Creating value and competitive advantage are two top use cases for Artificial Intelligence (AI). Now, it’s time for businesses to really get their hands on how to put it to work to those ends.
Most companies are ready and willing to take it on. In the case of large enterprises, they’re looking for ways to disrupt the disruptors – the smaller, nimble startups that harness the Cloud’s computing power to quickly innovate, taking a bite out of their markets. “Big companies know they need AI to get there,” said Jay Limburn, IBM Distinguished Engineer and Director of Product Management.
End-to-End Artificial Intelligence
IBM’s proposition is to start the ball rolling with its recently announced Watson Studio and the Watson Knowledge Catalog. The IBM Cloud Services support end-to-end AI workflows. Studio is an integrated environment for connecting to and refining data to build and train Artificial Intelligence models at scale; these can be continuously trained against new data as they infuse business processes. However, “data cataloging is the first step to become effective with AI,” said Limburn.
The Knowledge Catalog enters the picture here as a rich, intelligent Metadata Index that provides a view of and easy accessibility by knowledge workers to structured and unstructured data and Analytics assets that have been profiled and classified – no matter where they reside. Watson’s Natural Language Processing supports getting intelligence from structured and unstructured data to use in a uniform way.
Its AI-enabled search and recommendations based on Watson’s understanding of relationships between assets help users discover relevant catalog assets for their modeling needs. “We use AI to improve your AI,” he said – think of it as a ‘Spotify for data’ recommendation model. “This model self-trains over time, so the more you use it the more exhaustive it is,” he remarked. Last but not least, with a data policy activation engine, data can be opened up for self-service use by defining rules that explain how that data may be used and by whom.
Additionally, artifacts created on AI journeys can be shared back into the Knowledge Catalog and reused, Limburn explained, so that the efforts that went into preparing data for a Data Science project don’t go to waste.
The technologies give Data Governance a whole new purpose, he said –it’s not just for compliance but for driving innovation and disruption, too. One client using the Knowledge Catalog to index information from some 200 databases, which came with the acquisition of another company, found that the Knowledge Catalog discovered additional unexpected data sources relating to a retired product and then put them into shape, too. That information became part of a Data Science initiative exploring customer adoption of older and newer products.
“They built adoption models based on a whole wealth of information they didn’t know they had,” said Limburn. “Data Scientists didn’t have to fish around for the data. In fact, they wouldn’t even have known those data sources existed to build predictive models that can be used across the company.”
The link companies want to make between Artificial Intelligence and value is growing stronger, confirms Christoph Kögler, head of innovation at T-Systems Multimedia Solutions GmbH. At the Enterprise Data World 2018 Conference he discussed how companies have moved from getting a handle on what AI is – a way for machines to assess information, make sense of it and respond to it – to what AI can do for them. “It’s down to the business now,” he said.
And what the business wants to hear less about are the Sci-Fi scenarios of machines taking over mankind and more about the practical ways that Artificial Intelligence can be applied to deliver a significant return on investment. It’s true that many AI-based projects still stall out post-pilot, or close down because costs are too high relative to benefits, he said. But the good news is that there are examples today where AI efforts are creating real value that others can learn from.
As one of Germany’s largest systems integrators, T-Systems Multimedia Solutions has participated in student research, government and customer AI efforts, some of them still experimental. Even so, they help increase clarity around the real-world potential of AI. The industrial manufacturing environment examples include:
- AI models are being in these environments are being tried out to assess, process and respond in real time to situations where robots putting chips inside mobile phones on a sample production line make an error in positioning components. That should be a cheaper and more efficient approach than the traditional practice of using after-the-fact quality measures based on taking pictures at the end of the process. With Machine Learning having assessed data points related to the production process, it was possible to create AI models to detect correct or incorrect positioning of chips and send out alerts when a problem occurs.
- In another case, Machine Learning algorithms are behind an AI scenario where manufacturers in Germany can control factory operations handled by their partners in China. Having gathered enough data to build a model to detect operational issues, machines on the production line can send out alerts to a monitoring dashboard display. Via a speech interface, a manager in Germany can directly question the system to identify the problem and enables the manager to ask the system to infer the cause based on what it has learned previously and then propose a solution to solve it.
The data challenge to overcome for Artificial Intelligence to meet expectations, he explained, will revolve around going from sourcing data to engaging on business value. Optimizing data for building AI models, as Limburn also pointed out, will be key to get meaningful response information. That encompasses enriching data by including information from other sources for more context and trying to take the messiness out of sensor data, Kögler said.
As for today, though, he noted that it’s wise to never be sure that everything will work the way it’s expected to, and simulations are critical to get off on the right foot as best as possible. “Today it’s like the Wright Brothers’ first airplanes, and there’s a lot to do – like landing on the moon – before we get to the point where we can put a Tesla in outer space,” Kögler said by way of analogy.
Finally, he cautioned listeners to remember that “good AI starts with humans and business processes and even culture changes: “So center on people to get business value.”
Photo Credit: agsandrew/Shutterstock.com