Skip to main content

Leverage IBM Db2 for i in Building AI Models

Doug Mack and Gary Goldberg explain how to leverage your current database.

image of blue and purple tiles scattered and then organized.

The IT world is abuzz about the use of AI and machine learning to gain insights into data. With that excitement comes many questions to understand how to apply it. IBM i clients, who often have the majority of their data in IBM Db2® for i, are interested in how to leverage that data with these technologies to transform their business.

We’ve been seeing AI in science fiction for decades, from the murderous computer HAL in “2001: A Space Odyssey” to the bumbling robot maid in the Jetsons and the clueless “Robot” in Lost in Space.  Now we see it in our homes and our cars every day with voice activated products like Alexa, Google Assistant and Siri. We're seeing the beginning of intelligent computers with the emergence of preemptive braking features on the newest automobiles. Machine learning is one of the main ways that computers can be taught to do things on their own. This is accomplished through training the computer by applying a statistical model against a vast number of data samples and comparing the outcomes of that model against real-world results. This training allows the computer to learn to make better predictions of the outcome of the next sample of data it sees. 

AI at the Epicenter

Today, you most likely need a data scientist to use AI and machine learning. That’s because the first thing you need to do is build a statistical model and then create an application that uses that model to score your data against a large set of known examples. That effort is done using complex development processes and requires a knowledge of statistics, as well as tools like Watson Studio, R or Python (now both running natively on IBM i) or some other statistical technology. 

Next, you must establish a data set to train your model and subsequently validate that the model is providing the intended result. This often requires preparing, cleansing and formatting your data to feed into the training of the model. The model can then be applied against larger prepared data sets that you want to analyze to get the insight that you’re looking for. Finally, you need to create visual metrics from the outcome that can show the insight in a way that’s easily understood by the business user. 

If the future of AI is to be a part of everything we do, then that future has to provide the ability to include and use preexisting models that the business analyst can leverage without the need for a data scientist. In practice, it requires the ability to include the statistical models, which will always be the province of the data scientist, into the more broadly used workflow of a business analyst or end user.

2 Approaches to Models

Let’s describe, in general terms, two approaches to include preexisting models in that workflow. 

The first approach is to integrate sets of predefined models into the workflow. This means that the product you use to analyze your data needs to have pre-built machine learning functions that users can access as easily as they can generate an average or a total for any given data set. In many cases, users may want to apply multiple models to their data so they can choose the one that’s best at predicting the outcome for the data they’re analyzing. This approach is the closest to the idea of bringing AI and machine learning to the business analyst and only requires them to know what kind of analysis they need to apply. 

The second approach is a bit more technical, but also more flexible as it doesn’t require the tool to already contain all of the models. The process here is the ability to drive pre-built models from anywhere into the analyst workflow. This is done through a connector that interrogates the model and understands what data it needs to make a prediction the format of the prediction that it returns. The connector then automates the integration of that model into the analytic tool that the business analyst uses for his everyday work.

AI Tools Integration

As AI becomes a needed element in day-to-day data analysis, tools vendors will be required to add these kinds of capabilities into their product, including easy-to-use data preparation tools and the ability to integrate with AI and machine learning models. In addition, a truly modern business analytics tool will need to make these kinds of functions easy to consume by business users, as it’s unlikely that we can expect every analyst to become a data scientist any more than we can expect every end user to code their own programs. The methods that tools use and features they offer may be based on the approaches outlined in this article or, likely, on new approaches that will be found in the future.

PoV_Figure1-(1).jpg

IBM Systems Webinar Icon

View upcoming and on-demand (IBM Z, IBM i, AIX, Power Systems) webinars.
Register now →