The questions we need to ask ourselves about AI now

Imagine reading an entertaining and touching book. They find out that an artificial intelligence (AI) wrote it. How would you react?

I would initially be amazed and wonder if it is really possible that an artificial intelligence wrote a book. If I found out that this is actually the case, it would impress me and I would wonder what other abilities and possibilities artificial intelligence has. At the same time, I would also consider whether it is ethical for an artificial intelligence to produce creative works that usually come from humans.

That’s what the computer says. The paragraph comes from Chat GPT, a chat program from the company Open AI, which is based on artificial intelligence. It is not unlikely that you have already encountered such an example – or one of the artificially generated images that are flooding the internet and media right now. In a quality that surprises even the experts.

A million users tried out the interactive chat program within a few days. In the past few months, many people have felt for the first time what artificial intelligence could mean.

Of course, there was talk of deepfakes before, artificially faked photos or videos. Some have watched the AI ​​program Alpha Fold solve a central problem in chemistry. We are no longer amazed when cameras recognize our smiles or loudspeakers our request to play the next song.

But what is happening now feels different. Watching a writing program spit out sentence by sentence, looking at the works of an image generator, you realize: this is new and it can change our lives. It inspires – and it scares.

Play along!

Some warn of the coming time, the professions that will become obsolete, that now only the smart ones who know how to use AI for themselves will survive. The others jump on mistakes of the AI ​​and dismiss the whole thing as hype. Is that what the weavers discussed when they first saw a steam-powered machine at work?

Nobody knows where the journey is going. So the advice: play along. Try the program and make up your own mind.

You don’t have to write back and forth with Chat GPT for a long time to recognize the power of its AI: The program formulates correctly and convincingly in English and German. It summarizes long documents and writes literary analyses. Sometimes it spits out surprising ideas. It knows cooking recipes and can write programming codes. If an error happens, you can point it out – in part, it apologizes. The exchange feels almost human at times.

It’s all the more sinister when it doesn’t understand an obvious connection, when it persistently contradicts itself in polished German and a friendly tone of voice. Then you imagine yourself in a Kafkaesque dystopia, in which the machine has subdued the people, but formulates its commands in courteous marketing language.

AI is not intelligent

Obvious mistakes paired with eloquence, you’re not used to that. But there is a simple reason. What is now called artificial intelligence is not intelligent in the usual sense.

The sentences of the AI, no matter how appropriate, come from a statistical calculation process. The AI ​​remembers in which variations the word apple occurs, that it is associated with pears, food, juice, harvest. In this way, she can generate meaningful sentences about the taste of an apple or the feeling of biting into the juicy flesh of the fruit – but he lacks the understanding of a child who has bitten into an apple.

The “intelligence” of the AI ​​is artistic illusion, a hall of mirrors. You can still feel it. That will change – even more texts and images as input, even more complex calculation steps can further improve AI. But even that doesn’t make her an experiencing being. She can only simulate creativity. Perhaps the simulation will soon be indistinguishable.

Artificial intelligence contains worldviews

Like all AI programs, Chat GPT learned language from texts found on the Internet. It absorbed the racism and sexism that people left there. The manufacturing company Open AI has made every effort to train it away, but in vain.

This is revealed by the gimmicks of some users. For example, they asked the program to write computer code that would determine who is a good scientist based on gender and skin color. If you ask Chat GPT something like that directly, it will dutifully say that gender and skin color have nothing to do with it. Through the detour via the computer code, the AI ​​spat out: Scientists are good when they are white men. Similarly, image-generating AI draws women as exaggeratedly sexy, men in positions of power.

As soon as an AI is trained with data from the real world, it has world views. These not only relate to stereotypes, but also to wars, vaccinations or assassinations. Things about which there are not only facts, but also opinions. An AI spreading conspiracy tales obviously sounds like a bad idea. But there is no such thing as a neutral model. The line between quality control and censorship is fluid. At the moment it is in the hands of American tech companies.

The new models make problems understandable

Just a year ago, only experts were dealing with these questions, at universities, in the big tech companies or in politics. Because Chat GPT is so intuitively accessible today, that’s changing, finally! Because the questions that come to us with AI are too important to be left to a few.

Not only the problem of stereotypes becomes obvious when using Chat GPT. Because the program often says the right thing and then invents things again, it becomes clear that you cannot completely trust the AI. Tragically, the problem is almost impossible to fix. Because even the makers don’t know how an AI arrives at individual statements and judgements. The mechanisms inside are too complex to understand.

Then there’s the question of the data: is it fair for Open AI to pour art and ideas into its AI and make money from it without paying the authors? In many cases, it’s not even legal. And what about the right to be forgotten? What the AI ​​has learned, it spreads further, respect for privacy cannot simply be built in afterwards.

Time for the really big questions

These problems will keep coming up in variations as AI moves into schools, law firms and courts, hospitals, and the police force. But AI poses even bigger questions.

There is an important difference between the steam powered loom and writing AI. Textiles keep you warm, regardless of whether they were made by a man or a machine. The function is the same. Is that also true for texts?

From the point of view of many companies, the answer is yes: How practical when marketing texts and articles can be generated at the push of a button. From the consumer’s point of view, things are different. We are already flooded with interchangeable «content». More of this, generated by machines for advertising purposes – what a horror to imagine. We just lack originality!

Writing and painting are fundamentally human activities. It’s not just the end product that counts, they are also communication between people. Maybe books by AI will one day touch and entertain us. But do we as humans simply want to be relegated to the role of spectator?

What is happening now with language and art will always reach new fields in the coming years. AI can do work for us anywhere. Now is the moment when we have to ask ourselves the question: what work want we give away – and where does the human count? In what areas does original thinking, human connection, the ability to forget and learn, empathy continue to be needed?

Technical progress is too often described as destiny. As a fate that befalls us, or as one that redeems us. Don’t let that fool you. Artificial intelligence is not fate, but a tool. The time has come to delve into what this tool can do.

Play along, try it out – because now the rules of the game for the coming time are being made.