IBM Think Digital 2020: Build and scale AI with trust and transparency

There was a session titled 'Build and Scale AI with Trust and Transparency with an Automated AI Lifecycle' at the IBM Think Digital 2020.

Shadi Copty, VP, Offering Management, for Cloud Pak for Data, DataOps & Watson Tools/Runtime, IBM, said that an AI automated lifestyle can help you achieve the AI at scale for your business. Enterprises are accelerating their journeys towards AI. At least, 21% enterprises are making AI their largest tech investment. 57% have built AI skilled teams of greater than 50. Also, there is 2.5X greater investment on the AI lifecycle platforms.

Today, we are witnessing the use of AI in the post-Covid-19 world. The application of AI will be immensely valuable in helping companies adapt to the trend. Companies will be pressured into accelerating these activities, as per BCG Henderson Institute. Consumers coming out of the crisis will expect even better digital experiences. and more hyper-personalized interactions.

Companies will also rethink their supply chains to include redundancy, while still keeping costs lower. The only path to get there is through more intelligence. The post-Covid-19 enterprise will use AI to produce cheaper, better and faster decisions. They will expect these decisions to be safer at the same time. Companies will delegate more decisions to machines that are learning from data, with humans supervising these machines.

Value from AI

Most of the value from AI will come from leveraging valuable proprietary data to design their specific growth path. AI pioneers are more likely to build in-house AI capabilities and retrain employees. We have come to a prescriptive approach to accelerating the journey to AI at scale. The key word is scale. You need to be able to infuse AI into your business processes. You need to build and scale AI with trust and transparency. You have to create a business-ready analytics foundation. Finally, you have to collect and make the data simple and accessible. You will modernize and make your data ready for an AI and hybrid world.

An automated AI lifecycle and open source software have helped overcome the adoption challenges to drive outcomes. There are three factors: data, talent and trust. There are two critical ingredients in dealing with these challenges. First, automating the AI lifecycle, and extending it to the trust, and the management of AI, post deployment. Second, the reliance on open source frameworks that expand the pool of talent and innovation available to your AI builders.

Using IBM's platform and approach, data scientists at ExxonMobil were able to achieve 40% savings in time to prepare data in building new AI models. That means, this team is 2-1/2 times more faster in building more models. Data prep for modelling contributes a significant portion of the entire workflow. Data scientists at NYLINE were able to achieve a 10X jump in training productivity for the AI pipelines that read and interpret visual information.

KPMG was able to build trust in the AI outcomes by leveraging the ability to understand the biases and monitor models when approving a loan or a line of credit. An end-to-end lifecycle that leverages automation in all the steps of AI construction and deployment matters when it comes to scaling up these activities.

IBM's offerings are uniquely designed to provide clients a modeler set of tools for AI model creation and lifecycle management. This includes the ability to prepare and innovate for data discovery and activation. The Watson Knowledge catalog allows users to access, curate and categorize, and share data assets, wherever they are. To help simplify the AI lifecycle management, IBM offers the AutoAI as a capability within Watson Studio. IBM has the Watson OpenScale.

Role of automation

In performing feature engineering, where we rely on codified knowledge from a few of our Kaggle masters, powering up your data scientists to achieve the best, in tuning models. They will be optimizing for model accuracy where we can automatically detect triggers for dynamic model retraining.

Reliance on the open source frameworks is the other critical component for success. Open source helps you to simplify the multi-cloud data complexity, close the skills gap, and embed trust. You can build AI pipelines on top of the Red Hat OpenShift, making our platform available anywhere. There are seven data scientists with skill in Python and R. About 87% of the new data scientists have a clear preference for learning open source frameworks first.

Anaconda and IBM

Copty introduced Peter Wang, CEO, Anaconda Inc. Wang is also the founder of PyData community, and the initial creator of the of Bokeh and Datashader plotting libraries. Wang added that he is super excited to announce a strategic partnership between Anaconda and IBM.

Anaconda has been a leader in the data science movement for the past 10 years. Anaconda is a preferred software tool for learning data sciences. Wang noted that data scientists complain that their organizations don't understand the exploding use of open source in business analytical environments. I am very excited to get into this partnership with IBM. We can make sure that when open source arrives, it does not suffer.from enterprise-immune rejection.

Copty added that IBM is bringing the commercial version of Anaconda, which will integrate with the Watson Studio and Cloud Pak for Data platform. For customers, the innovation from open source is tested, secure, and governed.

Wang concluded that this partnership with IBM will be really great for the open source data science software community. The software is now powering the AI and ML revolution. Their challenge: how to build the open source software into business environments. This partnership will take the open source data science community from being just a movement into becoming established as an enduring transformation of business computing.