Artificial intelligence (AI) can be defined as intelligence that emanates exclusively from machines or robots and cyber-physical systems. The research area AI tries to understand and emulate human perception and human action. According to research, these are systems that approach rational thought, imitate it, replicate it, automate it, or even surpass and improve it.
Arthur Samuel, one of the pioneers of machine learning, defined machine learning as a kind of computer study in 1959. He was sure at the time that computers could learn specific tasks without being explicitly programmed. What at that time was still a theory due to the scarcity of experience, is now radically advanced thanks to the AI sections Machine Learning and Deep Learning.
The KI research area does not concentrate on a single, uniform technology. Instead, they are interrelated technological elements (software and hardware) whose use is very diverse and depends on the problem at hand. In general, it can be said that artificial intelligence consists of perceiving, processing and learning elements.
Application / Application areas KI general
There are numerous examples of the practical use in companies[1]
- Examination and assessment (sample classification): credit assessment, insolvency examination, image recognition
- Class formation (clustering): market segmentation, data mining
- Forecast (Prediction): Price forecasts, sales forecasts, cost forecasts
- Optimization: transport optimization (Travelling Salesman problem), sequence planning
AI and the subarea Machine Learning
To be precise, Machine Learning (ML) has nothing to do with artificial intelligence in the real sense of the word. Instead, ML relies on pattern recognition that can be performed by machines or computers. ML is thus the most straightforward and also the cheapest way to impart a kind of ‘artificial’ intelligence to computers. Training phases are coordinated by large amounts of data, so-called training data, and algorithms, which lead to the software used, coupled with the use of the processed data, acquiring the ability to modify itself or learn something new. Classic examples of machine learning are the learning algorithms ‘Naive Bayes’ and ‘Collaborative Filtering’. The latter is known from the online platform Amazon ‘Customers who bought this item also bought’. Naive Bayes, on the other hand, are used within spam filters for email management. The actual intelligence of machine learning systems is the ability to modify oneself – as soon as more information or data is available. If an ML system has high-quality data available (see also Smart Data), human intervention can even be dispensed with.
AI and the subdivision Deep Learning / Neural Networks
Deep Learning is technologically based on artificial neural networks (KNN). In most cases, the algorithms of in-depth learning are based on the functioning of the human brain. The KNN computational models and the algorithms used to learn, like their human counterparts, from experience by individually adapting the simulated neuron connections to fit precisely.
Establishment of a neural network
According to the Cluster für Künstliche Intelligenz (Bremen AI), “the neurons of an artificial neural network are arranged in layers and are usually connected in a fixed hierarchy. The neurons are mostly connected between two layers (Inter-Neuronlayer-Connection), but in rare cases also within one layer (Intra-Neuronlayer-Connection)”.
Editor’s note: The effectiveness of a neural network, therefore, lies in its networking. Thus the layers mentioned above or the individual layers are linked to each other using individual neurons – i.e., each neuron of one layer is always connected to all neurons of the next layer. Personal information flows into the so-called input layer, which in turn delivers the information to intermediate layers (hidden layers); at the end, the interpreted data is provided via the output layer – “where the output of one neuron is the input of the next.” When a neural network obtains new data, these connections are strengthened or reduced; likewise, each link in a neural network can be adapted by ascribing greater or lesser relevance to a characteristic.
Artificial neural networks have so-called core components or basic building blocks, which can be found in all network types[2].
- Processing units (processing elements)
- Connections between processing units
- The network topology
The dynamic core components listed below describe the information processing that takes place within the KNN. The processing intervals are divided into the following phases.
- Learning stage
- Processing stage
AI: Elements of perception
Artificial intelligence must be able to absorb information to be able to understand or perceive the real world. Thus, AI must have the ability to process text, capture images and videos, record sounds, and capture information about environmental conditions such as temperature, wind, and moisture. In this context, the Internet of Things (IoT) plays an important role, as it is about collecting heterogeneous data from many different devices, drawing conclusions and learning from them; a task that today also poses significant challenges for advanced data analysis tools. In a nutshell, IoT consists of the information collected (information and devices), how these things are connected, how data is collected, what can be learned from the data and, ultimately, what can be made different from it.
To ensure that the information is recorded, AI systems require sensors that detect (see/hear/ ‘smell’) the outside world for them to subsequently carry out specific actions and activities. The sensors are installed, for example, in cameras, microphones, and within robotic systems.
AI: Elements of data processing and learning
AI systems usually receive the information to be processed in two different ways. Either the human being supervises the so-called ‘naming’ of the data and manually ensures that the respective computer is provided with it. It thus also establishes the appropriate algorithm rules to achieve a specific output result, or it is unattended learning. The data records are fed to the system manually or automatically; artificial intelligence, however, draws its conclusions without feedback to humans. In both cases, information is collected, combined and repeatedly refined with the help of mathematical tasks. Since the naming of the data and the set of rules around the particular algorithm depend on the human being, this labor-intensive process is also the most cost-intensive. There is, therefore, a broad industry today that only names data for the training of computer programs.
