Speech by Federal President Dr. Frank-Walter Steinmeier before the members of the Order Pour le mérite for Sciences and Arts on June 11, 2023, in Berlin:

Germany Frank-Walter Steinmeier · Federal President of Germany

12 June 2023 · Order Pour le mérite for Sciences and Arts members meeting · Berlin, Germany

Context
Steinmeier addressed members of the Order Pour le mérite for Sciences and Arts in Berlin on June 11, 2023, reflecting on artificial intelligence's implications for democracy and human knowledge.
Main themes
Artificial intelligence and democracy, human-machine relationship, AI and knowledge, authorship and creativity in the AI era, AI threats to autonomy, democratic legitimacy, disinformation and factual basis, AI in science and arts, liberal democracy vs authoritarianism, AI regulation and responsibility
Duration
~14 min
48th percentile
Words
1,661
48th percentile
Sentences
75
49th percentile
Readability
46
50th percentile
Grade
12.3
49th percentile
Syntax
3.25
78th percentile
Vocabulary
2.14
35th percentile

Translated from German

Content
Actors
Style

The broad sphere of national life a sentence belongs to. Every sentence has an Arena, so it is the denominator the other indices are read against.

Other
  • Technological 68%67%
  • Social 11%11%
  • Institutional 11%11%
  • Ceremonial 5%5%
  • Geopolitical 4%4%
  • Economic 1%1%
  • Environmental 0%0%
  • Not in focus 1%

75 sentences in 19 paragraphs

"The purpose of our foundation is the recognition […] of the hidden forces in nature and the expansion of human dominion to the limits of what is even possible."

Here, I am not quoting the founding document of the Order Pour le mérite, although it admittedly sounds very much like it. No, the "expansion of human dominion to the limits of what is even possible" is the subject of a completely different document, a utopian work called "New Atlantis," published by Francis Bacon nearly four hundred years ago. In it, Bacon tells of sailors who were blown off course between Peru and China and were taken in by New Atlantis. There, they learn about astonishing experiments, such as those related to weather, newly bred plants and animals, or astronomical instruments.

When we talk today about hidden forces and the limits of what is possible, I think – and probably many of you do as well – immediately of the rapid development in the field of artificial intelligence. I know: Stories about human machines, about artificially created intelligence have always existed. This is not only the case with Francis Bacon, but I also think of Pygmalion, the homunculus, Frankenstein, or – those of you who are film enthusiasts will remember – HAL 9000. As I said, those were stories, modern fairy tales, science fiction.

But today, we are at a point with generative AI where we must seriously debate how closely a machine can and should come to our human cognitive abilities. And: How much can and should it perhaps even exceed these human abilities?

Fundamental questions about the relationship between humans and machines arise here, and I know how much these questions concern almost all of you in this room: What does knowledge mean in a time when an algorithm can seemingly explain everything to us? Why should our children, our students, or our pupils still learn, why should they strive for new insights when there is artificial intelligence? Who will actually be considered the author of scientific articles and papers in the future: Is it the AI, the person who queries the AI, thus using it, or the one who programmed the AI? Who will even be able to verify whether there is an original contribution behind texts and articles, whether the knowledge and necessary skills have truly been internalized? How creative can an AI be, at what point is a work of art created by AI truly art, and when is it merely technology? How does the self-image of artists and scientists change as a result?

These are just some of the questions that lie ahead of us. AI is a disruptive technology that will undoubtedly change all our lives. I am very glad that we have among us in this round minds who approach these questions with much experience, great wisdom, and foresight, regarding the immense positive potential that AI undoubtedly has, for example in medicine, but also of course the threats to individual self-efficacy or human autonomy.

However, there is one challenge that particularly concerns me as Federal President. Because I believe it has consequences, at least it could have, for our democracy. In other words: Where does it lead us when AI explains things to us with great plausibility, at high frequency, and with rapid dissemination, offers knowledge, creates images that have little or even nothing to do with reality? How can we debate and decide when we can no longer rely on facts? Or: Could – and should – AI even take important decisions away from us, should it make political decisions for us, for a government, since it seemingly knows so much more and comes to decisions so much faster than humans?

For us, I think it remains the case: Neither AI nor the companies that operate it are democratically elected. And an algorithm may offer good or even just decisions, but it is not a subject. It cannot take responsibility for these decisions itself and can never legitimize them. That remains the prerogative of humans and – in the great public questions – the democratic institutions of an open and free society, I say.

