Pređi na glavni sadržaj
AI Foundation trening (ICO AI FND)

AI Foundation trening (ICO AI FND)

Dvodnevni onlajn uvod uživo u veštačku inteligenciju: osnovni koncepti, stvarne primene, rizici i EU AI Act. Priprema vas za ispit ICO Artificial Intelligence FND i pomaže da ispunite očekivanja o AI pismenosti iz člana 4. Prethodno iskustvo nije potrebno.

Trajanje
12 sati
Format
Onlajn
Cena
390 kredita
Sertifikacioni ispit
Opciona dopuna
Sledeći termin
7. dec 2026. – 8. dec 2026.

Sledeći termini

Prijavi se za trening

O ovom treningu

Generativna AI ušla je u svakodnevni posao brže od bilo koje tehnologije pre nje: piše nacrte dokumenata, odgovara kupcima, sažima sastanke, piše kod. Ali dobri rezultati ne dolaze od nasumičnog eksperimentisanja. A otkako je EU AI Act stupio na snagu, razumevanje AI više nije stvar izbora: član 4 traži od organizacija koje razvijaju ili koriste AI sisteme da preduzmu mere koje podržavaju AI pismenost zaposlenih. Ne propisuje nivo znanja, ali očekuje da možete da pokažete šta ste uradili.

Ovaj Foundation trening vas tokom dva dana sa predavačem vodi od rasutog praktičnog iskustva do struktuiranog radnog razumevanja veštačke inteligencije: šta današnji sistemi mogu, a šta ne, gde stvaraju vrednost, gde podbacuju i šta zakon očekuje kada ih stavite u upotrebu. Posle treninga umete da procenjujete AI, a ne samo da je koristite.

Kome je trening namenjen

  • Profesionalcima u bilo kojoj funkciji, menadžment, operacije, marketing, HR, finansije, koji koriste AI alate i žele da razumeju šta stoji iza njih
  • Vođama timova i menadžerima odgovornim za ispunjavanje očekivanja AI Act-a o AI pismenosti svojih ljudi
  • Analitičarima, projekt menadžerima i konsultantima koji pripremaju svoj prvi AI projekat
  • Svima koje zanima ili koji moraju da uvedu AI u organizaciju, bez tehničkog predznanja

Šta ćete naučiti

  • Da objasnite osnovne koncepte i terminologiju veštačke inteligencije, mašinsko učenje, duboko učenje, generativnu AI i velike jezičke modele, jednostavnim poslovnim jezikom
  • Da razlikujete šta AI sistemi zaista rade dobro od onoga što samo naizgled rade dobro, i prepoznate njihova praktična ograničenja
  • Da prepoznate gde AI stvara vrednost u vašem sektoru, polazeći od stvarnih evropskih primera
  • Da prepoznate karakteristične rizike AI, pristrasnost, halucinacije, curenje podataka, i primenite svakodnevne navike koje ih obuzdavaju
  • Da koristite efikasne tehnike promptovanja i sistematski procenjujete kvalitet AI izlaza
  • Da se snađete u EU AI Act-u: kategorije rizika, obaveze koje iz njih proizlaze i šta AI pismenost iz člana 4 znači za vas i vaš tim

Program treninga

Prvi dan: Kako AI radi i kako raditi sa njom

  • Šta je veštačka inteligencija, a šta nije: ključni koncepti i terminologija
  • Od mašinskog učenja do generativne AI: kako se današnji sistemi grade i treniraju
  • Praktično: efikasan rad sa AI asistentima, obrasci promptovanja koji se drže u praksi
  • Šta AI radi dobro, gde podbacuje i kako uočiti razliku
  • Gde AI stvara vrednost: primene u različitim sektorima, sa razrađenim primerima

Drugi dan: Rizici, pravila i stavljanje AI u upotrebu

  • Pejzaž rizika: pristrasnost, halucinacije, privatnost i curenje podataka, praktično, sa merama za svaki
  • EU AI Act: kategorije rizika, zabranjene prakse i obaveze koje slede
  • AI pismenost iz člana 4: koje mere se očekuju od organizacija i kako se ovaj trening računa kao jedna od njih
  • Uvođenje AI u vašu organizaciju: izbor slučajeva upotrebe, postavljanje ograda, merenje vrednosti
  • Priprema za ispit i primeri pitanja

Preduslovi

Nema. Trening je zamišljen kao ulazna tačka i ne pretpostavlja tehničko predznanje ni programiranje. Ako ste ikada koristili AI čet-bot, imate svo potrebno iskustvo.

Ispit i sertifikacija

Trening vas priprema za ispit ICO Artificial Intelligence FND, koji dodeljuje ICO International Certification Organization GmbH, akreditovan sertifikacioni ispit, a ne potvrda o pohađanju.

