IBM Innovation Preview: Automation of AI
IBM’s second Innovation Preview event features AI experts and IBM executives as they share exciting new developments made by IBM Research in the field of Automation of AI.
In the event, now available online, these leaders present three ways to leverage automation for AI:
- Accurately digitize and streamline workflows to become more efficient and adaptable
- Reveal actionable insights with AI intelligence applied to automated workflows
- Give AI tools to your human workforce for better, faster options and increased productivity
Rob Thomas, Senior Vice President, IBM Data and Cloud Platform, gives an overview of IBM’s viewpoint on automating AI and explains how the technology can make work easier without sacrificing results.
Then Sriram Raghavan, Vice President, IBM Research AI, unveils the taxonomy of what IBM Research is doing in the space of automation of AI, and how IBM assists clients in building a better data foundation from which to launch transformative AI projects.
Finally, to illustrate how research makes its way into IBM’s portfolio of AI products, Dr. Lisa Amini, Director, IBM Research, and Shadi Copty, Vice President, IBM Data and AI, present a demo of AutoAI in IBM Watson Studio. With AutoAI, advertising giant Wunderman Thompson identifies new prospects and gain deeper human insight at scale.
The notion of automating work has been with us for as long as we’ve had work to do, says Thomas. It’s visible in the assembly lines of the Industrial Revolution, and as far back as ancient Greece and the automatic workers who assisted the Greek god Hephaestus in the pages of the Iliad.
Today, says Thomas, automation will allow patients to be processed seamlessly through the healthcare industry. It also gives site reliability engineers the ability to understand IT problems as they arise or predict critical issues before they occur.
The gift of automation is the gift of time. That time can be used to focus on more critical tasks, more cognitively creative assignments, or to take time away from work, knowing that the job will not suffer. The first step to success in infusing AI into your enterprise begins and ends with data and how it’s being used. IBM provides three critical ingredients for this success:
- Natural Language Processing
IBM’s agenda is to accelerate a business’s journey to AI by using cutting-edge technology to create, deploy, manage, monitor and evolve AI models. Businesses need fast data preparation, accurate AI models that conform to business constraints, governance of models across their lifecycle, and automated monitoring and improvement. This holistic approach breaks down into three pieces:
- Data Automation: Automatically selecting, ingesting and cleaning data to enable further automation
- Data Science Automation: Automating the arduous process of data science, allowing automation of engineering and even modeling
- Lifecycle automation: Testing and validation, deployment, monitoring and connecting the loop, so the model continues to perform and evolve
IBM Research has also added new capabilities within IBM AutoAI, which now includes the ability to automatically build models specific to a variety of constraints — anything from industry differentiators to fairness and bias. Research is updating model drift detection with a robust suite of algorithms, better scalability, and recommended next-best-actions for users to strengthen their models.
IBM Research’s comprehensive agenda provides a steady pipeline of innovation into the portfolio of products that live on the Cloud Pak for Data platform, including Watson Knowledge Catalog, Watson Studio, Watson Machine Learning, and Watson OpenScale.
In a demo, Dr. Amini explains how to use AutoAI in OpenScale to solve a problem relating to bias in your system. Usually, a data scientist would need to comb through the algorithm and decide what tools to use and how to appropriately assess the fairness metric, which could take weeks or even months.
IBM Research ensures that everything is transparent and visually oriented so users can understand issues at a glance. Users can save their in-progress model as an easy-to-read document detailing everything about the model for documentation or to make further experimentation easier.