AI's inherent big challenge: Dealing with surprises

by Roger Bourke White Jr., copyright April 2021


How big does an artificial intelligence (AI) system need to be, as in, how much data must be installed for it to function well in getting a mission accomplished?

The answer is: count the surprises it has to deal with -- the more surprises the more data needed.

Here is a simple example. Think of the thermostat in a room as an AI system. How much data does it need? All it needs to know is the temperature of the room and how to turn the on furnace when the room gets too cold. Pretty simple. Now add air conditioning to the the system. Now the AI needs to know if the room is too cold or too hot and it needs to know which to turn on: the furnace or the air conditioner. Adding this air conditioning feature has doubled the data needed.

What this means is that the more surprises an environment offers, the bigger the AI system has to be.

A real-world example of this is the difference between a delivery robot navigating a well-designed warehouse and a delivery robot navigating city streets -- the latter has a whole lot more surprises to deal with so the AI is going to have to be many times bigger.

In sum, AI size is all about surprises.



--The End--