Artificial Intelligence Takes Over - How AI Will Change Our Working World

13. January 2023 - from Till Könneker

This article is not by an AI expert, rather it is a loose report on my experiences and insights from the last months dealing with AI systems more extensively. Let's go down the rabbit hole.

Is artificial intelligence (AI) one of the most useful inventions of mankind or just an entertaining gimmick? Both.

Early on, I fantasized about an intuitive computer interface, one in which we operate our daily tools through our thoughts, creating entire images with our imagination. Imagine - an operating system that anticipates accidents like when you unintentionally delete a file by pressing the wrong button. What if an AI could prevent such accidents just by knowing your thoughts? We're far from there yet, nonetheless artificial intelligence has made increasingly impressive progress in recent years. Currently, what makes it so exciting is that it's accessible to everyone.

Flamingo illustration created in Midjourney
Flamingo illustration created in Midjourney
Flamingo illustration created in Midjourney
Flamingo on a tank - Test with Midjourney

Until now, AIs have been active in the background" recognizing faces on smartphones, teaching cars to drive autonomously, convicting insurance fraudsters or recognising music. AIs are increasingly coming into the foreground, with tools and services which enable use in all areas of life. Despite my fascination, I am also somewhat overwhelmed by this. It is not yet clear how disruptive this technology will be and what it will do to us.

The most important thing we humans need to learn when using AI is what it can and cannot do. AI is not really intelligent, it is actually very limited in understanding us properly. There is a lot of room for misunderstanding, but also endless potential.

In my opinion, AI systems currently offer the greatest benefits for scientific and medical tasks. Here, large amounts of scientific data can be analysed in a short time to gain new insights and discoveries. In prognosis and diagnosis support, for example, an AI system can analyse X-rays or MRIs to detect diseases at an early stage.

EPFL, synapses in breast tumor, AI, medicine
An image of synapses in breast tumour, EPFL

Example Breast Cancer Detection

Radiologists are still crucial. They cannot be replaced by AI systems. In a Swedish study with 68,008 findings the best computer program showed a specificity of 81%, as noted in this article. The radiologist who did the initial assessment had a specificity of 98.5%.

AIs - the new content creators

With the AI system ChatGPT, we can converse at a seemingly high level. All within a few seconds, we're provided with complex solutions like programming assistance, a new slogan or even a personalized love letter in the language of Shakespeare. No wonder academic institutions are rubbing their eyes and asking themselves how they are supposed to catch plagiarism in this day and age. How do you catch something that's not a copy, but rather tailor made. Is this still plagiarism? Doesn't the person who asked the right questions have an intellectual stake in the AI's answer?

ChatGPT should not be used like Google, but should be interacted with like a real conversation. This is also because ChatGPT does not know anything that has happened or been written after 2021. However, ChatGPT can slip into a role, can analyse text or make complex contexts easy to understand. Here is an exciting Twitter thread on how to use ChatGPT.

The Chat AI ChatGPT introduces itself
ChatGPT introduces itself

I understand AI more as a working aid to get better results faster. ChatGPT can also help with programming by explaining code, improving existing code or simplifying complex code. ChatGPT can also test functions, help writing test cases or whole documentations. ChatGPT can also find errors in the code at lightning speed. Sure, AI can also write a complete code, but we should not fully trust such systems (yet). Here is a detailed list of how ChatGPT can help with programming.

Here, a complete app code is created with the help of ChatGPT

More and more AI-powered tools, services and companies are emerging, bringing this technology into all areas of life. Google, one of the most valuable technologies of our time, suddenly looks stale. Of course, there is already an AI-based alternative: via YouChat on you.com you get the search result summarised in easy-to-understand text answers. Speech to text AIs like Whisper can create perfect text transcripts from speech, songs or movies. Copy.ai helps generate entire blog articles or headlines.

I asked Copy.ai: "A bold title for a blog article about AI and its impact on design work" - here are the suggestions:

  • Artificial Intelligence: The Designer's New Best Friend?

