Abstract: 

To sleep - perchance to dream: ay, there's the rub,
For in that sleep of death what dreams may come
-- Hamlet, Act 3, Scene 1

Powered by deep learning, we've taken amazing strides toward the ability of a machine to mimic human expression. In the case of text, if we provide enough training examples an LSTM can learn grammar, punctuation, sentence structure, and generate an essay that might be graded as a C. But it's a long way from freshman composition to The Divine Comedy. The missing ingredient is cohesive ideas that combine to reveal deep, human insights that people will be dissecting, studying and appreciating 500 years from now.

Androids may dream of electric sheep, but their attempts at artistic endevours (i.e. literatre, music, painting, etc.) reveal a distinct lack of soul. If we extrapolate this idea back to more tangible machine-learning problems, what can we learn about the limitations of the solutions that we build every day? It would appear that the popular assertion that AI is making humans obsolete is premature.

For scenarios like fraud, churn, recommendations, anomaly detection, tools like Auto-ML make it fast and easy to learn as much as you can from the examples you can provide to the algorithm. But these solutions look to the past, not the future. Semi-supervised methods and coaching can help, but that is like fixing an essay’s grammar – technically more correct but lacking innovation. How can a data scientist get real insights into a model?

During this session, live examples of machines attempting to emulate human expression (literature, music, art, etc.) will be shown using current, cutting-edge open-source models with the code repository provided. We will also review real-life examples of impacts to model performance. Maybe we will discover a machine with a soul, or at least a “one hit wonder”. Either way, attendees should come away inspired to get more of their soul into their models...

Bio: 

Joe Blue is a Customer-Facing Data Scientist at DataRobot and has with 20 years of hands-on experience including concentrations in financial services & healthcare. Prior to DataRobot, he built solutions for United Healthcare, HNC Software, ID Analytics and MapR. Currently helping to democratize data science with Automated Machine Learning - one organization, indivisible, with insights and value for all.