Aera Technology this week announced it has integrated the Microsoft Azure Digital Twins service with its Cognitive Operating System platform that employs crawling technology to provide artificial intelligence (AI) models with data in real time.
The integration with the Microsoft Azure Digital Twins service will provide organizations that employ the Cognitive Operating System platform with access to a source of data that many organizations are now employing to mirror physical environments, says Aera Technology CEO Frederic Laluyaux.
Rather than having to collect data from, for example, plant equipment, it’s simpler for that data to be collected from the digital twin that is synchronized with that environment, notes Laluyaux. That data can then be fed into an AI model alongside other data to enable organizations to make better business decisions faster, adds Laluyaux.
The Cognitive Operating Systems platform enables data scientists to employ low-code tools to construct AI models to automate a wide range of business processes using a set of business rules they can define and adjust as requirements change. That platform over time will surface recommendations to optimize those processes as it continues to receive guidance from end users.
Longer term, those recommendations will also incorporate the impact any decision might have on the profit and loss statement of an organization, adds Laluyaux.
“We’re moving from a world where people are guided by machines to one where machines will be guided by people,” says Laluyaux.
The main challenge in achieving that goal now has less to do with the technology itself than it does culture, notes Laluyaux. An AI model is not much different from a new employee that needs to be trained, he says. Decisions made by the new employee need to be reviewed until the organization gains confidence in the expertise of that new employee, said Laluyaux.
In the case of the Cognitive Operating Systems platform that review process is facilitated by providing full transparency into how the AI model was constructed to enable explainability, he added.
Also read: AIOps Trends & Benefits for 2021
Data Management Concerns
Ultimately, an AI model is only going to be as reliable as the data that was employed to train it. Most organizations, however, are discovering the quality of the data they have is suspect. Many data scientists today are still spending an inordinate amount of time on cleaning up the data before they even begin to train an AI model. The rate at which AI models are being deployed in production environments as a consequence is relatively slow given all the time and effort being made. It’s not uncommon for many AI models to never make it into a production environment because during the testing phase it becomes apparent there was an issue with the data used to train the AI model.
However, as more organizations become proficient in best practices for both data science and engineering, it’s now only a matter of time before thousands of AI models are deployed across an enterprise IT environment. The next big challenge will be finding a way to manage and retrain all of those AI models as the assumptions that were made when they were deployed prove to be less accurate as circumstances continue to change and evolve.
Read next: Google Makes Case for Managing AI Models