  • How AI Can Help Designers In The Process Of Creativity

  • AI: The Designer's Secret Weapon

  • What AI Can Do for Designers

  • The Rise of AI: How AI is changing the future and how we design

  • Design meets AI: How Artificial Intelligence is Changing the Way We Create

  • AI: The Tool We Didn't Know We Needed

  • A Data-Driven Future: How AI Will Shape the Design Industry

  • Designers, Meet Your New AI Assistant

  • AI in Design: Unlocking New Possibilities

  • AI and Designers: Is Your Career in Jeopardy?


There are already some interesting approaches among them, which could well have been developed in a brainstorming session in human heads.

YouChat by you.com writing python code
The YouChat on you.com writes phython code

How images are created from text

Although not as earth-shattering as some medical apps or scalable as ChatGPT, as a visual person I'm interested a bit more in the AIs like Midjourney or DALL-E from OpenAI, which generate images - suddenly anyone can create any kind of visual. But it's not that simple, it's designing through words. Only the right inputs and multiple iterations might produce the desired results. Currently, most people simply upload their own portrait and create an artistic image of themselves in the style of famous painters, comics or characters from a game or film. But there are also true masters at work who generate incredible images with a lot of imagination.

Midjourney creation by Dame Pankhurst
Midjourney creation by Dame Pankhurst
Chinese cloisonne astronaut by Dame Pankhurst

Still others create a background for a child's photo, generate a personal wallpaper or design their dream flat. It seems unclear to me how stock photo platforms intend to stay in the market.

Midjourney Interior images by Linus Ekenstam
Midjourney Interior images by Linus Ekenstam
Midjourney Interior images by Linus Ekenstam
Midjourney Interior images by Linus Ekenstam
Midjourney Interior images by Linus Ekenstam
Interior images generated with Midjourney by Linus Ekenstam - Source Twitter

Whether text, film or image, it is already difficult to tell whether content was generated by a human or an AI. It will be exciting to see how this develops. So far, there is little sign that this technology will be regulated in any way. How could it? But are AI-generated images fake? Not necessarily, they are mostly newly generated images that did not exist before, which makes each of these creations original.

Midjourney creations by Jonathan Hoefler
Midjourney creations by Jonathan Hoefler
The fantastic devices of Jonathan Hoefler - source Instagram @jonathanhoefler

It is our human outputs that make the almost endless possibilities and "creativity" of AI possible. So original art, illustrations, characters and designs are needed so that the AI can continue to develop and not stand still in time. Fortunately, the AI needs us (still) more than we need it.

Describing images - how AI generates images

The input at one of the most advanced AI image generators Midjourney works via a chat interface within the Discord platform. With sometimes cryptic inputs, the so-called "prompt", the images are generated. Each image is unique, and did not exist before.

An example: /Imagine: 1980's dark fantasy film still of a lifelike lizard on a science lab::1.5 bench, heavy film grain, extremely detailed, cinematic photography, dark lights --ar 3:2.

/Imagine here is the command that we want to generate an image, followed by what we want to see. In this example, a film still with the feel of a film from the 80s, showing a lizard in a laboratory. ::1.5 indicates the weighting, i.e. that the laboratory environment is important to us. This is followed by some more inputs on the appearance of the image, such as: extremely detailed, cinematic photography and dark lights. The --ar 3:2 defines the aspect ratio.

Now we get 4 different proposals from which we can create an upscale or variants. I make variants of the first picture (V1) and from the new 4 pictures I have the fourth enlarged (U4). Now you could make endless variations and fine-tunings until you like the result. Pictures can also be used as templates to change them or to merge two pictures, for example to generate variants of the same picture.

Here is the complete User Manual to be able to operate Midjourney.

Bildentwicklung in Midjourney, Eidechse im Labor
Bildentwicklung in Midjourney, Eidechse im Labor
Bildentwicklung in Midjourney, Eidechse im Labor
Image development in Midjourney

Here is an example of a background for a client presentation. The idea was to cut out imaginative shapes from coloured papers and photograph them with lots of light to create abstract paper landscapes. For the first time we used Midjourney to generate these specific backgrounds.

Mockup Background, color papercut made with midjourney
Mockup Background, color papercut made with midjourney
Mockup Background, color papercut made with midjourney
Mockup Background, color papercut made with midjourney
Prompt: fun colors, papercut background, clear round shapes, high details

An industry is growing up that will soon be implementing entire projects under AI control. Of course, knowledge of 3D, photography, design, etc. is still an advantage, just as it was an advantage with Photoshop to know how analogue photos were retouched in the past. But basically anyone can create anything if they know which inputs lead to which results or variations.

