Ted Talk Summary: The danger of AI is weirder than you think | Janelle Shane

 Video Link: https://www.youtube.com/watch?v=OhCzX0iLnOc

AI is known for disrupting many industries.

In an experiment, the speaker fed an AI application all the different flavors of ice cream in order to obtain a new flavor of ice cream from the AI application. The AI came with flavors like Pumpkin Trash break, Peanut Butter Slime, Strawberry Cream Disease. Clearly, all of them not fit for human consumption. In movies, when something goes wrong with an AI robot, it is usually because it doesn’t want to obey humans any more. However, in real life, the AI is not smart enough for that.

While AI is able to understand things, it is not able to make sense of the world as well as humans. In reality, AI does not even know what a human really is. While AI will do what we tell it do, it may not entirely do it the way we want to, resulting in an outcome that is not desirable for us. For example, if we ask AI to assemble a robot and get from point A to point B, it may not assemble a robot that has legs that can walk. It may simply create a bridge to connect point A with point B. This is not exactly what the human may have in mind. So the problem with AI is not that it won’t obey us, but it will do exactly as we tell it to do.

So working with AI is lot less than working with a human, but more like with a weird force of nature. And it’s very easy to give an AI the wrong problem to solve.

There was another example given of Tesla car accident in 2016. In that, the car was taught to run on autopilot on highways. However, the driver was using it in the city. The car got confused when it saw a truck coming from the side.