Salesforce.com Inc. has found a new application for artificial intelligence.

Researchers from the cloud giant on Tuesday released a paper detailing the architectural details of Seq2SQL, an internally-developed tool designed to help business users interact with databases. The software provides a natural-language interface for asking questions about the records in a system. According to Salesforce, each query is automatically translated into the SQL language typically employed to query databases.

The idea is to spare workers the hassle of learning the syntax before they can dig into their data. This is facilitated by a specialized AI that implements reinforcement learning, a method of improving model accuracy over time.

The technique involves creating the AI equivalent of a reward system. When the system produces a favorable result, it receives positive reinforcement and can thus learn what calculations are most effective. The process is used to enhance different aspects of the analysis until an acceptable level of accuracy is reaches.

Seq2SQL applies this method to tackle the interpreting mistakes that can occur when converting user requests into database-speak. A question about the number of customers that a company has in London, for example, could potentially be understood to refer to London, Ohio. Seq2SQL learns from common database interactions to try to avoid such errors.

Salesforce trained its internal implementation of the system using a custom dataset that was released alongside the paper. Dubbed WikiSQL, it consists of information from Wikipedia that has been structured into tables and manually enriched for development purposes.

Companies could use Seq2SQL with the sample data to build custom natural-language search features for their business systems. It’s the latest addition to a growing list of projects that employ machine learning to streamline information access. 

Another is Learning-to-Rank, an AI plugin that Bloomberg LP released for the open-source Solr search engine earlier this year to improve the relevance of results. More recently, Elasticsearch Global BV added machine learning capabilities to its rivaling platform. The latter offering can be used to find points of interest in infrastructure data and business records.

Salesforce has been working to embed AI, through its Einstein technology (pictured), in all its services. It unveiled Einstein last year at its Dreamforce conference in San Francisco.

Image: Salesforce