How far should AI’s decision-making authority go?

Perhaps the leading value proposition for artificial intelligence  (AI)  is its propensity to help humans make decisions. With the right data and the right analytics, people can choose a course of action based on solid information, not hunches or guesses, and this is expected to bring dramatic improvements to the business model.

Elsewhere, however, AI won’t focus on the choices that humans must make, but the ones it will make for itself. In the realm of automation, in particular, AI will be tasked with broad decision-making capabilities — all data-driven, of course — to streamline data flows, improve manufacturing processes, direct traffic and perform a wide range of other functions.

This begs the question, where is the line between what AI should decide and what is best left for humans?

Being careful with autonomous AI

Implementing autonomous AI across the full scope of the enterprise data ecosystem – from the data center to the cloud to the edge to connected devices – will require careful coordination between a number of emerging data initiatives. Alan Young, chief product officer at automation firm InRule Technology, recently highlighted the intersection of machine learning (ML), decision automation and process automation and how it will drive better business results. With ML providing the probabilistic decisioning logic and both decision and process automation contributing consistent, orchestrated rules-based governance of operations and behavior, processes gain the ability to act on real-time, dynamic inputs and values without the need for constant, direct human oversight.

With this framework in hand, Young says organizations can not only produce more successful outcomes from its data process, but do so at scale rapidly and consistently. Already, this is evolving beyond a mere competitive advantage to an operational necessity by allowing organizations to detect and respond to both opportunities and threats as they emerge.

Still, there is an element of the slippery slope to all of this as AI becomes more infused into the digital universe. As Talend’s Julinda Stefa noted recently, allowing AI to choose your playlist or manage your die-cutting process is one thing, but off-loading decisions on stock trades, healthcare choices and other critical activities is quite another. Unless and until AI is empowered with some degree of humanity, businesses and people should tread carefully as to what it should and should not do. Fortunately, the human touch can be added to AI using three basic techniques:

• Ensure data quality – not just to detect errors but to ensure data is timely and relevant.
• Make data accessible to all – users should have complete visibility into the data used to train models, and that data should be comprehensive.
• Prioritize security and compliance – security and privacy policies should be reliably documented, regularly updated and consistently enforced.

Clearly, rigorous monitoring of AI decision-making will have to be a top-priority going forward, even if the model is aimed at rote, routine tasks. Michael Ross, senior vice president of retail data science at EDITED, offers an example of a retail bot empowered to mark down prices under certain conditions. If, say, at the start of swimwear season, sales of a particular line haven’t hit their targets due to unseasonably cold temperatures, an AI model may execute its logic and put the entire stock on clearance, losing millions of dollars when normal buying patterns resume. The best way to prevent this is to keep humans in the decision-making loop so they can prevent mistakes like these, or correct them quickly if they do happen.

What is intelligent automation?

What we’re seeing in the integration of AI and automation is the emergence of a new class of platforms called Intelligent Process Automation. In Cognizant’s view, the incorporation of tools like robotics process automation, cognitive technologies, optical character recognition and natural language processing produces a digital Swiss army knife that allows the enterprise to manage its exponential growth, provide clear and consistent results and accelerate business outcomes. Along with this new technology, however, organizations will have to re-orient themselves to the new operational paradigm by addressing gaps in the workforce skillset and overcoming cultural resistance to change.

It’s important to remember that intelligent automation should not become an excuse to put systems and processes on autopilot. Even as the knowledge workforce evolves from an operational resource to a strategic one, someone still has to keep an eye on how the business is running. And while technology may become empowered to make more, and more important, decisions, it still must answer to someone for its actions – if only to let the good decisions multiple while the bad ones are neutralized.