What would you like to know about DataRobot and the future of Artificial Intelligence (AI) and Machine Learning (ML)? In this inaugural Ask the Expert event, you will be able to chat with Phil and ask questions relating to DataRobot's role in the future of AI and ML. While these topics are complicated, Phil is available to help clarify and answer your questions.
|Phil Gurbacki is the SVP of Product and Customer Experience at DataRobot. He has a passion for building and bringing new machine learning and artificial intelligence software products to market. Phil is very adept at leading the overall product development process from concept to launch.|
This Ask the Expert event is now closed and we will be providing a summary of all the excellent questions and responses.
Thank you Phil for being a terrific event host!
Let us know your feedback on this event, suggestions for future events, and look for our next Ask the Expert event coming soon.
@MFerraro7 DataRobot works with many of the large Global System Integrators and Consulting firms, to enable their teams with automation-first AI and advanced machine learning, and to help our mutual clients achieve transformative value. Please private message me if you have a specific interest in learning more.
@george1986, Great question! It’s true...governments, think-tanks, and even companies have proposed standards helping society cope with AI explainability and trust. We have multiple engineers and data scientists solely devoted to developing trust specific features in our platform. DataRobot embeds trust throughout the modeling workflow from data warnings like target leakage, model-specific interpretations such as feature importance and even models in production with automated compliance documentation. In terms of the regulatory and compliance aspect, DataRobot continues to be involved by drafting principles, providing guidance and informing public officials so AI can be widely adopted while balancing societal protections.
@ErnieZ, Capabilities from the Paxata product will be increasingly tightly integrated with the DataRobot platform and augmented with new feature development for DataRobot’s existing data preparation functionality, with the ultimate objective of delivering on our vision of a unified, end-to-end AI platform.
Hi Phil - first time caller. Thanks for taking my question.
Imagine for a moment that we used machine learning to determine what gift to get you for your birthday that may or may not be today, the model is saying you’re a gift card guy, but I took you for more of a nice bottle of wine guy. Can you validate our findings?
Also, would love to hear your thoughts on your vision and the investments we’ve been making around Customer Experience as a company?
Hi Phil - So great that we get to ask you questions! One of the hardest tasks ML tasks we've had to date is getting enough data to build good ML models. (It's just the nature of our industry and regulations and not likely to change in the near future.)
Do you see a time when ML models may become even better than today with even fewer data requirements?
Looking forward to hearing your thoughts. thankyou.
Phil, We build lots of models, and the few that get deployed never seem to be updated. I’m excited about the potential for MLOps to help identify when models need to be replaced – what’s your vision for MLOps?
@BrianOblinger, thanks for the question Brian. You are spot on, I accept gift cards to any amazing restaurant and will certainly use it for a good bottle.
In terms of Customer Experience, I think the first and most important step of the process is to learn and understand what your customers are going through across their entire journey. Tracking this information allows you to gain insights in to the problems you need to correct and the opportunities you can take advantage of. The second thing is to really take care to understand which personas are engaging with you and to make sure you have tailored experiences for those personas that really resonate with them.
@sallyS, it's certainly true that getting data to build models can be difficult but I think people spend way too much time on this step of the process. There are many algorithms that will do a great job on small sets of data. I recommend an agile approach to data science where you try to build models with the data you have and see what the results are. Then add more data, build models, and see how that improves your baseline performance.
@c_stauder, access to data and data silos. Healthcare data by nature needs to be protected and the controls around that data stifle AI innovation. The other problem I see is the lack of central repository or wholistic view of the data. Many times the data is siloed inside of departments or organizations and rarely tells the entire story.
@rick-wheller, I view MLOps as a critical capability for anyone that is serious about running their business on machine learning models. The capability needs be able to give confidence & trust in machine learning to the executives in an organization. My view on the market is it's really unacceptable and irresponsible for models to used without performance tracking, without monitoring of the underlying machine learning, and without being frequently refreshed.
Lots of good questions on this thread. So can you share your thoughts on how machine learning models might manage themselves after deployment? For example could they monitor themselves for drift, etc., then rebuild and redeploy as needed? Might this be a future possibility?
@krista-kelly, my view is that you have to think about the business problem and what it means for a model to solve that problem. For example, if you are forecasting sales for this month then you wouldn't want last year's model to do that forecast. There needs to be a model created based on the most up to date information available that is used for this forecast. The world we are moving toward is one where models will be upgraded when new information becomes available and the old model does not provide the performance required. This has been happening in the predictive maintenance space for a long time and I think you will start to see predictive maintenance for machine learning models becoming more mainstream in the near future.
@DaveTheMaster, Every user of DataRobot needs to prepare their data before building models. We are starting out with an extremely tight integration of Paxata and DataRobot. Our ultimate objective is to have a native prep capability inside of DataRobot focusing on Data Prep for AI and delivering on our vision of a unified, end-to-end AI platform.