Salesforce has built artificial intelligence into more and more of its tools since it introduced its Einstein Platform a few years ago.
For instance, the CRM giant announced new tools to customize voice assistants at this year’s Dreamforce conference and to incorporate AI into contact centers. These capabilities demonstrate how Salesforce, the world’s number one customer relationship management company, is progressively making things easier for its trusted and esteemed customers – albeit in incremental steps.
The Salesforce Research team showcased some of its breakthroughs in areas such as natural language generation and conversational artificial intelligence to offer a forward-looking glimpse into the product capabilities of Salesforce at the Dreamforce 2019 event. Salesforce Chief Scientist Dr. Richard Socher remarked their research is focused on creating an AI-driven world that is so far only found in sci-fi. Socher added Salesforce is emphasizing on future predictions and the best way to predict is to create.
The Research Group division of Salesforce includes four sub-groups: new product incubation, applied research, fundamental research, and AI platform research. Socher added this is a change from what used to happen until just a few years ago when Salesforce was primarily focused on fundamental AI research.
Salesforce Senior Research Scientist Victoria Lin remarked lots of enterprise data exists in relational databases and deriving insights from the data usually requires users to know a query language such as SQL. Alternatively, they would require the expertise of analysts to run a report for them. Lin added this is a big step towards democratization and a possible move towards the start of data conversion and the end of data entry.
Salesforce Senior Research Scientist Nitish Keskar unveiled Conditional Transformer Language (CTRL) model that has the potential to generate text conditioned on control codes that highlight style, domain, dates, entities, topics, plot points, task-related behavior, and relationships between entities.
This open-source language model has been trained on 143 GB of text, including thousands of books, millions of documents, and all of Wikipedia besides having an unprecedented 1.63 billion parameters.
Salesforce Einstein VP of Products Marco Casalaina stressed that the world’s number 1 CRM giant does not make use of customer data for training or building global AI models. Salesforce built an end-to-end automated machine learning library for structured data – TransmogrifAI – for assisting customers to build customized machine learning models. Casalaina added the Einstein platform now powers as much as 10 billion predictions a day.