The benefits and harms of AI / Why we only use AI very little.

"We are now doing something with AI," we didn't just hear once in 2024. Almost as often, our counterparts were almost a matter of course that we also use artificial intelligence for our tasks — especially those in the field of individualization and personalization. This is by no means so. For reasons.

People are convinced regularly, when we show our projects, that we have used artificial intelligence to create these results. But: In most cases, no AI was actually used. Especially not in the curating and generation of content. Occasionally we expand a picture that has the wrong format; Occasionally we optimize the resolution of an image or improve it in terms of contrasts or brightness. We generate initial translations of our content. In our running diary for the Drupa 2024, we generated individual videos including an individual audio using text-to-speech.

It will not surprise anyone if we say that artificial intelligence as a topic, of course, has accompanied us throughout the year and more than once. A few more readers may be astonishing that we do not share the hype about artificial intelligence, although we are very facing the topics of digitization and automation: many people seem to think that AI will do miracles.

In fact, we dealt intensively with artificial intelligence, read and experimented about them and we too can not escape a certain curiosity whether the possibilities; Experience and findings have also shown us that depending on the task, we achieve at least as well in other ways and/or that the price we are too high for the use of artificial intelligence to justify the results.

But now a step back.

What are we actually talking about when we speak of artificial intelligence?

There is no uniform definition for artificial intelligence as there is no the an artificial intelligence. The term is usually specified context -related.

The EU regulation of the EU, which came into force in 2024, defines a "AI system" in Article 3-Definitions-as follows: "For the purposes of this regulation, the expression [...] AI system A machine-based system, which is designed for a different degree of autonomous operation and that can be adaptable after its operation and that derives from the received input for explicit or implicit goals, such as expenses such as predictions, content, recommendations or decisions, the physical or virtual Can influence environments."

A division of the KIS is, for example, into "strong AI" and "weak AI" or into different AI types, of which the types "Reactive Machines" and "Limited Memory" are in use, while others ("Theory of Mind", "Self-awareness") are still future music.

Artificial intelligence can also be divided into different areas. Sub -areas that we have tried and tested are

• Image processing, optimizations and supplements (Image Enhancement) (Adobe Photoshop AI, Topazlabs Photo Ai, Leonardo.ai, Pictid, Remove.bg, Viesus)
• Image-to-text conversion (Astica Ai, Google Vision AI)
• Language models (LLMS) and word processing (ChatGPT, ClaudAI, LanguageTool, SUMM AI)
• Text-to-image processing (Midjourney, Dall · E 3, Stable Diffusion)
• Text-to-language processing (Google Text-to-Speech AI, Elevenlabs AI)
• Tools for code optimization in programming (Github Autopilot, Jetbrain Ai Assistant)
• AI search engines (Perplexity AI)
• Translation tools (DeepL, ChatGPT)

Almost exclusively large Big Tech groups are responsible for the AI ​​applications, the seat of which is located in the United States.

Operations of AI in communication

Like every invention, artificial intelligence is a tool that can be used for "good" or "bad" purposes depending on the intention and perspective.

"Good" from our communication point of view are products that offer the recipient: inside and users: offer relevant information on the inside and address them in the best possible way-regardless of the type of product it is. Personalized images and texts, predesions and other opportunities for curating and generation of content seem to be exactly the right tools.

So is artificial intelligence the chance for needs -based printing products?

In fact, thanks to artificial intelligence and on the basis of information about people or groups of people, selections can be made and pictures and texts can be generated on the basis of prompt, which then become individualized products via layout automation. And you might think that with AI the egg -laying wool milk sow is standing in front of us.

In fact, this thousandsassa animal also brings with it a whole swing of challenges and problems, so that from our point of view, in every application, it must be critically questioned whether the use of AI actually makes sense or not.

The price we pay

The price we pay for the use of artificial intelligence can be quantified on several levels.

Social consequences

Probably the best -known weakness of generative artificial intelligence is the appearance of hallucinations in image and text. By using probabilities instead of facts, AI language models, for example, spread false information with convincing formulations. In some cases, this is obviously and sometimes amusing, in other cases annoying or even dangerous. In the speed in which AI produces information, she has the potential to make out most of the available information and thus shake its reliability and trust and our education.

AI is able to filter information from a large amount of information, both as well as to generate new information. There are risks in both:

• The curating of information through AI decisions takes place on a not fully understandable basis so that no one can evaluate how this selection comes about.
• With the scalability by AI, it can be accompanied by the fact that recipients: inside with large amounts of information are flooded and the sheer quantity means that they can no longer be evaluated and can be distinguished in important or unimportant.

AI generated results reproduce and reinforce prejudices, stereotypes, discrimination and social injustice. The reason for this lies in the training data.

Only a few people understand — and then not completely — why AI delivers which results and how to enable other results. In the incomprehension of the processes, on the one hand, there is a hidden risk that a troubleshooting will become impossible, on the other hand, it enables manipulation by people who have a greater understanding of AI.

Especially in marketing, which has always been manipulative features, AI and the possible scaling strategies are driven to the extreme and human weaknesses are used.

Financial consequences

While free versions of AI tools suggest that good results can be generated with artificial intelligence with little financial effort, the opposite is the case in many cases.

The training of your own model requires high initial costs, which usually only pay off if there is extreme scalability. The scalability in turn increases the problems of ecological (see "ecological consequences").

The new challenges are often not covered by the existing employees: so that further personnel costs may need to be added to the high initial costs.

The prompt, i.e. the command in order to generate pictures and texts that meet both the qualitative claim and are reproducible for other products in your style, must be learned time-consuming ("Prompt Engineering"). In addition, for good results, it is usually necessary to take several attempts and to revise texts or pictures in several steps, possibly even with several applications in succession.

Consequences for the quality and our quality standards

The high scalability leads to foregoing the quality of the products to be generated, since the investment can be higher in quality than that for the generation, production and delivery and thus rather (again) high litter losses.

In addition, loss of quality caused by self -reference. So many texts were generated with artificial intelligence within a short time and have been incorporated into KIS's continued training session that subsequently lost the generated texts of quality ("text trial").

Ecological consequences

The use of artificial intelligence creates high energy and water needs, which are reinforced by the high training needs or the many attempts to refine on the one hand and the high degree of scaling on the other. The greatest consumption arises from the high cooling requirement of the server. Artificial intelligence is in contrast to the global efforts to save energy, the responsible handling of water and the limitation of the inheritance.

Monopolization and dependency

The use of AI supports few dominant companies that have a kind of monopoly position, and by exploiting their existing dominance and by targeting buying AI startups, the development of artificial intelligence. In doing so, you are all the more strengthening your monopoly-like market positions with all the associated disadvantages. These companies also use partly questionable business practices (lining up copyright for the training data, fake-AI projects such as Amazon's "Just Walk Out"; underpaid clickworkers, etc.). Few earn a lot of money and strengthen their own power.

The alternatives, which are hardly existing in the respective sub -area, lead to a dependency on a few companies and their likelihood.

Incidentally, there is no AI company a profitable business model today.

So what to do?

We have decided not to use artificial intelligence to be used in the curating or the creation of data to be used. In experiments, she has not helped us to add any added value that would be appropriate to the price that we as a company pay for it in the long term.

Instead, we continue to refine the automated approaches that we use to provide recipients: inside and users: to provide information tailored to the inside based on their own high-quality data. This does not result in the results that arise using artificial intelligence. But they don't have to. The possibilities of variable concepts without AI have not yet been exhausted and can do with significantly fewer and lower disadvantages.

How does automation with the help of regular curating differ from automation using artificial intelligence?

Automation performs predefined tasks, reduces manual interventions and increases efficiency. This makes it much faster than with human use to create many different variants of a product.

In order to decide which data is used, there are different types of curating: the control via a trained artificial intelligence or — our way — a definition of comprehensible and adjustable rules at any time?

We decide on the combination of several rules about the complexity of projects and are at any time able to adapt these rules and to refine results in iterative processes. Every desired change must be entered into the application.

The curating via a AI is via training data that the AI ​​uses to make connections independently and to transfer it. In addition to the desired contexts, both no longer up-to-date and unwanted can arise, which due to the difficult traceability can only be excluded with great effort and not reliably.

And how about the use of generative AI results in layout automation?

For us, which we also focus on fully automation of our layouts, the use of generative components is not an issue. In fully automation, we must be able to ensure that the data used is flawless and do not land six fingers or textual false information in our products that are no longer checked individually. Therefore, each of our products is based on a data pool, the content of which has been individually checked for their quality and suitability. The individual texts or pictures for this can theoretically be created with AI.

There are also products with just-in-time generative content-but so far our quality has not convinced us. For example, there is a children's book personalized on the basis of a photo, in which, in the example shown, the child has hair in each picture of different lengths and the hangover, who also occurs as the main character in history, is different in every picture.

Our conclusion

So far, we have not been able to determine any added value for our applications that are superior to our previous automation solutions without AI and for which the associated disadvantages would be worthwhile. On the contrary: since the results of complex ideas are usually — still? -are not satisfactory, but many projects should quickly carry some AI stamp, simple, effective individualizations are often used without added value. Some — even excellent — projects only exist as — convincingly presented — idea.

We will keep an eye on the developments around AI or their respective applications. At the same time, however, we continue to use reliable and comprehensible rules to curate content. Hallucinations or the reinforcement of advantages and discrimination have as little space as solutions that lead our global efforts to limit global warming or absurd.

We hope that the recipients: inside and users: Inside our products, continue to recognize their added value and look forward to the diverse projects that may still come there.