The corona pandemic poses major challenges for logistics and delivery traffic. How can Artificial Intelligence support them?
Sören Kerner: The logistics network in Germany is highly optimized, as are warehousing and intralogistics. Even today, this is still achieved through a high level of manual input in planning. At the same time, supply chains are becoming increasingly interlocked, and the complexity of logistics is growing super-exponentially. Corona has confronted logistics with a state of emergency - the general conditions were turned upside down virtually overnight by market behaviour. This is where the greatest weakness of manual optimisation became apparent - the lack of flexibility. A vivid example that has affected everyone is toilet paper. This is not a seasonal article. But due to the Corona-related hamster purchases, demand has changed dramatically, so that customers stood in front of empty shelves for weeks and at the same time videos of forklift drivers in warehouses full of toilet paper circulated on the Internet. The reason was rigid supply chains. This is where AI can help in a targeted manner to optimize processes at the push of a button and as needed - right up to autonomous, self-optimizing flows of goods.
Why is logistics particularly suitable for AI-based optimization?
Sören Kerner: Logistics is highly complex, but follows very simple rules - similar to the game of Go. As is well known, this is a prime example of the use of Machine Learning methods to achieve "superhuman" results. The complexity in both cases results from the diversity and multimodality of the decision options at any given time. In logistics, this applies both to optimizations of supply chain decisions and to intralogistic systems. But there are also differences: Go can be perfectly simulated. As a result, Artificial Intelligences such as Alpha Go Zero can train against itself in a highly scalable manner and generate vast amounts of data for optimization. Logistic systems can also be simulated in a discretely scalable way. However, something gets lost in the step from strategic game to reality - the perfect model. Therefore, research on the modelling of logistic systems is a crucial prerequisite for the transferability into practice.
Can AI contribute to making global flows of goods more resilient in the future?
Sören Kerner: Absolutely! The changeability that AI promises for logistics is decisive for the resilience of the flow of goods. In addition to the algorithmic requirements for training AI methods, there are other hurdles to overcome - such as data transparency. The optimization of complex flows of goods with a large number of participating companies can be successful if AI can access data across company boundaries. The Silicon Economy initiative launched by the Fraunhofer IML has set itself the goal of creating a B2B AI ecosystem in the coming years in close cooperation between research and industry. Based on business models of a platform economy in combination with a secure and sovereign communication infrastructure using Industrial Data Space and GAIA-X, Silicon Economy will enable the free development of AI algorithms to an unprecedented extent.
That leaves the all-important question: trust in the algorithms. When AlphaGo first took on the world's strongest player, Lee Sedol, the 37th move of the AI system provoked reactions from commentators that ranged from astonishment to amusement. AlphaGo seemed - in human terms - to have made a catastrophic mistake. However, subsequent analysis showed that this move in particular was the cornerstone of AlphaGo's victory in this game. Logistics is a rather conservative industry. It still needs to be persuaded to give up the reins of action through AI optimization. The Silicon Economy initiative will lay the foundation, but it will also require courage on the part of the industry.
The application scenario On the Way to Intelligent Mobility developed by Plattform Lernende Systeme shows what AI-supported logistics could look like in the future.