What it is: Neural networks is an artificial intelligence technique to model how the brain works in software.
Artificial intelligence (AI) has been promised since the 1960’s, but it’s finally here. For the longest time, AI has been touted as the next big thing,b ut each time it’s proven disappointing. Early attempts at AI involved something called expert systems, which was nothing more than a series of if-then rules.
The idea was to take an expert’s knowledge and translate them into multiple if-then rules. This required the use of a knowledge engineer who would act as an intermediary between the expert and the programmer. Once the expert system was designed, it would theoretically encapsulate the expert’s knowledge so others could query the expert system without bothering the real expert.
The huge problem with expert systems was that they took too long to create. With a knowledge engineer working with a programmer and an expert, creating an expert system was a time-consuming affair. Even worse, once the expert system was done, it couldn’t learn anything new without someone manually modifying it.
Since knowledge in any field changes rapidly, this meant the expert system would become increasingly obsolete over time. Eventually it would get so antiquated that it would be essentially useless. After all, how many people would want to take advice from an expert system containing medical information from the 1950’s?
Because expert systems took so long and couldn’t be modified, they were largely ignored as too troublesome to create and too cumbersome to constantly update. That’s why AI remained absent from the commercial world for decades.
Neural networks are a different approach to mimicking artificial intelligence. The goal with neural networks is that they train themselves. Feed data to a neural network and it eventually learns from that data. This makes neural networks easy to create and easy to modify.
Credit card companies use neural networks to monitor fraud. By studying normal buying behavior, a neural network can detect abnormal purchases from someone’s daily routine. Then the credit card company can let a human investigate further.
Creating a neural network involves mathematical calculations, but Apple has now created a new software framework called basic neural network subroutines (BNNS), which is a collection of functions to implement and run a neural network. By creating this framework, you can easily create your own neural network without worrying about the details of calculations. This makes it easy to add neural networks to any macOS, iOS, tvOS, or watchOS apps you might create using Objective-C or Swift.
The easier it is to create a neural network, the more likely developers will create them in their apps. That means more neural networks and more AI. By making it easy to create your own neural network, Apple has helped make AI more accessible for developers.
This doesn’t mean we’ll necessarily have smarter devices, but it does mean we’ll have the power of neural networks in more apps. A long time ago, artificial intelligence could barely play a decent game of chess. Today, artificial intelligence and neural networks are picking stocks, checking for credit card fraud, looking for terrorists, and approving bank loans.
Artificial intelligence is here to stay.