  • 30 pitanja sa višestrukim izborom, jedna, više ili sve ponuđene opcije mogu biti tačne
  • 45 minuta, bez literature
  • Prag prolaznosti 60%
  • Polaže se onlajn, u vreme i na mestu po vašem izboru
  • Sertifikat važi 36 meseci

Ispit je dostupan kao opciona dopuna treninga. Sve detalje objavljuje telo koje dodeljuje sertifikat na ico-cert.org.

Šta je uključeno

  • Dva dana (12 sati) onlajn obuke uživo sa predavačem, na engleskom
  • Male grupe, nikad više od 16 polaznika
  • Kompletni materijali, vežbe i primeri ispitnih pitanja
  • Praktičan rad u svakom bloku: AI sisteme koristite tokom treninga, a ne posle njega
  • Potvrda o pohađanju; ispit ICO Artificial Intelligence FND kao opciona dopuna

Program kursa

Poglavlja prate zvanični ICO syllabus. Procenat je udeo poglavlja u ispitu.

How the course is organised: the four chapters below cover terms and definitions, an introduction to artificial intelligence, an introduction to AI management systems according to ISO/IEC 42001, and the EU AI Act with other laws on the use of AI. The ICO Artificial Intelligence FND exam has 30 questions in 45 minutes with a 60% pass mark; weights per chapter are shown next to each chapter.

Introduction

This chapter covers the essential terms and definitions crucial for understanding the field of Artificial Intelligence (AI) and related management systems. This foundational knowledge helps beginners grasp AI concepts, applications, and organizational integration.

Key Terms and Definitions

  • Artificial Intelligence (AI)

    Source: ISO/IEC 42001:2023 - A branch of computer science focused on creating machines capable of performing tasks that typically require human intelligence, such as problem-solving, learning, perception, decision-making, and understanding language.

    Artificial intelligence is intelligence demonstrated by machines, in contrast to the natural intelligence displayed by humans and animals.

  • Machine Learning (ML)

    Source: ISO/IEC 42001:2023 - A subset of AI involving algorithms and statistical models that enable computers to perform tasks without explicit instructions by identifying patterns and making inferences.

    Machine learning is a type of artificial intelligence that allows software applications to become more accurate at predicting outcomes without being explicitly programmed.

  • Neural Network

    Source: ISO/IEC 42001:2023 - Algorithms designed to simulate human brain functions by recognizing underlying relationships in data, forming the basis of deep learning.

    Neural networks are a series of algorithms that endeavor to recognize underlying relationships in a set of data through a process that mimics the way the human brain operates.

  • Deep Learning

    Source: ISO/IEC 42001:2023 - A specialized subset of machine learning that employs multi-layered neural networks to learn from extensive datasets by simulating human brain functions.

    Deep learning is part of a broader family of machine learning methods based on artificial neural networks with representation learning.

  • Natural Language Processing (NLP)

    Source: ISO/IEC 42001:2023 - A field of AI that enables machines to interpret and understand human languages, used in translation and chatbots.

    Natural language processing is a subfield of linguistics, computer science, and artificial intelligence concerned with the interactions between computers and human language.

  • Computer Vision

    Source: ISO/IEC 42001:2023 - A field of AI that enables machines to interpret and understand visual information from images and videos.

    Computer vision is an interdisciplinary scientific field that deals with how computers can be made to gain high-level understanding from digital images or videos.

  • Supervised Learning

    Source: ISO/IEC 42001:2023 - A machine learning approach where models make predictions from labeled training datasets.

    Supervised learning is the machine learning task of learning a function that maps an input to an output based on example input-output pairs.

  • Unsupervised Learning

    Source: ISO/IEC 42001:2023 - Machine learning using unlabeled data, allowing algorithms to independently identify patterns.

    Unsupervised learning is a type of algorithm that learns patterns from untagged data.

  • Reinforcement Learning

    Source: ISO/IEC 42001:2023 - A learning method where an agent learns to make decisions by performing actions within its environment to maximize cumulative rewards.

    Reinforcement learning is an area of machine learning concerned with how software agents ought to take actions in an environment to maximize cumulative reward.

  • Algorithm

    Source: ISO/IEC 42001:2023 - A step-by-step procedure for data processing that enables predictions or decisions.

    An algorithm is a finite sequence of well-defined, computer-implementable instructions.

  • Training Data

    Source: ISO/IEC 42001:2023 - The information and examples used to teach AI systems how to perform specific tasks.

    Training data is a dataset used to train machine learning algorithms.

  • Model

    Source: ISO/IEC 42001:2023 - The mathematical representation that an AI system uses to make predictions or decisions based on input data.

    A machine learning model is a file that has been trained to recognize certain types of patterns.

  • Edge AI

    Source: ISO/IEC 42001:2023 - AI processing that occurs locally on hardware devices rather than in cloud servers, reducing latency and improving privacy.

    Edge AI is the deployment of AI applications in devices throughout the physical world.