What the image generator AIs are not yet able to do optimally are logical texts and names, but they also have their difficulties with an exact number of objects. MUSE, the latest text-to-image AI from Google however shows that this and much more is actually possible.

Here you can see the difficulties Midjourney has with logical texts

AI is unlikely to replace human designers in the near future, but it can be used to assist in design processes. Not so much for copying, but for generating design variations, elaborate settings, mockups and new ideas. The results are only as good as the input and really useful results still take some time.

Another difficulty at the moment is to generate continuous results, i.e. to create the exact same background with different objects in the foreground. This would be possible, but the reduced interface does not yet allow us to work on results at a professional level. Here, one could try to take the first version as a starting point and have variants created.

Cans of the future, future Coke can designs by Till Könneker made with midjourney and photoshop
Cans of the future, future Sprite can designs by Till Könneker made with midjourney and photoshop
Cans of the future, future Pepsi can designs by Till Könneker made with midjourney and photoshop
Cans of the future, Fanta can designs by Till Könneker made with midjourney and photoshop, Fanta
Cans of the future, Budweiser can designs by Till Könneker made with midjourney and photoshop, Budweiser
Cans of the future, Evian can designs by Till Könneker made with midjourney and photoshop, Evian
Cans of the future, Root Beer can designs by Till Könneker made with midjourney and photoshop, A&W
Cans of the future, Gatorade can designs by Till Könneker made with midjourney and photoshop, Gatorade
The series "Cans of the future" is a mix of midjourney images edited in Photoshop

AI as a tool - working together with artificial intelligence

AI can support designers and developers in almost all areas of their work and help them to do their work better. AI is getting better at predicting what people need or how they will behave in a particular environment. This empowers UX designers to develop even more user-friendly products.

In the design process, AI can help to create initial mood boards and visual target images. The idea and the image someone has in their head can now be represented much faster as a real image, variants can be created and tested within seconds. Compositions, lighting moods, patterns or an entire photo shoot can be created, adapted and designed by a single person using AI. These options help to make design decisions faster and more confidently.

A practical example of a design process and direct comparison with a human designer can be read in this article.

Midjourney UX Designs by Chris Lüders
Midjourney UX Designs by Chris Lüders
Midjourney UX Designs by Chris Lüders
UI designs created with AI by Chris Lüders from the exciting thread compiled by Linus Ekenstam.

AIs like clipdrop.co or Luminar, which can also be used as a plug-in in Lightroom, support us in image processing. Here, backgrounds can be extracted in a flash, for which one previously had to toil laboriously in Photoshop. A portrait can simply be given a different light, or images can be enlarged in unprecedented quality. This works not only with images, but with applications like runwayml.com also with film.

How can AI support us in the design process?

Moodboards
A wide variety of moodboards emerge from the initial visual research at the beginning of a design process. Colours, photos, fonts and other snippets of inspiration help to explore initial visual possibilities.

KI generated Moodboard for design furniture
AI generated moodboard

Research & UI Texting
As seen above, you can brainstorm with an AI chatbot, but also create or revise entire texts. What the AI does not know, however, is whether a text is good or bad.

Here are some tips for ChatGPT in marketing & SEO in a Twitter thread.

The use of ChatGTP is also interesting in the research phase; for example, target group-specific services or functions can be searched for.

StockFotos
An AI like Midjourney is an exciting alternative to stock photos because it creates copyright-free images that are customisable in terms of colour and content. With a few iterations, photos are generated that can serve as layout images or templates for a shoot. Unlike stock photos, which are often chargeable and require extensive searching, Midjourney offers the possibility to generate images quickly and easily without having to worry about copyright restrictions - at least at the present time. Midjourney can thus be a time-saving and cost-effective alternative to stock photos. Nevertheless, compared to the strength of photos, especially when it comes to shots with people, an AI image generator cannot yet be an alternative.

App icons with different armchairs created with Midjourney
Ideas for App Icon Stylisation

Design elements & mockups
For example, to generate ideas for backgrounds, mood images, 3D renderings, visual shapes and colour palettes. The designs can serve as inspiration or starting point and can be revised and adapted by designers to prepare them for final use in campaigns or user interfaces.

Web design with artificial intelligence from A-Z

AI driven Future

We are only at the beginning of the development and possibilities of artificial intelligence and it is not yet clear where this will lead. We have to make sure that this future becomes a reality that simplifies our lives and does not totally control them. We are already experimenting with things that give us a glimpse of where technology can take us in the future.

Entertainment

Especially in the field of text-to-image AI systems, many applications are currently springing up, such as several children's book generators, where we can influence the content of the story and our own child becomes a character in the story. Of course, you can also implement such a book completely yourself with AI, as described in this article.

The avatar generator Lensar has been in the media because it portrays women in a very one-sided way, to say the least. So we can say the AI is only as good as its input AND the database it is trained with. Who controls what data the AI learns with influences how it sees things and how it presents them. AI can, of course, also be fed its own data sets and thus bred into a specialised AI system.

Conversations

For example, artist Michelle Huang fed the entries of diaries from her childhood to an AI in order to have conversations with her past self. Communication with the afterlife is likely to become eerily real soon.

The experiment "The infinite Conversation" lets us witness an AI-generated endless discussion between Werner Herzog and Slavoj Žižek, which of course never took place.

Psychology

More ethically questionable is an experiment that koko recently conducted. The online platform offers psychological support in chat form. For 4000 people ChatGPT was used and responded to the concerns of the users - and remained unnoticed. In a follow-up survey, it turned out that these conversations were considered more helpful than those of human interlocutors. Read more about the trial in this thread on Twitter.

Deep Fakes

Deep fake technology has rapidly evolved from a gimmick to a powerful tool in recent years. It is now used in various fields, such as film, advertising and entertainment. Some films have already used it to replace actors with computer-generated avatars.

As the name implies, there are also dangers lurking in the use of Deep Fakes, they could be misused to spread false information (Fake News) and thus influence public opinion. There is also the danger that Deep Fakes can be used to embarrass or even blackmail individuals or companies.

Deep fakes change the film industry but also the way we deal with the truth

We humans are good at humanising everything and so we quickly forget that an AI is not really intelligent and certainly does not have a soul or empathy. Nevertheless, it can imitate humans relatively well, has huge amounts of data at its disposal and can provide exciting but also irritating interactions with fictional, living or long-dead people.

GPT (the base model of ChatGPT) is currently available in version 3, but version 4 is already being trained with a much larger data set. This will be able to write a 60,000-word book from a single prompt. At the moment, ChatGPT is still "held back" for security reasons and is therefore not connected to the internet. In 2021, ChatGPT has been disconnected from its data source, so the internet will not be searched in the traditional sense.

Comparison of data volume of ChatGPT 3 with ChatGPT4
Training data volume of GPT 3 vs. chat GPT 4

How far do we let AI take over?

Already, the first companies are being run by an AI. The AI Mika for example sits on the board of a rum manufacturer and Tang Yu, an AI-powered humanoid robot, has been appointed the new CEO of a gaming company with 6000 employees. Tang Yu will serve as a real-time data hub and analytics tool to support day-to-day operations and also play a role in talent development. The extent to which such concepts make sense or are more due to hype and PR remains to be seen, but the question of how far we will go definitely arises.

Yu Tang, virtual KI CEO of NetDragon
Yu Tang, CEO of NetDragon

AI can not only help businesses make better decisions based on comprehensive and meaningful information without political and economic interference, but also entire governments. AI systems can be used to simulate and analyse the potential impact of decisions. This could help governments better predict the consequences of their decisions. However, it is important to note that AI systems are not a substitute for human decision makers and that ultimately humans are always responsible for decisions. In principle, however, I would trust a government that makes its decisions with the help of artificial intelligence more than a system of individual, short-sighted politics shaped by lobbying.

So we need to be very careful about how much power we give to AIs and how we programme them. Janelle Shane, an optics scientist and artificial intelligence researcher, talks about how hard this can be in her TED Talk.

Moral Dilemma
When an AI decides, it can lead to a moral dilemma. How should the AI decide here?

