Artificial intelligence (AI)

Categories of artificial intelligence

Weak/low AI

Weak/narrow AI describes systems that are specialized for a specific task or a limited area of application. Examples include systems for speech recognition, image recognition, chess or self-driving cars. These AI systems are very powerful in their respective area, but cannot transfer their capabilities to other areas.

Strong/general AI

Strong/general AI would be systems that have a general thinking ability comparable to human intelligence. Like the human mind, they could solve problems in a wide variety of areas, recognize correlations and acquire new knowledge independently. This form of Artificial General Intelligence (AGI) has so far only existed in theory and is the subject of intensive research (and many Hollywood films).

Core areas and technologies

Machine Learning

Machine learning (ML) is one of the central areas of artificial intelligence. Computer systems should learn from data themselves and not work with rigid program rules. ML algorithms independently recognize patterns and correlations in large amounts of data and can thus make predictions or decisions.

The most important categories are:

• Supervised learning: The system is given input data and expected outputs as training examples so that it learns the underlying patterns. Applications include image classification or spam detection.
• Unsupervised learning: Here, only input data is available and the system has to find structures and clusters in the data itself. Applications include customer segmentation or recommendation systems.
• Reinforcement learning: An algorithm learns the optimal actions in an environment through trial and error. These are signaled to it through “rewards” and “punishments”. Applications include robotics, game strategies and process optimization.

Well-known ML techniques include decision trees, support vector machines and neural networks. Basically, almost all known applications with AI are currently based on some form of machine learning.

Deep learning

Deep learning is a branch of machine learning that is based on artificial neural networks. These mimic the way the human brain works by processing information in networked, hierarchical layers.

Thanks to enormous computing power and huge amounts of data for training, deep learning models can now independently recognize complex patterns in images, audio and text. Applications include

• Computer vision: object recognition, image segmentation, autonomous driving
• Speech recognition: speech recognition, translation, speech synthesis
• Natural language processing: text analysis, sentiment analysis, dialog systems

Transformer architectures such as BERT or GPT for text processing are particularly powerful. Generative models for image synthesis such as DALL-E or Midjourney are also based on deep learning.

Natural language processing (NLP)

Natural Language Processing (NLP) is concerned with the interaction between computers and human speech in written and spoken form. Its core tasks are

• Speech recognition: converting speech into text
• Text analysis: recognizing words, themes, moods in texts
• Translation: Automatic translation between languages
• Speech synthesis: Generating natural language from text
• Dialog systems: understanding and generating conversations

Many NLP systems use deep learning, but other methods such as rule-based systems or statistical models are also used. Applications include virtual assistants, chatbots, translation services and more.

Generative AI

Generative AI models (such as ChatGPT) are able to generate completely new content such as text, images, video, audio or program code that is similar to the training data. The models are usually based on deep learning and specially trained transformer architectures.

Large Language Models (LLMs) / Foundation Models

Large Language Models (LLMs), also known as Foundation Models, are gigantic language models that have been pre-trained on huge amounts of data, for example with content from the entire Internet. They are extremely powerful and can be adapted for a wide range of applications. They are currently mainly used in generative AI. They are considered a promising basis for general artificial intelligence.

Further Whitepaper

We show you the most important AI trends in the areas of sales, merchandise management and service.

Request now for free

AI or AI – is there a difference?

The terms artificial intelligence ( AI) and artificial intelligence (AI) basically mean the same thing: the ability of machines to carry out human-like thinking and learning processes. While the term AI is usually used in German-speaking countries, the abbreviation AI is more commonly used internationally . There are only differences in perception: for many, AI sounds more modern and technological, while KI is often associated with German or European developments. In terms of content, however, there is no distinction – regardless of whether AI or AI, the goal remains the same: intelligent systems that automate workflows and support decision-making processes.

Artificial intelligence in the ERP environment

Artificial intelligence is no longer just a theoretical concept. It is already an integral part of modern ERP systems, especially in SMEs. Integrated AI functions analyze automation potential in real time, create forecasts for stock levels, support automated orders and enable predictive maintenance.

In addition, AI identifies promising sales and leads, suggests targeted up- and cross-selling options and creates significant efficiency gains through automatic document recognition, OCR and intelligent process analysis.
Of particular interest: for many companies, AI is no longer a “nice-to-have”, but a real game changer – whether in the warehouse, sales or service. The technology enables ERP processes to be made more resource-efficient, faster and more user-friendly.

Areas of application and examples

The list shows some of the current areas of application for AI. It is by no means exhaustive. AI solutions are currently developing at breakneck speed. New functions are presented almost every week.

Production/manufacturing

• Industrial robots for assembly, packaging and welding: AI-controlled robots are increasingly taking on repetitive and dangerous tasks in production facilities, from automotive assembly to the packaging of goods.
• Production monitoring and quality control through image analysis: Camera systems with AI image analysis detect faults and quality defects in real time, enabling seamless monitoring.
• Predictive maintenance through monitoring and prediction models: Sensor data from machines is analyzed in order to predict impending failures and schedule maintenance in good time.

Traffic/mobility

• Driver assistance systems such as lane departure warning and emergency braking systems: AI systems in cars provide support by intervening in dangerous situations or unintentionally leaving the lane.
• Autonomous driving through object recognition and situation analysis: Self-driving cars use AI to recognize objects, signs and traffic situations for safe navigation.
• Intelligent traffic control to optimize traffic flows: By analyzing traffic data, traffic lights and routes can be optimized in real time.

