Here's how much smarter Google search results have become using artificial intelligence

Google has a special way to improve the quality of search results: artificial intelligence (AI.) The company uses a type of AI known as machine learning to figure out what its users really want to search for.

Machine learning is where a computer gradually teaches itself how to perform a task. One example of that is Google DeepMind, which learned how to play retro arcade games over time.

Google's machine learning system for search is called RankBrain, and it tries to figure out what a user is searching for.

Digital marketing agency Stone Temple Consulting took archives of Google searches before it introduced RankBrain and compared them to present-day searches. The difference between the old, pre-machine learning search results and the new ones are interesting.

Stone Temple Consulting used a non-obvious search to test RankBrain. It searched for "why are pdfs so weak," a phrase that seems to be missing a word or, at least, makes no sense unless you can guess what the searcher has omitted from the word-string:

The search on the left is an older search from before RankBrain was introduced. And the image on the right is a more recent search. There's a big difference between the two: The search on the right seems to have understood what the user has typed, and is showing information that answers the question in a meaningful way rather than just matching for those keywords.

Stone Temple Consulting says that 54.6% of searches conducted using RankBrain were improved since late last year. It also says that RankBrain may have caused Google to include more special elements in search results.

Special elements include answer boxes and dynamic charts (above) as well as maps and other things that save users from clicking through to web pages.