CL Simplex

Artificial Intelligence: Practicality

Artificial Intelligence: Practicality

We discussed that artificial intelligence is fundamentally a computer using rules to search a graph to find the best result given constraints. These constraints are things like legal moves in chess, the work hours in a week (if you’re building a scheduling tool, for example), government regulations, and other facts of life we use computers to represent when attempting to use computational rationality to solve problems.

Some problems are more practical to solve than others. Interestingly enough, what was originally thought to be hard problems are easier to solve using computers than expected. It is much easier for a machine to detect credit card fraud than it is to have a machine stack cans in a supermarket. The most difficult problems are ones that humans find the most intuitive, but perhaps that should be unsurprising. Using our stack of cans example, machines can only deal with the information given to them. A human can make additional judgements like putting cans of the same variety together, not blocking an aisle with a stack of cans, etc.

Unlike that stack of cans, the infusion of massive data sets has given previously academic techniques commercial application (like detecting credit fraud.) In the past, language translation software was built by creating complex rules based on grammars and expertise. Language is a slippery thing - it evolves, there is slang, misspellings, new vocabulary, and compositions. Turns out - when you throw all those assumptions out the window and merely use the data of how people use the language (with the web this data is quite substantial) - software performs pretty well! Things like slang, popular misspellings, and new grammar and vocabulary are captured in the patterns found in the data sets.

This data revolution is the real driver behind the recent explosion of “Artificial Intelligence.” Better data techniques and an ocean of information have made it possible to let the trends and patterns in the data do the talking. Metadata collection has opened up powerful insights even when the data itself isn’t directly accessible. Datasets and data processing is reaching the point where the results are beyond understanding the processes used to reach the conclusion (for example - if a computer calculates 10 million digits of Pi, do you trust it? You can verify, but it will take a long time - is it worth it?) Our next part in the series is how humans can understand AI, and how it can be accessed to make life better.


Tap or click on these posts to navigate to the next or previous posts.

Post Series

This post is part of a larger series. Tap or click on a post to view more in this series.