Today’s fast-changing business climate demands real-time visibility and up-to-the-minute recommendations from data and analytics. To survive the post-pandemic world, organizations need to be able to predict what’s going to happen next based on the data they have; and develop more discipline around decisions and actions. Processes for measuring impact and closing the decision intelligence loop will sharpen their focus.
According to Michael O’Connell, Chief Analytics Officer at TIBCO, policymakers are now dealing with challenges of AI-driven decisions including self-driving cars, and medical implants. Congressional committees are considering how to regulate social media platforms to reduce fake news, hate speech, and terrorist recruitment. These policy concerns are usually viewed through an ethical lens, like responsible, fair, transparent, and accountable AI.
“In 2022, we will see more AI startups focused on ethical AI and tools enabling regulation of AI applications, incorporating industry standards like the Robotics Industry Association and IEEE. While some argue this will limit innovation, ensuring ethical AI will only lead to better data insights, as organizations must ensure all data feeding algorithms are unbiased, trustworthy and clean,” he said.
Anomaly Detection is radiating into new applications
Anomaly detection, combined with root cause analysis and next best actions, is adaptively radiating into many application areas. Applying these methods to cheap and prevalent data sources such as sound (mp3s), in edge AI scenarios, is creating value. Coupled with sophisticated analytics techniques such as long short-term memory, this is opening up new application areas such as Cybersecurity Analytics.
“Human Centered AI (HCAI) is on the rise in 2022. AI is delivering on both dreams and nightmares when machines and automated software are in control. HCAI aims to amplify human abilities enabling people and software to create in extraordinary ways. For example the human brain can process entire images that the eye sees for as little as 13 milliseconds. Combining these human capabilities with AI and machine learning is opening new frontiers for adaptive machine learning systems,” he explained.
HCAI involves building reliable and transparent systems based on sound software engineering practices, such as testing software to check AI algorithms, using visual tools to reveal anomalies, and testing databases to enhance fairness in machine learning training datasets. A big step forward is to record activity that allows retrospective analysis after failures and near misses. Another vital feature is explainable AI, to understand AI decisions and address unfair or incorrect decisions.
In 2022, working from home has accelerated collisions and collaborations between teams of data scientists, devops and model ops developers to get data science apps into production. Emerging from this is a focus on converting ad-hoc processes into a controlled environment, for managing low code and code first components, processes for data flows and model connections, along with rules, actions and decisions.
“Continuous analysis of models actually in operations is also in focus to assess ROI of the data science app, model drift and model rebasing.”