The Realities of AI

This week, I was able to explore the realities of AI development for corporations and the implications/challenges of modern day AI. I was able to meet with a TRI representative and discuss some of the process of research and development for AI in robotics and driving. A major limiting factor for AI in the modern day is the narrow applicability of each algorithm. While an algorithm can learn and recursively apply its AI principles, when an AI is placed out of its narrow circumstances, the metaphorical 1s and 0s behind its operations quickly fail. At a place like the Toyota Research Institute a major focal point of AI is self-driving technology. Driving is a very unstructured and unpredictable activity. This makes AI a very difficult concept to apply to a situation like driving where anything could happen. Ideas like defensive driving, driving around children, and congested traffic are all abstract concepts that are difficult to train into the neural network of any AI. Anybody who regularly drives a car knows the importance of intuition in the driver to account for unexpected scenarios. No matter the sensor data, efficiency, and speed of the hardware and the complexity of the software in the car, human intuition is extremely difficult to account for. This lack of abstraction is one of the reasons why AI took so long to prosper outside of closed, curated environments, such as factories and labs, into the real world. Once this problem is addressed, AI will truly be able to expand into our daily lives.