  • Explainable AI (XAI)

    Source: ISO/IEC 42001:2023 - AI systems designed to provide clear, understandable explanations of their decision-making processes.

    Explainable artificial intelligence (XAI) refers to methods and techniques in the application of AI such that the results can be understood by human experts.

  • Transfer Learning

    Source: ISO/IEC 42001:2023 - A technique where knowledge gained from training a model on one task is applied to a different but related task.

    Transfer learning is a machine learning method where a model developed for one task is reused as the starting point for a model on a second task.

  • Narrow AI (Weak AI)

    Source: ISO/IEC 42001:2023 - AI systems designed to perform specific, limited tasks - all current AI applications fall into this category.

    Weak artificial intelligence is artificial intelligence that implements a limited part of mind, focused on one narrow task.

  • General AI (AGI)

    Source: ISO/IEC 42001:2023 - Theoretical AI systems with human-level intelligence and understanding across all domains - not yet achieved.

    Artificial general intelligence (AGI) is the hypothetical ability of an intelligent agent to understand or learn any intellectual task that a human being can.

  • Super Intelligence

    Source: ISO/IEC 42001:2023 - Hypothetical future AI that would surpass human intelligence in all aspects.

    A superintelligence is a hypothetical agent that possesses intelligence far surpassing that of the brightest and most gifted human minds.

  • Symbolic AI

    Source: ISO/IEC 42001:2023 - AI approach using logical rules and symbols to represent knowledge and reasoning.

    Symbolic artificial intelligence is the term for the collection of all methods in AI research that are based on high-level symbolic representations.

  • Statistical AI

    Source: ISO/IEC 42001:2023 - AI approach using probability, statistics, and data analysis to make decisions and predictions.

    Statistical AI refers to artificial intelligence methods that use statistical techniques to enable machines to improve at tasks with experience.

  • Knowledge-Based AI

    Source: ISO/IEC 42001:2023 - AI systems that use explicit rules and expert knowledge to solve problems in specific domains.

    A knowledge-based system is a computer program that reasons and uses a knowledge base to solve complex problems.

  • Expert System

    Source: ISO/IEC 42001:2023 - AI designed to replicate the decision-making ability of human experts in specific fields.

    An expert system is a computer system that emulates the decision-making ability of a human expert.

  • Deterministic System

    Source: ISO/IEC 42001:2023 - AI systems that produce the same output for the same input every time, with no randomness.

    A deterministic system is a system in which no randomness is involved in the development of future states of the system.

  • Stochastic System

    Source: ISO/IEC 42001:2023 - AI systems that incorporate randomness or probability in their decision-making processes.

    A stochastic system has a random probability distribution or pattern that may be analyzed statistically but may not be predicted precisely.

  • Robotics

    Source: ISO/IEC 42001:2023 - Field combining AI with mechanical systems to create autonomous or semi-autonomous machines.

    Robotics is an interdisciplinary field that integrates computer science and engineering.

  • API (Application Programming Interface)

    Source: ISO/IEC 42001:2023 - A set of protocols and tools that allows different software applications to communicate with AI systems.

    An API is a computing interface that defines interactions between multiple software intermediaries.

  • Prompt

    Source: ISO/IEC 42001:2023 - The input text or instructions given to an AI system to generate a desired response or output.

    A prompt is a text input given to an AI system to elicit a specific response or output.

  • Fine-tuning

    Source: ISO/IEC 42001:2023 - The process of adjusting a pre-trained AI model to perform better on specific tasks or domains.

    Fine-tuning is a transfer learning approach that involves training a pre-trained model on a new, related task.

  • Data Mining

    Source: ISO/IEC 42001:2023 - The process of discovering patterns in large datasets.

    Data mining is the process of discovering patterns in large data sets involving methods at the intersection of machine learning, statistics, and database systems.

  • AI Bias

    Source: ISO/IEC 42001:2023 - Systematic errors in AI models arising from faulty or unbalanced training data, leading to discriminatory outcomes.

    Algorithmic bias describes systematic and repeatable errors in a computer system that create unfair outcomes.

  • Ethical AI

    Source: ISO/IEC 42001:2023 - Designing AI systems to be fair, transparent, and aligned with human values.

    AI ethics is a set of values, principles, and techniques that employ widely accepted standards to guide moral conduct in the development and use of AI systems.

  • Generative AI

    Source: ISO/IEC 42001:2023 - AI systems that create new content based on existing data and patterns.

    Generative artificial intelligence is AI capable of generating text, images, or other media, using generative models.

  • AI Chatbot

    Source: ISO/IEC 42001:2023 - Software that communicates with humans in natural language to perform tasks or provide information.

    A chatbot is a software application used to conduct an online chat conversation via text or text-to-speech.

  • Large Language Model (LLM)

    Source: ISO/IEC 42001:2023 - Extensive language models like GPT, capable of understanding and generating human-like text.

    A large language model is a language model consisting of a neural network with many parameters, trained on large quantities of unlabeled text.

