Overview: Artificial intelligence in intralogistics
Artificial intelligence (AI) has long since arrived in the everyday and business world, and no longer just the target of a science researching at the technological maximum. Whether it is a matter of translations or image recognition, of cures for diseases or the development of new materials, AI can apparently be used to solve a wide variety of problems that have so far seemed difficult or even impossible for human beings to solve.
Triggered by the sheer endless growth of data and ever-increasing computing power, automation is being driven forward in all sectors of the economy by the ongoing digitalization process. On high-performance hardware and software platforms, the machine learning methods of AI provide the necessary tools to extract complex dependencies from large amounts of data without having to be explicitly programmed for it.
Previously published articles of the series:
Strong and weak AI
Not all AIs are the same. Experts differentiate between weak and strong implementation achievements of the effort to transfer human learning and thinking to computers. A so-called weak AI is usually used to make intelligent decisions for specific sub-areas, such as process automation. In contrast, the strong AI tries to emulate or exceed human intelligence. Thus, it no longer acts reactively like a weak AI, but of its own accord. According to experts, however, a strong AI is still beyond the current technical possibilities, while great progress has already been made in the field of weak AI in recent years.
Functionality and application areas of AI
In present times, when reference is made to so-called artificial intelligence, the weak AI is usually meant. It uses methods from mathematics and computer science to simulate intelligent behavior. Mathematical approaches from statistics, mathematical programming and approximation theory are used to accomplish the desired tasks. An artificial system learns with the help of large amounts of data to derive patterns and then to generalize them. With this system, decisions can be made that produce a desired result even with unknown data.
Typical fields of application for AI are tasks in which it is necessary to search for the best solution, find the most efficient way, plan procedures or optimize processes. Examples of applications can be found in our everyday lives in thousands of ways: search engines structure the flood of information on the Internet for us, computer vision programs enable safer road traffic, and the combination of speech recognition with the use of knowledge-based systems serves us as an everyday helper in the form of Amazon’s Alexa application or Apple’s Siri.
AI in intralogistics
But AI is not only used in our private environment. The volume of data generated in intralogistics alone is an argument in favor of using machine learning to optimize processes and exploit synergies. This involves algorithms that derive new data and information based on the analysis of historical data. This opens up a wide range of possible applications, for example in the design of new picking orders and the avoidance of incorrect picking, in the field of predictive maintenance or in the simulation of intralogistics systems.
TUP master’s degree candidate Sina Hill distinguishes between five areas of intralogistics in which AI is valuable: Big data analysis, optical recognition, process optimization, forecasting and human-machine interaction.
Big data analysis is used to process historical data sets and to identify structures and interrelationships, for example with regard to the various facets of warehouse optimization. Possible areas of application include the intelligent selection of storage locations, strict inventory control and an optimal layout of the storage facility to ensure process flexibility. AI works most efficiently on the basis of large and highly complex data volumes that a human being could never handle fast enough.
Both in the tracking of objects in storage facilities and in the optimization of processes, AI offers a high potential for improvement. The more the warehouses are filled with heterogeneous goods or the more structured data is available from operating processes, the greater the potential gain from AI decisions is.
The use of AI enables analytical predictions to be made for the intralogistics sector, e.g. on market demand, returns or personnel requirements. These forecasts are no longer based only on existing data, as in the past, but also on patterns developed through machine learning.
TUP position on AI
TUP maintains an ambivalent view of the importance of AI for intralogistics. We welcome the fact that the industrial application of AI systems is a key trend in modern intralogistics. On the negative side, however, we see the inflationary use of the term AI, which could stir up fears among our clients and business associates and hamper possible process optimizations. Especially medium-sized companies, which often find it harder to introduce new technologies than large corporations or agile start-ups, do not usually see AI as a universal remedy for improving business processes.
Concluding words
The hype about artificial intelligence can be taken into account in so far as it is certainly a seminal field with great potential. However, the prerequisite for the in many cases somewhat early statements is the appropriate scientific and empirical foundation. If this basis leads to a working weak AI (see above), this will undoubtedly have advantages, but no AI is infallible.
One factor that can clearly be optimized in the context of intralogistical porcesses is decision time. Since the computational effort for a decision is largely shifted to the machine learning phase, only a fraction of the necessary work is required at a given point in time when a state is to be assessed. Consequently, decisions can be made very quickly by AI. In addition, an AI-supported system can be trained continuously or trains itself. Thus, it can automatically adapt to changing boundary conditions without the need for an explicit definition of these conditions.
One disadvantage of AI is that a decision cannot be substantiated. AI therefore sometimes makes correct decisions for the system, which must be evaluated as errors by humans. Amazon provides an example: The company was supported by an AI with regard to personnel decisions. The AI was trained with the data of successful employees from the past. To ensure that no one is discriminated against according to gender or race, the data was entered anonymously. Nevertheless, the percentage of women was much lower than expected. Machine learning established patterns based on other factors, including language, which were part of the data set and thus reflected the more male-dominated technical industry.
This kind of consideration therefore suggests the use of AI where many decisions have to be made, but where it is not fatal if a decision is suboptimal or wrong. The idea of using AI in storage location determination seems to be fruitful since intralogistics can come up with these boundary conditions: If one or ten items out of 10,000 are in a bad position, this is acceptable.
And if an AI, after analyzing many hundreds of parameters, comes to a result that a user finds unacceptable, he or she can still intervene. The employee who does not accept the advice and consciously changes a parameter will have appropriate reasons for doing so.
To the overview of the articles
Teaser image: Franki Chamaki – Unsplash