How much should AI decide on its own, and when should human oversight intervene? Eugene Amigud of Infios explains to Paul Hamblin how supply chain operators should answer this key question.
Try some word association games with the now-ubiquitous term ‘AI’. Chances are the words ‘fear’ and ‘trust’ (or lack of it in the latter case) are likely to feature quite highly. This matters, says Eugene Amigud, because he argues that lack of trust is a key reason why AI projects can fail. On the other side of the coin, when trust is sought, fostered, nurtured and achieved, great things can happen.
As a 25-year veteran supplying technical solutions to the supply chain and retail sectors, Eugene Amigud (pictured, below) is well-placed to advise. He serves as Chief Innovation Officer at Infios, the global supply chain execution software company formed by Körber AG and KKR that unites warehousing, transportation and order management into a seamless, adaptable network and now serves over 5,000 customers across 70 countries.
He likens the path to autonomous AI to a loop, or a flywheel. “You sense, you decide, you act, you learn, you repeat. If you keep this framework in mind, you embark on the road to successful autonomy. If you simply take unilateral action outside of this repeating framework, you’re unlikely to be successful.”
The process is known as Graduated Autonomy.
“Graduated Autonomy is about building the trust with the system,” he explains. “Start small, optimise a single decision, then you learn, build the trust, decision by decision, until you have the trust and confidence to build a fully autonomous supply chain.”
How might this work for an operator in a warehouse? He cites the example of a weather event with the potential to disrupt schedules and operations.
“The AI may ask itself: ‘Do I need to replan my loads within my inbound transportations, because the weather event will affect them? Alternatively, it may be that no loads are due in the next few days from the weather-affected region, so I do not need to take action. Or do I need to escalate the decision to humans, based on the criteria that I have been set?”
The point is that decisions are made, some autonomous, some manual, constantly collecting data, learning and assessing the impacts of those decisions on customers and on efficiencies. The loop is in place – sensing, deciding, acting, learning and repeating. It’s the Graduated Autonomy concept in practice. “As AI builds the trust with the operators, it is about how much it can now be autonomous with decisions and actions, rather than simply informative.”
It is about action, rather than information.
“Let me give the example of analytics dashboards,” he says. “They look great, they are clean and smart, easy to use. They show a lot of information, but the crucial difference is that they do not act on that information. Their usefulness is limited.”
Intelligent supply chain execution is very action-oriented. “This might include altering loads, meeting flexible customer commitments, optimising labour in the warehouse, continuous optimisation on many fronts. It’s that change from simply showing the data to acting upon it. That’s the Graduated Autonomy difference – it’s pretty easy to just show the data, to see how your systems are performing, but now you have a system of actions, such as sending trucks to different locations, informing customers of delays, projecting different staff requirement on warehouse floor. So it’s very impactful.”
Fear of AI often stems from a perceived fear of losing control. Amigud points out that this is a fallacy, because control always ultimately rests with the user.
“The user is always in control. But, over time, the user is happy to cede control to the supply chain software, allowing it to start making adjustments once that operator is confident that the system can be trusted to do so.”
He divides systems into three separate categories – Assisted, Automated, and Autonomous. “You could start with Assisted which makes recommendations only, and the user approves every decision. So for instance, in bad weather, a shipment will be rerouted, but the user supervises. The next stage is more automated. The user sets new criteria and policy, so for instance, the system might be enabled to make decisions on all projected delays of up to 48 hours, with further exceptions reported for human supervision.
“The third stage is autonomous and again, the user will apply guardrails. For instance, instead of an explicit number of hours, the system evaluates dynamically the amount of delay and intelligently notifies users.”
Transparency is front and centre at all times. “Bear in mind that with all decisions, there is a full audit trail, a full evidence file explaining why decisions were made, what constraints were considered and what were not. The objective is to build and retain trust with the operators. Once that trust is in place, autonomous AI can be allowed to take over with confidence.”
What advice would he give to companies embarking on an AI-powered supply chain software transformation? “My advice would be to address the business need first, which may include some capabilities from warehouse and transport powered by AI. There is no point considering AI capability without defining the business need it serves first. That connection is vital to successful implementation.”
And it doesn’t have to be a huge process, he confirms.
“Start small. But with an understanding of where you want to end up. That’s vital.”