  • Deep Fake

    Source: ISO/IEC 42001:2023 - AI-generated or manipulated content designed to appear real but is fabricated. (EU AI Act §1 a3)

    Deepfakes are synthetic media in which a person in an existing image or video is replaced with someone else's likeness.

  • AI Hallucination

    Source: ISO/IEC 42001:2023 - AI-generated incorrect or misleading information that lacks factual basis.

    An AI hallucination is a confident response by an AI that cannot be justified by its training data.

  • Governance

    Source: ISO/IEC 42001:2023 - The framework ensuring effective, ethical, and compliant organizational management.

    Governance is the way rules, norms and actions are structured, sustained, regulated and held accountable.

  • Compliance

    Source: ISO/IEC 42001:2023 - Adherence to legal, regulatory, and internal requirements.

    Compliance means conforming to a rule, such as a specification, policy, standard or law.

  • Management System

    Source: ISO/IEC 42001:2023 - A structured approach for achieving organizational objectives through planning, implementation, and improvement of processes.

    A management system is a set of policies, processes and procedures used by an organization.

  • AI System Impact Assessment

    Source: ISO/IEC 42001:2023 - Identifying and addressing AI's effects on individuals and society.

    An impact assessment is a process for identifying, predicting, and evaluating the potential impacts of a proposed project or development.

  • ISO/IEC 42001

    Source: ISO/IEC 42001:2023 - International standard specifying requirements for establishing, implementing, maintaining, and continually improving an AI management system.

    ISO/IEC 42001 is an international standard that specifies requirements for an artificial intelligence management system.

  • EU AI Act

    Source: ISO/IEC 42001:2023 - European Union regulation establishing rules for the development and use of AI systems.

    Source: European Commission - The AI Act is a European law on artificial intelligence – the first comprehensive AI law.

    The Artificial Intelligence Act is a European Union regulation on artificial intelligence in the European Union.

  • OECD AI Principles

    Source: ISO/IEC 42001:2023 - International guidelines promoting innovative and trustworthy AI that respects human rights and democratic values.

    Source: OECD.AI - The OECD AI Principles promote AI that is innovative and trustworthy and that respects human rights and democratic values.

    The OECD AI Principles are the first intergovernmental policy guidelines on artificial intelligence.

  • Information Security (CIA Triad)

    Source: ISO/IEC 42001:2023 - Ensuring the protection of information from unauthorized access (Confidentiality), maintaining data accuracy (Integrity), and availability.

    Information security is the practice of protecting information by mitigating information risks.

  • PDCA Lifecycle

    Source: ISO/IEC 42001:2023 - A continuous improvement cycle (Plan, Do,

    PDCA is an iterative four-step management method used in business for the control and continuous improvement of processes and products.

  • Statement of Applicability

    Source: ISO/IEC 42001:2023 - Documentation and justification of control measures implemented.

    A Statement of Applicability is a document that lists all the controls from a standard and whether they apply to an organization.

  • Corrective Action

    Source: ISO/IEC 42001:2023 - An action to address the causes of problems and prevent recurrence.

    Corrective action is a process designed to identify and correct problems that affect an employee's work performance.

  • Audit

    Source: ISO/IEC 42001:2023 - Systematic assessment to verify compliance with established criteria.

    Source: ISO 19011:2018 - Audit is a systematic, independent and documented process for obtaining objective evidence and evaluating it objectively.

    An audit is an independent examination of financial information of any entity whether profit oriented or not.

  • Model Context Protocol (MCP)

    Source: Anthropic Documentation - Model Context Protocol is an open protocol that enables seamless integration between AI applications and external data sources and tools.

    The Model Context Protocol (MCP) is an open standard, open-source framework introduced by Anthropic to standardize the way artificial intelligence (AI) models like large language models (LLMs) integrate and share data with external tools, systems, and data sources.

  • AI Agent

    An intelligent agent is an autonomous entity which acts, directing its activity towards achieving goals upon an environment using observation through sensors and consequent actuators.

1. Overview of Artificial Intelligence

  • Definition of AI and core concepts: Explanation of AI as machines performing tasks requiring human intelligence, such as reasoning, learning, perception, and decision-making.
  • Difference between traditional software and AI: Traditional software operates with predefined rules and logic, while AI systems learn and improve based on data exposure and experiences.
  • Machine Learning (ML), Deep Learning (DL), Knowledge-Based AI: ML includes algorithms that improve through data exposure (e.g., voice assistants); DL involves complex neural networks for advanced tasks like facial recognition; Knowledge-Based AI uses explicit rules, as seen in expert systems for medical diagnosis.

2. Evolution and Brief History of AI

  • Early AI developments (1950s–1980s): Milestones include the Turing Test by Alan Turing, Dartmouth Workshop (1956), and early expert systems.
    • 1997: IBM's Deep Blue defeats chess champion Garry Kasparov
    • 2011: IBM Watson wins Jeopardy!
    • 2012: AlexNet breakthrough in image recognition
    • 2016: AlphaGo defeats world Go champion Lee Sedol
  • AI winters and resurgence (1990s–present): Phases of reduced research due to unmet expectations followed by resurgence with neural networks and computational advancements.
  • Recent breakthroughs and widespread adoption: Significant innovations like AlphaGo beating human champions in Go, the release of ChatGPT for natural language understanding, and advancements in autonomous driving technologies.

3. Categories and Types of AI

  • Narrow vs. General vs. Super Intelligence: Narrow AI excels in specific tasks (e.g., virtual assistants); General AI aspires to human-level versatility; Super Intelligence envisions AI surpassing human capabilities comprehensively.
  • Symbolic vs. Statistical AI: Symbolic AI uses logic-based rules (e.g., early chess-playing programs); Statistical AI employs probabilistic methods and data-driven approaches.
  • Rule-based vs. Learning systems: Rule-based systems depend on explicit programming; learning systems use data-driven adaptations (e.g., recommendation engines).
  • Deterministic vs. Stochastic systems: Deterministic outcomes are predictable and consistent; stochastic systems incorporate randomness to explore multiple outcomes, typical in reinforcement learning scenarios.

4. Core AI Subfields and Techniques

  • Machine Learning: Includes supervised learning (spam detection), unsupervised learning (market segmentation), and reinforcement learning (AI in gaming).
  • Deep Learning and Neural Networks: Techniques involving multiple neural layers, critical in tasks like image and speech recognition.
  • Natural Language Processing (NLP): Enabling machines to comprehend and generate human language, such as chatbots and language translators.
  • Computer Vision: Technologies interpreting visual data, crucial for self-driving cars and facial recognition systems.
  • Robotics: Autonomous or semi-autonomous robots used in manufacturing, service industries, and exploration (e.g., drones, industrial robots).
  • Knowledge Representation and Reasoning: Structuring domain-specific knowledge to simulate human reasoning, as found in diagnostic expert systems.
  • Planning and Search: Algorithmic solutions for optimizing paths, schedules, or resources (e.g., GPS route optimization, task scheduling).
  • Emerging fields: Generative AI producing original content (e.g., ChatGPT); advanced data analytics for predictive business insights.

5. Major Application Areas of AI

  • Healthcare: AI-driven analysis of diagnostic imaging, personalized medicine.
  • Finance and Banking: AI for fraud detection, credit risk assessments, and algorithmic trading strategies.
  • Manufacturing and Industry: AI predictive maintenance to prevent equipment failures, automated quality control.
  • Transportation and Logistics: Autonomous driving systems, logistics optimization, and fleet management.
  • Education: Adaptive learning platforms providing personalized education experiences.
  • Security and Defense: Real-time surveillance analytics, cybersecurity threat detection.
  • Law and Government: AI-assisted document analysis, regulation compliance checks.
  • Entertainment and Media: AI-powered content recommendations, generation of virtual environments.

6. Benefits and Opportunities of AI

  • Increased productivity and efficiency: Automation reducing repetitive tasks and streamlining workflows.
  • Cost reduction and improved accuracy: Minimizing human error, enhancing reliability in data-driven processes.
  • Data-driven insights and decision support: Leveraging large datasets for improved forecasting and strategic planning.
  • Innovation and new capabilities: AI-driven development of novel products and services, transforming industries.

7. Challenges and Limitations of AI

  • Data dependency and quality issues: Necessity for vast, accurate datasets to ensure robust AI performance.
  • Computational Limitations:
    • High energy consumption of complex AI models
    • Infrastructure and hardware constraints
    • Scalability challenges for large-scale AI systems
  • Lack of common sense and contextual understanding: Difficulties in nuanced human-like reasoning and interpretation.
  • Transparency, bias, and ethical concerns:
    • Challenges in ensuring fairness and accountability
    • Algorithmic discrimination risks
    • Difficulty in creating truly unbiased AI systems
  • Regulatory and legal uncertainty: Navigating evolving regulations, such as data privacy laws.

8. Ethical and Societal Considerations

  • Fairness, bias, and transparency: Preventing biased outcomes and promoting transparency in AI operations.
  • Privacy, safety, and accountability: Ensuring ethical use of personal data and safe AI deployment.
  • Human autonomy and social impact: Balancing AI's capabilities with preserving human autonomy.
  • Malicious use prevention: Strategies to mitigate harmful applications of AI.

9. AI Governance, Regulations, and Standards

  • Internal governance frameworks: Implementing ISO/IEC 42001 standards for systematic AI management.
  • Regulations: Complying with frameworks like the EU AI Act.
  • Standards and ethical frameworks: Adhering to global guidelines such as OECD AI Principles.

10. Future Trends in AI

  • Progress towards general intelligence (AGI): Advancements toward versatile, general-purpose AI systems.
  • Human-AI collaboration and augmented intelligence: Enhancing human decision-making and productivity.
  • AI in edge computing and IoT: Expanding AI capabilities in decentralized environments.
  • Quantum computing impact on AI: Exploring quantum enhancements in AI processing and problem-solving.

11. RAG vs. Fine-Tuning: Modern Approaches to Optimizing AI Systems

  • Retrieval-Augmented Generation (RAG): Combines large language models with external knowledge sources to dynamically retrieve relevant information for each query.
  • Fine-Tuning: Adapts an existing model to specific use cases by training it further with domain-specific data.
  • Flexibility vs. Specialization: RAG provides flexible and up-to-date knowledge integration, while fine-tuning delivers tailored performance improvements.
  • Sources: Red Hat, IBM

12. Stages of Artificial Intelligence Development

  • Level 1 – Chatbots: AI capable of holding natural language conversations – e.g., today's ChatGPT systems.
  • Level 2 – Reasoners: AI able to solve complex problems with PhD-level reasoning capabilities. According to internal sources, OpenAI is nearing this stage.
  • Level 3 – Agents: AI that can autonomously execute tasks over time with planning and decision-making abilities.
  • Level 4 – Innovators: AI that supports creative processes, research, and the invention of novel ideas or tools.
  • Level 5 – Organizations: AI capable of running an entire company or organization independently.
  • Source: Inc.com – “5 Steps That OpenAI Thinks Will Lead to AI Running a Company” (July 2024)

13. Types of Artificial Intelligence

  • Reactive Machines: Respond to current inputs without storing past experiences (e.g., chess programs).
  • Limited Memory: Temporarily use past data for decision-making (e.g., self-driving cars).
  • Theory of Mind: Hypothetical AI that understands emotions, beliefs, and intentions of others.
  • Self-aware AI: Still theoretical; would have consciousness and a sense of self.
  • Artificial Narrow Intelligence (ANI): Specialized in one specific task (e.g., voice assistants, facial recognition).
  • Artificial General Intelligence (AGI): Would be capable of performing any intellectual task a human can do — not yet achieved.
  • Artificial Superintelligence (ASI): Would surpass human intelligence in all domains — purely theoretical.
  • Source: Viso.ai – Types of Artificial Intelligence

14. Types of Machine Learning

Artificial Intelligence (AI) uses different types of learning to find patterns and make decisions.

  • Supervised Learning: Learns from labeled examples: The model adjusts its prediction algorithm until it classifies new data correctly.
  • Unsupervised Learning: Works with unlabeled data to independently find patterns and structures.
  • Reinforcement Learning: Uses rewards and penalties in a dynamic environment to optimize strategies through trial and error.
  • Semi-supervised Learning: Combines both approaches with a few labeled and many unlabeled data samples.
  • Source: IBM – Supervised vs. Unsupervised Learning

15. Typical GPAI Application Areas

General Purpose AI (GPAI) describes AI systems that can be used flexibly across different tasks and contexts.

  • Text Creation and Translation: Automated generation and translation of content.
  • Prediction and Analysis: Recognizing patterns and making forecasts in large datasets.
  • Classification: Automatically categorizing information, e.g., spam filters or image recognition.
  • Generative Applications: Creating new text, images, or music.
  • Automation: Supporting processes through virtual assistants or chatbots.
  • Source: EU AI Act Article 3: Definitions
  • Source: EU AI Act Article 5: General-Purpose AI Models

Difference: GPAI vs. Generative AI

Term Description Example General Purpose AI (GPAI) Versatile AI that can perform multiple tasks in different contexts. Large language models that translate text, answer questions, and analyze data. Generative AI Creates new content like text, images, or music. ChatGPT, DALL·E, Midjourney.

16. Prompting and Variants

Prompting is the art of obtaining desired answers from an AI through targeted inputs (prompts). The quality of the prompts directly influences the accuracy and relevance of the results. Definition “Prompt”: An input in the form of a specific instruction or question given to an AI to receive a contextually appropriate output.

  • Zero-Shot Prompting: The model is asked to perform a task without any prior examples. Example: “Summarize the following text: …”
  • One-Shot Prompting: One example is provided for the model to follow. Example: “Here is how a summary looks: … Now summarize this new text: …”
  • Few-Shot Prompting: Several examples are given to help the model understand the desired pattern. This improves performance on more complex tasks.
  • Source: OpenAI – GPT Best Practices

17. Advanced Prompting – Types of Prompts

Advanced prompting includes different types of prompts that help achieve more precise and useful results. The clearer the prompt, the better the AI can respond.

  • Standard Prompt: A simple instruction or question to receive a clear answer.
  • Reasoning Prompt: Asks the AI to show its steps or reasoning, e.g., “Please explain your answer step by step.”
  • Deep Research Prompt: A detailed prompt with context, source hints, or multiple sub-questions for in-depth analysis.
  • Custom-GPT or Agent System Prompt: Specific system prompts that define roles, rules, or capabilities for a Custom GPT or agent system.
  • Source: OpenAI – GPT Best Practices

18. Neural Networks & Training

Neural networks form the foundation of modern artificial intelligence. They consist of layers of neurons that process data step by step and identify patterns. Through training, the connections between neurons are adjusted to improve predictions or classifications.

Training means adjusting the neuron connections of a neural network so that it produces increasingly better outputs. Technically, the parameters (weights, thresholds) are set using training data so that input values X map as closely as possible to output values Y. The goal is an optimized set of parameters that contains all learned information and enables the model to deliver useful results.

  • Architecture: A typical neural network consists of an input layer, several hidden layers, and an output layer.
  • Training: During training, the error between prediction and actual result is calculated and minimized using backpropagation.
  • Activation Functions: Functions like ReLU or Sigmoid determine whether a neuron is activated.
  • Learning Rate: Determines how much the weights are adjusted during training.
  • Source: IBM – What is a Neural Network?

Management system in general

A management system is a structured framework that supports organizations in systematically achieving specific objectives. It comprises guidelines, procedures, processes, responsibilities and resources. The aim is to ensure the effectiveness and efficiency of the organization in a specific area - such as quality, information security or here: Artificial Intelligence - and to continuously improve it.

Management systems are characterized by the PDCA cycle (Plan-Do-Check-Act):

  • Plan: define objectives, resources and processes.
  • Do: Implementation of the planned measures.
  • Check: Review the results (e.g. through internal audits).
  • Act: Correction and improvement.

Summary of the Seven ISO 9001 Quality Management Principles

  • Customer Orientation
    Identifying and meeting customer requirements to build trust and ensure long-term satisfaction.
  • Leadership
    Establishing a clear vision and direction, motivating leaders, and empowering employees to actively shape quality objectives.
  • Engagement of People
    Leveraging the competence, motivation, and knowledge of all employees through participation to foster ownership and gather improvement ideas.
  • Process-Oriented Approach
    Managing activities and resources as interconnected processes to enhance consistency, efficiency, and transparency throughout the value chain.
  • Improvement
    Systematically optimizing the organization, its products, and processes through goal setting, action planning, and regular review.
  • Evidence-Based Decision Making
    Basing decisions on reliable data and facts, conducting analyses, and monitoring key performance indicators.
  • Relationship Management
    Maintaining partnerships with suppliers and stakeholders to ensure sustainable supply and mutual benefit.

Artificial Intelligence Management System (AIMS)

An Artificial Intelligence Management System (AIMS) in accordance with ISO/IEC 42001 is a management system specifically for the use and control of artificial intelligence (AI). It supports organizations in identifying and managing the opportunities and risks of AI systems and ensuring the responsible use of AI.

Key aspects of an AIMS are

  • Responsibility and leadership: defining leadership roles, responsibilities and communication channels.
  • Risk-based approach: Identification and assessment of risks associated with AI systems.
  • Documentation and control: Creation and maintenance of relevant documents, e.g. guidelines, procedural instructions and records.
  • Internal audits: Regular reviews by the organization itself to ensure compliance and effectiveness of the system.
  • Management review: Regular assessment by top management to ensure the adequacy and effectiveness of the system.
  • Continuous improvement: Identification of weaknesses and derivation of improvement measures.

Important elements in the AIMS according to ISO/IEC 42001

Leadership
  • Top management is responsible for the implementation and continuous improvement of the AIMS.
  • Clear definition of roles, responsibilities and communication channels.
  • Promotion of an organization-wide "AIMS culture", e.g. through training and awareness campaigns.
Planning
  • Identification of risks and opportunities in connection with AI.
  • Setting goals for the safe and responsible use of AI.
  • Definition of resources and timeframe for implementation.
Support
  • Provision of necessary resources, such as qualified personnel, technology and infrastructure.
  • Documentation of guidelines, processes and procedures to ensure consistent workflows.
Operation
  • Implementing the planned measures, monitoring processes and ensuring compliance.
  • Dealing with changes (change management) and incident management for AI systems.
Performance evaluation
  • Internal audits carried out by the organization itself to evaluate the effectiveness and appropriateness of the AIMS.
  • External audits are not part of the standard requirements, but can be carried out voluntarily (e.g. for certification).
  • Management reviews are to be carried out by top management in order to regularly assess overall performance and identify potential for improvement.
Improvement
  • Recognize and implement corrective and preventive actions.
  • Promote a culture of continuous improvement.

Overview of Annex A (Normative Controls)

Annex A of ISO/IEC 42001 lists control objectives and controls that are relevant to an AIMS. These controls are not all mandatory, but serve as reference and guidance. Organizations can develop their own controls as long as they meet the requirements of the standard.

Annex A is divided into the following sections:

  • A.1 General: Basic controls to support the system objectives.
  • A.2 Guidelines: Specifications for creating and implementing policies for AI.
  • A.3 Internal organization: Responsibilities, roles and resources.
  • A.4 Resources for AI systems: Identification and control of resources.
  • A.5 Risk management: Processes for identifying and managing AI-related risks.

Overview of Annex B (Implementation Guide)

Annex B provides practical guidance on implementing the controls in Annex A. It explains:

  • How the individual controls can be implemented in practice, e.g. through specific processes or technological solutions.
  • Examples of best practices and adaptations to the specific organizational context.
  • Guidance on documentation and verification.

Annex B is particularly helpful for organizations that want to set up an AIMS or adapt existing processes. It serves as a practical guide.

Summary of the EU AI Act

The EU AI Act is a comprehensive regulatory framework of the European Union for Artificial Intelligence (AI). The aim is to make the development and use of AI safe, transparent, and ethical. It is based on a risk-based approach and contains the following key points:

Classification of AI systems according to risk levels

  1. Prohibited AI: Systems with unacceptable risks (e.g. systematic evaluation of socially unacceptable behavior, manipulative practices).
  2. High-risk AI: Areas of application with high potential impact, e.g.:
    • Medical diagnostics
    • Critical infrastructures
    • Evaluation of work performance
  3. Limited-risk AI: AI systems with limited risk, but which must meet transparency requirements (e.g. chatbots).
  4. Minimal risk AI: AI systems with no or low risk (e.g. AI-powered games).

Central requirements for AI

  • Transparency: Users must be informed about how it works and how it is used.
  • Data management: Ensuring quality, accuracy and fairness of data.
  • Monitoring: People must be able to maintain control and monitor AI systems.
  • Compliance assessment: Manufacturers must ensure that systems comply with EU regulations.

Aim of the law

To promote innovation while minimizing risks to citizens' rights and security.

Article 4 of the EU AI Act

Article 4 of the EU AI Act, as amended by Regulation (EU) 2026/1744, asks providers and deployers of AI systems to take measures that support the AI literacy of their staff and of other persons who operate or use AI systems on their behalf, taking into account their role, their prior knowledge and the context in which the systems are used. It does not define a required level and it does not ask for a formal proof; it does expect organisations to be able to show which measures they took. Training with a defined syllabus and an externally set exam is one such measure.

The exact wording is in the consolidated text of Regulation (EU) 2024/1689 on EUR-Lex (see Sources).

Summary of Regulated AI in the EU AI Act

Regulated AI includes all systems that are either prohibited, classified as high risk, or limited risk and thus subject to the requirements of the Act. The aim is to minimize risks for citizens, companies, and institutions while promoting innovation in AI.


Other relevant laws in the EU

1. General Data Protection Regulation (GDPR)

  • Regulates the handling of personal data.
  • Relevant for AI that processes personal data:
    • Transparency obligations
    • Rights of data subjects: (e.g. information, erasure, automated decisions)
    • Data protection impact assessment for high-risk systems.

2. Product liability law

  • Addition to the AI Act: manufacturers can be held liable if their AI systems cause damage.
  • Planned adaptation of the Product Liability Directive to include AI.

3. Labor law

  • Use of AI for monitoring or making decisions about employees (e.g. hiring) must be legally sound.

4. Copyright

  • Clarification of property rights for AI-generated content in accordance with copyright law.

Sources

Akreditacije i sertifikacije

Ovaj trening vas priprema za sledeće sertifikacione šeme.

Vaš put do sertifikacije

Vaši treneri

Predstojeći događaji

Česta pitanja

Ispit je opciona dopuna. Možete ga rezervisati zajedno sa treningom ili odlučiti posle prvog dana, cena je ista.

Ne. Trening ne pretpostavlja ni programiranje ni prethodno iskustvo sa AI. Ako ste ikada koristili AI čet-bot, imate svo potrebno iskustvo.

Uživo i onlajn, sa predavačem, dva dana uključujući pauze i praktične blokove. Nastava je na engleskom.

Pitajte AI agenta o ovom kursu
Termini, cena, sertifikat, za koga je kurs — agent poznaje ovaj kurs i odgovara odmah.
Pozovite AI agenta

Ili pozovite telefonom:011 635 5918

Imate pitanje o ovom treningu?
Pitajte o terminima, obuci za tim ili bilo čemu drugom — odgovaramo u roku od jednog radnog dana.

Srodni treninzi

AI Professional: tehnička implementacija i automatizacija (ICO AI PRO)

Trodnevni inženjerski kurs onlajn uživo: fino podešavanje, generisanje potpomognuto pretragom, AI agenti, bezbednost LLM-ova i produkciono postavljanje. Priprema vas za ispit ICO Artificial Intelligence PRO. Za programere i data profesionalce sa osnovnim znanjem Pythona.

Napredni nivoOnlajn