Healthcare

• Disease detection through image analysis and prediction models: Based on image data and patient histories, AI models can detect diseases such as cancer or Alzheimer’s at an early stage.
• Virtual assistants for triage, treatment recommendations, patient advice: AI systems can use speech processing to provide initial assessments and advise patients.
• Drug development through virtual screening and effect prediction: AI can be used to identify promising drug candidates from test data and predict their effects.

Finance

• Credit risk analysis and credit scoring: AI models analyze customer data and make predictions about credit risks for lending.
• Fraud detection for transactions and activities: By recognizing anomalies in data patterns, fraudulent activities can be identified at an early stage.
• Portfolio management and optimization of investment strategies: AI systems support the optimization of portfolios and investment strategies through predictive models.
• Algorithmic trading and automated securities trading: Complex trading systems execute transactions in seconds on the basis of market data.

Customer service/marketing

• Chatbots and virtual assistants for customer service: AI systems can use voice processing to automatically answer incoming inquiries and escalate them to humans.
• Personalized product recommendations based on customer profiles: Analysis processes create individualized recommendations for each customer.
• Customer satisfaction analysis from interactions and feedback: evaluation of customer conversations and ratings to optimize service quality.
• Predictive analytics for future customer needs: Predictive models identify trends and future demand for products and services.

Entertainment/Media

• Personalized recommendations for films, music and games: Individually suitable content is recommended based on usage history and preferences.
• Automatic subtitling and translation of content: Voice processing allows content to be subtitled and translated in real time.
• Virtual characters and digital actors: Motion capture and speech synthesis can be used to create photorealistic, animated characters.

Security

• Surveillance and object detection for security systems: camera systems with AI image analysis detect people, objects and dangerous situations.
• Prediction and detection of threats and attacks: By analyzing patterns in data, cyber attacks and other threats can be detected at an early stage.
• Biometric identification and facial recognition: For access control and identification of persons using features such as face, iris or fingerprint.

Ethical aspects

In addition to its enormous potential, the use of artificial intelligence also poses numerous ethical challenges and risks that need to be carefully considered:

Accountability and transparency

One of the main problems is the question of responsibility when AI systems make mistakes or cause damage. Who is liable in such cases: the developers, the operators or even the AI (as a legal entity) itself? Closely linked to this is the demand for transparency and traceability in AI decision-making. This is a major challenge, especially for systems such as deep learning, which work as a “black box”.

Data protection and security

Many AI applications are based on the processing of huge amounts of data, often including personal and sensitive information. The use of AIs in organizations must therefore be regulated and monitored in order to prevent violations of privacy and data protection.

Effects on the world of work

The use of AI for automation will lead to job losses in certain sectors and occupational fields. At the same time, however, completely new job profiles and qualification requirements will emerge. This transformation of the world of work must be designed in a socially responsible way. This includes preparing employees for change through training and further education.

Artificial superintelligence

Will it one day be possible to develop a powerful artificial intelligence that is on a par with or even superior to human intelligence? This would raise fundamental ethical and existential questions. How can we ensure that such a “superintelligence” respects human values, interests and safety? Many experts see uncontrolled superintelligence as one of the greatest potential risks for humanity.

Discrimination

AI systems adopt the biases and prejudices contained in their training data. On this basis, they may make discriminatory decisions, for example when granting loans, recruiting staff or prosecuting offenders. This development must be specifically counteracted by programming the algorithms and carefully selecting the training data.

Disinformation and manipulation

The ability of modern generative AI to realistically generate content such as text, images, audio and videos harbors enormous risks. They can be used to spread fake content that looks real in order to deceive people. Systems for recognizing AI-generated content as well as regulations and transparency rules are possible countermeasures. However, it is debatable how effective these are.

Protection of intellectual property

Generative AI systems are trained with huge amounts of data that often contain copyrighted works. Many authors believe that the unsolicited use of this data for AI training infringes their rights. Court cases are currently underway.

In addition, the output of AI-generated content raises questions of intellectual property. Who owns the works created by an AI? The developer of the AI system, the operator, the user or even the AI itself? Existing laws are not clear on this.

Technical challenges and limitations

Overall, AIs are still a very young, immature technology with many limitations such as these:

Scalability and computing power

Very large AI models such as GPT require enormous computing power and vast amounts of training data. Training such models requires data centers with thousands of GPU accelerators and consumes as much energy as entire cities. Scalability to even larger models has so far come up against practical and economic limits.

Data scarcity and quality

In many areas, there is a lack of available and high-quality training data. Especially for rare use cases, new languages or specialized fields, there is often a lack of annotated data sets in the required quantity and quality.

Robustness and fault tolerance

Today’s AIs are still quite susceptible to interference or targeted attacks. AI systems can easily be misdirected by manipulating the input data and led to incorrect results. Fault-tolerant AI architectures that are protected against attacks are therefore an important field of research.

Transferability of skills

Even powerful AI models are often unable to transfer the knowledge they have learned in one domain to new, unknown tasks and areas. If they are given a task that they are not yet familiar with, they produce nonsensical or incorrect results. This is one of the major limitations of today’s “weak” AI.

Further Whitepaper

We show you the most important AI trends in the areas of sales, merchandise management and service.

Request now for free

Interesting facts from the blog

13 use cases for AI in the ERP system
These factors prevent the efficient use of AI
Material requirements planning in ERP with AI