Salesforce Chief Scientist Richard Socher Debunks the Biggest AI Myth
Previously CEO and founder of MetaMind, a startup acquired by Salesforce, Richard Socher now leads the Salesforce’s research efforts and works on bringing state of the art artificial intelligence solutions to Salesforce. We spoke with Richard about his top priorities in advance of AI By The Bay and asked him to debunk the main myth about AI.
What is the most exciting project you are working on right now as a Chief Scientist at Salesforce?
At Salesforce I wear two hats, one as Chief Scientist leading AI research and another in applied AI where I help bring state of the art aI technology to our customers with Salesforce Einstein. I’ll choose an exciting project from each.
On the AI research front, it’d have to be the “joint multi-task” learning model we recently introduced, a single deep neural network model which can learn five different NLP tasks and achieve state-of-the-art results. Our model starts from basic tasks and gradually moves to more complex tasks, which is continuously repeated until our model finishes learning all of the tasks. By doing this, our model allows the tasks to interact each other and improves accuracy.
In terms of applied AI, with Einstein AI features and capability are embedded directly into the entire Salesforce platform and empower everyone– from engineers to customers– to leverage AI to be smarter, more productive and predictive. By doing the heavy lifting and removing the complexity of AI we are able to deliver seamless and scalable AI to Salesforce customers of all sizes.
What are the key challenges for companies that are getting onboard with AI and machine learning?
Data is crucial for AI to work and most struggle with accessing historical data needed to build predictive models. Wrangling this data is expensive both in terms of time and resources and technical expertise is required to properly label outcomes for the AI model to predict. For most companies, the technical complexity and resources is just too much.
What are your priorities in terms of AI and machine learning at Salesforce going forward?
I’m focused on creating the world’s smartest CRM with Salesforce Einstein by advancing the science behind deep learning. This enables the creation of new applications of this technology.
The other priority is creating applications that thoroughly understand language, image and structured data in order to help businesses get answers to any question they have about their data and make more accurate and personalized predictions.
What is the biggest AI myth you would like to debunk?
There are so many floating around today, but I’d say the biggest myth is that AI will become independent, self-aware and take over. We are currently nowhere near anything resembling such a system, and these myths can be distracting from the actual, more pressing issues that could arise.
How can developers, data scientists, and software engineers engage with Salesforce?
Our platforms like Force.com and Heroku make it easy to build add-on apps that integrate into Salesforce’s main applications. We’re also continuing to add new services on top of these platforms that will empower anyone to build AI-powered apps.
You’ve spoken at Text By the Bay in 2015 and Data By the Bay 2016 on Deep Learning with NLP and computer vision. Where are we headed in 2017?
I think it’s time we build a larger, more versatile language and vision system that can do more than one thing, a.k.a multi-task learning.
Right now no one is able to train a single model to do many different kinds of things as once. For example, the question, “Who are my customers?” presents a simple listing task. “Who are my best customers?” adds a complex layer that requires numerous integrated tasks to answer many qualifying questions and make some hard decisions. With questions like the second one in mind, the research team at Salesforce recently created a “joint multi-task” learning model that successively trains with basic tasks, gradually moves to more-complex tasks, remembers all the tasks, and allows them to interact with each other. I’m really excited about it and I think we’ll see more on this front in 2017.
See Richard Socher at AI By The Bay on March 6-8 at The Pearl in San Francisco.