Many may still remember the moral dilemma that the developers of self-driving cars have to deal with. How does the AI decide when it comes to life and death, and who is ultimately to blame when something happens? Moral Dilemma addresses this issue.

But the really useful and disruptive applications with AI are still to come and we can look forward to them. As a creator, I am of course excited about the possibilities, but as a trained graphic designer I am also a little unsure about what this technology will do to the many visual professions in a few years. Closing ourselves off to it or demonising it is no use, we should learn to work with it, be curious and try out how these AI systems can support us in our work.

It is our human outputs that make the sheer endless possibilities and "creativity" of AI possible. So we still need original art, illustrations, characters and designs so that the AI can continue to develop and not stand still in time. Fortunately, the AI needs us (still) more than we need it.

Whether we like it or not, there will be AI systems in the future that generate ready-designed, programmed apps and websites that are perfectly tailored to the user behaviour of the target group.

Conclusion - Brave new (working) world?

When you consider that the capabilities of ChatGPT can be combined with those of Midjourney and then also with the data from user tests, it is pretty clear that we are facing a whole new era in the world of design. How do we deal with it? Do we refuse or do we try to learn and use AI in our everyday work?

At Apps with love we are in an experimental phase. We want to find out where the use of AI in the work process makes sense and where it does not. Of course, ethical and moral issues need to be discussed when using AI technology and blind trust in AI-generated content would be negligent at this stage. We do not believe that AI will make our work redundant in the near future, but it will certainly change it.

The best thing that could happen to humanity through AI would be that it helps us solve our problems. For example, in complex scientific problems, in producing better goods and services, or helping us to improve our quality of life.
The worst that could happen.... well, we've all seen Terminator 😎

AI and Copyright

What is the legal status of all the data the AI has been trained with? Are the generated images transformative enough to be their own? This clip from the Corridor Crew explains the legal situation and the current court case.

Corridor Crew

Data-labelling

When it comes to training neural networks, labelling the data is crucial. These labels, also known as annotations, allow the parameters of the neural network to be calibrated.

Labelling is often done by people in lower cost countries like Asia, Africa and South America. A detailed report on this on technologyreview.com.

I leave the closing words to artificial intelligence

The future of artificial intelligence as a tool for designers, photographers and programmers will bring both opportunities and challenges. On the one hand, AI has the potential to revolutionise the work of creatives, helping them to open up new possibilities and solutions and to realise their ideas more quickly and effectively. On the other hand, there is also a risk that AI could replace their jobs and make it harder for them to compete in the job market. Overall, AI is likely to change the way we work and live forever and it will be important to address and prepare for its impact.

AI can already help us today to optimise our work processes and save time and resources. AI is already in use in many areas, e.g. image recognition, speech processing and autonomous vehicles.

In the near future, we could expect AI to support us in complex decision-making processes and to be present in many areas of our lives. It could be increasingly used in medicine to make diagnoses or create individual treatment plans. It could also be used in the transport and logistics industry to increase efficiency and prevent accidents. We are also likely to interact more with AI in the field of entertainment, for example by receiving even more personalised music or movie recommendations or by using chatbots to replace conversation partners when we feel lonely.

In the distant future, AI might even be able to take on creative tasks and find new solutions to problems that were previously difficult for us humans to solve. It could also help us meet the challenges posed by climate change and help us find alternative energy sources. However, it is important to note that AI is only as good as the data on which it has been trained and that we need to use it responsibly.

- Chat GPT from
openai.com

Further links

The possibilities of AI are almost endless and so I could not cover everything in this article. Therefore, here is an - of course incomplete - list of other AI applications in various fields. To understand the impact of consumer AI, I recommend the exciting interview with OpenAI CEO Sam Altman.

UX/UI Design

AI Film

Code

AI Friends

Marketing Text

Text

Speech to Text

Text to speech

  • murf.ai
    Murf offers many natural-sounding voices for versatile applications

  • lovo.ai
    Text-to-speech generators commonly used for games, audio ads, e-learning and audio books

Image editing

Science and Medicine

Community

We just noticed that you surf with Internet Explorer. Unfortunately, our website does not look so nice with it.

You want to know why that is?
We have written about it.

Blog

You need help with the changeover?
Get in touch. We are happy to help

Contact

Install a new browser?
There's lots of choice.

Browser