Can AI help human decision makers fill in the gaps missed in data?

Tackling Big Data

Analytics and forecasting is likely to be one of the areas where AI makes the biggest difference – enormously extending human abilities when it comes to crunching Big Data and drawing meaningful conclusions from it.

But can AI tackle ever-evolving information streams faster and more accurately than humans? “On a fixed given processing task, machine-based processing will be faster than human based processing,” says David. “As data and tasks evolve, the greater adaptability of human cognitive processes will still give humans an advantage. Existing AI-based processes will need modifying to the new tasks and data. These modifications will be done by Data Scientists and AI specialists, though the time taken to make such modifications will shorten as Data Scientists become more experienced in adapting existing AI approaches and methods to new business tasks.”

And the machines are changing. David explains, “We are seeing an increase in interest in AI meta-learning systems, such as recent research from Google on meta reinforcement learning, [that’s to say] systems that learn how to learn and whose scope to adapt to new tasks is greater. Such systems will be better than humans at analysing certain evolving data and tasks, but may still be limited in their scope.”

So what will ultimately be automated?

The key question of whether tasks can be automated can be answered by weighing up the positives and the negatives, says David.

“Where the value and/or risk in making a decision is high, the decision will still be made by a human, possibly augmented by AI, where the AI predictions can be shown to be accurate,” says David. “Where the value of the decision being made is intermediate with intermediate or low downside risk, then AI systems can replace human decision makers where the AI systems are accurate. Where the value of the decision is low and the downside risk of the decision is low, then AI systems can replace human decision makers as there is little benefit to using human resource to make the decision.”

What this means is that human supervision is a must for decisions that are particularly important or sensitive. Where algorithms are training themselves, meanwhile, inspection is absolutely critical – as the machine is working entirely outside of human direction and failures will go undetected without testing.

The upshot of all this is that humans and AI aren’t generally going to be working in isolation– but instead will be working closely together for a long time to come.