Predictions, predictions vol 2
Last time we discussed about the theory of determinism in the sense of believing that our universe is strictly governed by cause and effect and thus our future is pre-decided. In this post, I'd like to explore a more mundane and earthly application of determinism. An application that has to do with modern engineering and how we design systems that are safe and more importantly understandable to human users.
Determinism as a property of a system can be thought of as reliability or predictability. A working concept of this type of determinism is the black box approach. The black box approach essentially implies that it is not necessary to know how a system (let's say, a website) operates: the only requirement to the system is that given the same input (press the same button on the keyboard) one always gets the same output (my post gets published in my blog). The black box approach is thus very helpful; it lets salespeople sell clear features and users to interact with technology with clear expectations. Everybody likes the black box!
Expect when they don't like the black box... What happens when we have a rare deviation? Is the website still deterministic if I click the button once and the post doesn't get published? How often should this be accepted to happen? If we rely on the system working properly, should we perhaps know a bit more of how the system actually works?
An example where the black box approach to determinism doesn't work is Artificial Intelligence applications and especially those built on neural networks. Neural networks are by nature probabilistic: they are trained to associate specific inputs to specific outputs, always with certain probability. It is thus very hard to argue that your self-driving car will always stop at a red light! Consequently, to prove that a self-driving car is safe we have to resort to extensive testing, with hours and hours of real life driving tests.
Alternatively, we could instead consider a different conception of determinism - one borrowed from the complex world of computer science and logic. Determinism is when the current state of the system and the input is known, then there is only one state that the system can move to. Suddenly, our self-driving system is simultaneously deterministic and non-deterministic depending on the agreed definitions. Definitely a confusing world - no wonder self-driving cars are still not a commercial reality!
Comments
Post a Comment