RankBrain uses machine learning to filter results and improve the way Web pages are positioned in search engine results pages (SERP). Machine learning software gathers information that is not specifically provided to it. Like data mining, it searches through data to find patterns and uses that data to improve the software's own understanding.
The RankBrain technology bases its interpretation of ambiguous search queries -- and especially complex, multi-word queries -- on knowledge about similar searches it has experienced. According to Google, their technology recognizes like patterns among apparently dissimilar complex queries and shorter, more easily understood queries. Those patterns help RankBrain identify how those queries are actually similar. That information can help the search engine algorithm interpret long and complex search queries in the future and identify topics they might relate to.
See a video introduction to RankBrain: