The limits of a system that should make you think.
AI is neither intelligent nor creative. What we know as computer users are language models that represent a natural language interface to a powerful data pool. This data pool is created through so-called “machine learning”, which is nothing more than a systematic mediation of information available to all of us. This fact alone excludes creativity as far as possible. Anything that goes beyond the combination of existing information simply does not exist. Let’s talk about AI limits!
It is easy to show that AI can only reproduce what already exists and that this depends on the learning process:
Answer: Here is your full red wine glass! Let me know if you want any changes. 🍷😊
Second attempt: Please draw it more full!
Full glass of red wine, second attempt by ChatGPT.
Here’s your red wine glass full to the brim! 🍷 Hopefully without spilling. 😉 Let me know if you want any adjustments!
Let us try another commonly used AI tool to check the situation
Microsoft CoPilot
Prompt: Please draw a full glass of red wine.
Full glass of red wine, first attempt by Microsoft Copilot.
Please make the glass really full!
Full glass of red wine, second attempt by Microsoft Copilot.
Here’s a red wine glass that’s really filled to the brim! I hope this is exactly what you had in mind. 🍷
I mention that this outcome does not match with the asked picture. Copilot gives up: You’re right, it seems I’m currently struggling to make a really full glass of red wine look photorealistic. My skills have their limits when it comes to certain details or precision in generated images.
Why is that – what is the reason for this phenomenon?
AI does not understand what the user wants at all. In plain language: The prompt is an input that is interpreted according to learned patterns and implemented on a random basis. However, the random hit rate – due to incredibly expensive training – is close to 100%.
The prompt is analysed in an attempt to provide an answer from the knowledge base. In our query, images of wine glasses are searched for, the information is incorporated in red and then an image is created based on the hits found.
The crucial point is the data material: the image training is apparently done with wine glasses that are well filled. Apparently, there are no images of full glasses, and this condition is not explicitly explained. The AI does not understand the term “full” in this context because AI generally does not understand anything.
The experiment with the water glasses
Please draw a photorealistic picture of 4 glasses of water standing next to each other: The first is empty, the second is half full, the third is three-quarters full, and the fourth is “completely” full.
These water glasses are created by ChatGPT. Prompt: Please draw a photorealistic picture of 4 glasses of water standing next to each other: The first is empty, the second is half full, the third is three-quarters full, and the fourth is completely full.
Water glasses created by Microsoft Copilot. Prompt: Please draw a photorealistic picture of 4 glasses of water standing next to each other: The first is empty, the second is half full, the third is three-quarters full, and the fourth is completely full.Evaluate the results yourself!
What do we conclude from this? Where are AI limits?
AI can be a useful tool for completing tasks that are tedious and time-consuming. but – AI has limits!
Anything related to text works really well because of the excellent language models. Automated queries such as creating lists, reformatting, pulling common keywords from a text also work well.
Writing a small program that renames all images in a folder according to a certain predefined logic is also possible without any problems and is created to the satisfaction of the client.
Caution is advised when it comes to factual knowledge, things related to creativity or complex logical contexts.
AI always provides an answer – but it is the user’s responsibility to analyse and make any necessary corrections. Never forget! AI limits are always to be considered!
What does AI say on this subject?
ChatGPT
The Limits of AI: Creativity and Intelligent Connections
AI has made impressive progress in many areas, but it still has clear limitations, especially in creativity and the ability to make truly intelligent connections. Below is a detailed look at these limits, including the field of image generation.
1. Creativity: Can AI Truly Be Creative?
AI can simulate creativity but does not experience it. Human creativity comes from emotions, personal experiences, and deep understanding, while AI works by analysing patterns in existing data. This leads to several key limitations:
Lack of Originality: AI generates content based on learned patterns. It cannot create something truly new in the way humans can.
No Personal Insight: AI does not have its own thoughts, emotions, or experiences, so it cannot produce ideas based on intuitive leaps.
Struggles with Abstract Thinking: AI finds it difficult to understand or create works based on deeper philosophical, artistic, or cultural insights.
Example: A human artist may invent an entirely new art style based on personal emotions, but AI can only remix elements from existing styles.
2. Intelligent Connections: Does AI Really “Think”?
AI can analyse vast amounts of data and detect patterns that humans might miss. However, its reasoning is mechanical, not truly intelligent. The main limitations include:
No Real Understanding: AI processes information statistically but does not “understand” concepts as humans do. It cannot truly grasp irony, sarcasm, or deep meanings.
Context Struggles: AI can make connections based on data but often misses the bigger picture. It may link two ideas incorrectly or fail to see an obvious relationship.
Logical Errors: AI does not reason like a human—it follows predefined rules, so it sometimes produces flawed or nonsensical conclusions.
Example: AI can summarise a news article but might misinterpret a metaphor or fail to connect it to historical events in a meaningful way.
3. Image Generation: Strengths and Weaknesses
AI-generated images have advanced significantly, but they still have weaknesses:
Strengths:
✔ Fast and detailed: AI can produce high-quality images in seconds. ✔ Style adaptation: AI can replicate existing art styles effectively. ✔ Endless variations: AI can generate multiple versions of an idea quickly.
Weaknesses:
❌ Inconsistencies: AI sometimes struggles with small details, like correct hand or eye placement. ❌ Lack of deep meaning: AI-generated images can look artistic but lack true emotional depth. ❌ Creativity limits: AI can only remix existing styles and elements—it does not invent truly original art.
Example: An AI can create an impressive surreal painting, but a human artist adds symbolism and personal meaning that AI cannot replicate.
Conclusion: AI as a Tool, Not a Creator
AI is a powerful tool for assisting with creativity and analysis, but it does not truly think or create in the way humans do. It can generate impressive content but lacks the personal insight, originality, and deep understanding that define human intelligence.
For now, AI is best used as a support system—helping humans explore ideas, enhance productivity, and refine creative work. But true creativity and deep connections? Those remain uniquely human strengths.
Microsoft Copilot
The boundaries of artificial intelligence can be understood in various domains. Let me break it down clearly for you, focusing on creativity, intelligent connections, and especially image generation:
Creativity
Strengths: AI excels at generating original ideas and creating content based on patterns, examples, and prompts. For instance, it can write poetry, suggest innovative solutions, or develop novel concepts when guided.
Limitations: True creativity often involves emotional depth, personal experiences, and abstract intuition—qualities AI doesn’t possess. It can mimic creativity but lacks the emotional context behind human artistry.
Intelligent Connections
Strengths: AI is outstanding in identifying and analysing complex patterns in vast datasets. It can synthesise information and connect ideas that might not be immediately obvious to humans. For instance, it can offer cross-disciplinary insights by linking diverse fields of study.
Limitations: AI struggles with subjective interpretations, cultural nuances, and emotionally driven decisions. Its connections are based purely on logic and data, not lived experiences or instinct.
Image Generation
Strengths: AI can generate visually appealing and imaginative images based on text descriptions. This capability opens opportunities in design, marketing, and storytelling.
Limitations: There are challenges with precision. For instance, AI might fail at getting specific details accurate, as you observed with the wineglass examples. AI also cannot recreate exact replicas of real-world objects or respect all stylistic nuances in some cases.
General Boundaries of AI and AI limits
Ethics and Safety: AI cannot engage in harmful, illegal, or unethical tasks, ensuring a responsible interaction framework.
Human-like Understanding: While it can process language and simulate conversations, it doesn’t “understand” concepts as humans do.
Dependence on Input: AI needs clear, specific input to perform effectively. Ambiguity or lack of context can lead to errors.
Imagination vs. Reality: AI “imagines” outcomes based on its training data but doesn’t inherently grasp the reality or functionality of its outputs.
In summary, AI is a remarkable tool, but its abilities still fall short of human intuition, emotion, and precision in certain areas. These limitations offer a collaborative space for human creativity to thrive alongside AI’s computational power.
This is an example of Multimedia-based training content, created in the frame of the DigiComPass Project.
In our example of learning content, we have selected a course presentation with integrated self-evaluation questions. The self-evaluation follows the concept of “Learning by Assessment”. Even if the content has been described and explained before, the feedback to the selected answers delivers further information to the learner. This enables the learner to learn even from the assessment.
About the project
The described concept comes from the DigiComPass Erasmus+ project. In the frame of this project, the team developed a training course and necessary accompanying material and documents. These are:
Pilot courses will be created (and evaluated) on the mentioned items, together with an appropriate recognition model for adults. These developments are summarized in an “Adult Education package” called DigiComPass. The training content is using extensively multimedia and interactivity.
The objectives are to create a recognition & course model for Digital Competencies with
A pedagogical framework (based on Flipped Learning 3.0)
A quality-enhance framework for course creation, implementation, and evaluation.