Example AI service chatbots
In the future, artificial intelligence will replace people where computer-aided service leads to the desired result faster and with higher quality on the customer side. Service chatbots, for example, make it easier for customers to search for information and carry out simple transactions via voice or chat interfaces. Machine learning algorithms scan a variety of products and technical documentation and automatically answer frequently asked questions. It should be noted that such systems focus exclusively on fundamental and straightforward issues. “In the future, modern methods of Natural Language Processing, as well as frameworks and platforms, will facilitate Chatbot development; the creation of powerful bots will then also be standard,” Rudolf Grötz, Senior Technical Engineer at Raiffeisen Bank International[3].
In the future, users will search less and less on websites for relevant information, but merely ask the chatbot a question.
Rudolf Grötz – ix Magazine for Professional Information Technology / german / 06/2018, page 50
AI and Logistics/Intralogistics
The application of artificial intelligence is particularly suitable in the area of logistics or intralogistics, and almost all areas of the supply chain. In the logistics area of a company as well as within the logistics chain of all participating companies, enormous amounts of data are generated daily, both structured and unstructured; artificial intelligence is virtually predestined for exploiting this information. Thus, AI can support the development of new methods and behaviors, for example, the proactive generation from reactive processes; and in the sense of planning security, concrete predictions can be made instead of assumptions and rough estimates. Of course, it is also possible to concentrate solely on the existing processes; manual, as well as automated processes, can only be optimized regarding time. Also, services cannot be standardized but personalized and thus made more customer-friendly.
An example shows how glaring the advantages of artificial intelligence can be in intralogistics: In a market, 7.5 percent of articles are not available due to gaps in the shelves due to manual ordering. The error rate drops to five percent when a special AI software makes recommendations to a human dispatcher. If no human corrections are possible and artificial intelligence performs warehouse and logistics tasks completely autonomously, the error rate drops to 0.5 percent.
Joachim Bengelsdorf / diyonline Magazine[4]
Logistical examples: Artificial Intelligence
Warehouse logistics is regarded as the first location for the widespread use of artificial intelligence. We also speak of the Learning Warehouse. A large amount of information is already being analyzed by algorithms there today.
- This is preceded by the fact that the AI is already able to predict with approximate accuracy the occurrence of certain events (via e-commerce, multi-channel, cross-channel). By, for example, analyzing order behavior more deeply, strong statements (and thus measures) can be made regarding future orders/orders, also to enable a faster shipping process. Important: Within a warehouse, the methods are automated in such a way that the respective processes, for example, goods receipt, storage, stock removal/translocation, picking, finishing and goods issue, are clearly defined regarding time; a positive statement based on intralogistics is therefore always possible.
- Another common practice is the recognition of trend articles at a certain point in time. If necessary, the responsible employees can then adjust the product allocation in the warehouse in order, for example, to force route optimization during picking; similar to the fast- and slow-moving principle.
- This also makes it possible to identify pending maintenance work or potential improvements in the material flow. This represents an essential step toward cognitive logistics. The latter is undoubtedly the big goal within the supply chain. The semantics required for this not only enables objects to be linked in a machine-understandable way; employees and leading management can navigate in real time from a pick (e.g., Pick-by-MDE) to the order or the original advice note in incoming goods to obtain relevant conclusions. Especially the semantic analysis of logistic processes is a unique method of information retrieval, which allows the interpretation of key figures and delivers exact results thanks to the inclusion of domain knowledge (synonyms, similarities, ontologies as well as taxonomies). For the future, it is conceivable that intelligent systems will make decisions based on predictive analytics entirely without human intervention.
- It is also worth mentioning that the optimization of the cooperation between humans and robots (human-machine interface) is also a practical field of application of artificial intelligence; namely by learning automata from humans. Using virtual reality and 3D modulation, humans can automatically generate digital information, such as particular hand movements that are accurate to the millimeter; to grasp individual positions, which in turn improves the grasping of robots within picking (pick-by-robot).
Summary Artificial Intelligence
From tour optimization, inventory management through to batch formation and the resulting synergy effects in order loading; artificial intelligence and its algorithms already support warehouse employees through targeted process optimization (see also control center in intralogistics). In the future, so-called cognitive systems will have the ability to learn, recognize patterns and derive recommendations for action – without being dependent on human cognitive ability. This enables them to support warehouse employees in making decisions or to inform them in advance of events that are likely to occur.
Sources:
[3] ix Magazin für professionelle Informationstechnik / german / 06/2018, Seite 50