But looking at the world out there, I must add a caveat: We see it this way, we as liberal democracies. But we all know that there are authoritarian and libertarian-technocratic powers and forces that see it quite differently. All the more important that we are clear in our stance.

The data and patterns of yesterday, of past decisions, are not suitable as the sole basis for political decisions for tomorrow. In a democracy, it is not about reproducing and automating decisions, but about societal negotiation, about gaining insights, about changing and improving, about pluralism of opinion, about the integration of the deviant, about civilized dispute, and hopefully also always about finding common solutions. Normative questions probably cannot be solved mathematically, not solely based on data. In this respect, I do not see the end of our democracy nor the emergence of a "minimal state" through AI.

But there are indeed dangers: AI can be used manipulatively and maliciously. You may recall: Recently, an image of an explosion at the Pentagon circulated, disguised as serious news, viewed thousands of times, shared hundreds of times. The responsible fire department and the official spokesperson for the U.S. Department of Defense denied it. The fake was exposed, but on the financial markets, this fake had already created a new reality, causing stock prices to plummet. Clarifications are not always so quick and rarely so clear, and above all: Not everyone who saw the news and shared it receives the clarification.

Even in Germany, fake images have often circulated, most recently a supposed photo of angrily shouting men with black hair, accompanied by a xenophobic slogan. By the time it was recognized that these men do not exist at all, that one or another of those depicted even had a few too many limbs, the photo had already circulated on many channels. Spreading fear with fakes works unfortunately all too well.

This is already the case now, but of course even more so with the possibilities of AI. Now, you all know the objection: Yes, fake and retouched photos have existed more or less since the discovery of photography. But the difference is this: Today, almost anyone can create one with an image generator in seconds and, in principle, spread it indefinitely with just a few clicks, and the fakes are getting better and better. If you google "fake photo," the search engine first suggests "create fake photo," and only afterward follows "recognize fake photo." Exactly in this order. Once the fake is out in the world, as everyone who has dealt with this knows, it is hard to recapture.

How do we recognize what is reality and what is merely the product of a manipulatively deployed AI? How do we confront the rapidly increasing and ever more deceptively presented disinformation that delegitimizes democratic politics and fact-based decisions, that may discredit science and also corrupt the markets?

I believe: We will not achieve a comprehensive fact-check with a single solution, and technology alone will not provide this solution. Rather, we need a kind of digital general education. We all need to learn to verify the answers of AI instead of simply relying on the plausibilities of AI. We need an ethical and legal framework with effective standards and control bodies. Where AI is involved in decisions, we must ensure the verifiability of those decisions. We must be able to uncover when AI is used maliciously and distorting – and for that, we need, among many other things, a strong and independent journalism. We need media that do not ultimately make themselves dependent on the use of AI when they research. And perhaps it is time for a kind of watermark that becomes visible as soon as AI produces content, in images and in texts. And we need, I say this after an expert discussion here in the house, AI developers who do not merely understand or misunderstand humans as a deficient algorithm.

In these days, leading AI developers are warning, as you have seen in the newspapers, about their own technology. I cannot assess the intention behind this. Some believe they know it is simply marketing and sensationalism. Others see it as an attempt to shift the responsibility for potential serious damage onto politics, onto the lack of regulation. But what matters is: The warning that potentially uncontrollable risks are coming our way is out in the world, and it deserves at least attention.

For today, at least, I can say: Among the new members of the Order Pour le mérite, whom I have just had the honor to award, ChatGPT is not represented, but rather with David Chipperfield, Peter Gülke, Peter Schäfer, and Michael Tomasello, we have concentrated natural intelligence. Jürgen Habermas cannot be here with us today, but I was able to speak with him extensively at his home a few weeks ago and discuss with him.

All of you belong to the Order Pour le mérite because of your extraordinary achievements. I asked ChatGPT whether AI assumes that it could someday achieve such excellence as you have. The answer was essentially: "Interesting question, I am not there yet – but I learn so quickly and so much that I could eventually surpass certain aspects of human excellence in some areas."

I would like to respond to this self-assessment of AI: It is – at least so far – still mediocre, the mediocrity of the data underlying it. You, who are gathered here this evening, know: True excellence requires more. Namely, the unique artistic and intellectual achievements of people like you. Science and the arts gain another potent tool with AI, no more and no less – a tool that would have excited Francis Bacon and the inhabitants of New Atlantis, and a tool with which we must learn to deal.