarify our view and comprehension of the terms so that the designed AI can comprehend safety and security more clearly. Thus, everything depends on how much humankind can refine its thoughts and understanding to develop ‘safe and secure’ AI systems that precisely understand the meaning of the same.