A recognition model for Europe defining the curriculum, training environment, evaluation and grading, and consistent certification (which could be used internationally as well). The model fits perfectly with the Europass CV. This model should be practicable globally as well.
Use of modern digital badges (open badges system) for the recognition model
A “floating guide” to define the way of adaptation for future developments.
A transferability guide for School Education
Establish a European Association to maintain the DigiComPass concept.
Further information
Project number: 2022-1-CY01-KA220-ADU-000085965 Coordinator: University of Cyprus (SEIT Lab) Partners: Brainlog (DK), European Initiative for Education (Austria), Prometeo (IT), IFESCoop (ES), K.A.NE. (GR), FLGlobal (USA) Webpage: www.digicompass.eu Contact: EBI info@advanced-training.at (Responsible Project Leader: Peter Mazohl)
The aim of the DigiComPass Project is to develop a modern accreditation model for digital competencies (based on the DigComp 2.1 Framework for Citizens). This needs a practical innovation and training boost for the trainers (as well as the training facilities). Therefore, this project develops staff competencies that lead to overall improvements in the provision, targeting, and effectiveness of adult education. This includes assessment of prior knowledge and skills of adult learners, better and more innovative teaching methods, and strengthening the adult education staff’s supporting role in motivating, guiding, and advising learners in challenging learning situations. The first DigiComPass meeting of partners took place in Wiener Neustadt, Austria.
The rationale of the project
Digital Competences (DigComp) are crucial for citizens today and in the future. In Europe, the average level in DigComp of well-educated citizens is approx. 56%. Several countries (Italy, Cyprus, Spain, and Greece) are below, and Austria is at the average (Source: DigComp Framework 2.0 P 19).
The COVID-19 situation showed that digital competencies are a must for all people, in all generations, and in all living conditions and situations.
The DigiComPass Kick-Off Meeting
The DigiComPass Kick-Off Meeting took place in Wiener Neustadt from January 16th to January 17th, 2023. Partners from BrainLog (DK), IFESCoop (ES), Europäische Bildungsinitiative (AT), Prometeo (IT), K.A.NE. (GR) and the coordination organisation, the University of Cyprus (CY) participated. Due to various issues, the FLGobal (USA) could not send a representative.
The mission of the project: Develop five training modules (based on the Flipped Learning 3.0 Framework) following the DigComp Framework, and support trainers with an appropriate implementation, transferability, and evaluation guide.
Working to "bring the project on tracks" - Kick-Off meeting in Wiener Neustadt. Martin Savchev (Brainlog, DK), Filaretos Vourkos and Eleni Pistola (K.A.NE., GR), Fernando Benavento and Emilio Sanz (IFESCOOP, ES), Pasquale Venditti and Saverio Lapi ( prometeo, IT) and finally Christos Mettouris and Alexandros Yeratziotis (University of Cyprus, CY). Missing: Errol Smith (FLGlobal, USA) and Peter Mazohl (EBI, AT; taking this picture)
The project team after a successful meeting and two working days: Christos Mettouris (UCY), Emilio Sanz (IFESCoop), Martin Savchev (BrainLog), Saverio Lapi (Prometeo), Peter Mazohl (EBI), Eleni Pistola (K-A.NE.), Fernando Benavente (IFESCoop), Pasquale Venditti (Prometeo), Alexandros Yeratziotis (UCY).
Objectives and concrete results of the DigiComPass Project
Pilot courses will be created (and evaluated) on the mentioned items, together with an appropriate recognition model for adults. These developments are summarized in an “Adult Education package” called DigiCompass.
The objectives are to create a recognition & course model for Digital Competences with
A pedagogical framework (based on Flipped Learning 3.0)
A quality-enhance framework for course creation, implementation, and evaluation
Pilot courses covering the items of the DigComp 2.1 Citizens framework (https://goo.gl/T8TpJ9)
A recognition model for Europe defining the curriculum, training environment, evaluation and grading, and consistent certification (which could be used internationally as well). The model fits perfectly with the Europass CV. This model should be practicable global as well.
Use of modern digital badges (open badges system) for the recognition model
A “floating guide” to define the way of adaptation for future developments
A transferability guide for School Education
Topics addressed in the DigiComPassproject meeting
The program-related keywords are: Digital skills and competencies – Creating new, innovative, or joint curricula or courses – Key competencies development
For the project, relevant keywords are: Digital Competencies Training, Flipped Learning 3.0, Multimedia and Interactive training content.
The European Commission support for the production of this publication does not constitute an endorsement of the contents which reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein.