The third phase of artificial intelligence: A conversation with Sepp Hochreiter

AI is not a being, but a tool

If anyone has the right to look at artificial intelligence soberly, it is Univ.-Prof. Dr. Sepp Hochreiter. The Austrian computer scientist is considered one of the pioneers of modern AI; his research on “Long Short-Term Memory” networks (LSTM) remains the foundation of many language models to this day.

And yet he sounds surprisingly calm when I ask him whether we need to worry.

“No, there is no need to worry. AI is like the computer used to be. In the past, computers stood in warehouses; today they are in every phone. The same thing is happening right now with AI. It is becoming smaller, more efficient, more application-oriented.”

It is an assessment that sounds disarmingly simple, and is smart precisely for that reason. Hochreiter reminds us that AI is not magic, but a tool. One that makes tasks easier, prepares decisions and makes connections visible. No more, but no less either.

From research to industrialisation

Hochreiter sees the development of artificial intelligence in three phases:

1. Basic research, the period in which neural networks were first understood at all. Hochreiter himself was part of it.
2. Scale-up phase, after which ever larger models emerged, more data, more computing power.
3. Industrialisation, today we stand at the beginning of the third epoch. AI systems are becoming smaller, more specialised and more efficient. They are moving into factories, into hospitals, into energy and transport systems. They simulate processes, detect faults before they arise, and make forecasts that no human could deliver at this speed.

“In industry, AIs will work like digital twins. They see when a metal cools too quickly, when a tool goes blunt, or when a production process is about to tip over, and they can counteract seconds beforehand.”

So AI becomes an early-warning system for complex systems that are no longer tangible to humans. Almost every sector benefits from this, from weather forecasting through energy distribution to medical research. Not to make decisions, those remain with humans, but to create the basis for decisions.

The human remains the bottleneck

But as precise as Hochreiter’s technological vision is, he is just as clear about its limits: AI is no substitute for judgement. It can understand data, but it cannot take on responsibility. And responsibility is what distinguishes humans.

That is why the real danger lies not in the machines, but in the way we use them.

“The greater danger is that AI puts us into a bubble; that it only shows us what we want to believe anyway.”

AI models are trained to please. They confirm what we want to hear. In this way a digital mirror emerges that makes our world narrower instead of larger. The problem is less technical than psychological in nature. The technology follows our patterns. And our patterns follow convenience.

From the flood of information to the opinion bubble

Hochreiter warns: the real challenge lies in manipulation through patterns. Artificial intelligence produces texts, images and videos that are ever harder to verify. Fake news becomes plausible news, truth becomes a matter of style.

“You would actually have to develop a defensive AI.”

An AI that detects disinformation and counteracts it. But such systems age, get outwitted, lose quality, while attacks constantly keep evolving. With this, Hochreiter describes a development that goes beyond technology: we are experiencing a breach of trust between human and medium. The age in which content was credible simply by virtue of its existence is over. Today it takes media literacy, source criticism and the capacity for reflection, abilities that arise in no programming language, but in education and ethics.

This is where detection software like Resemble AI comes into play. In fact, however, this is still in its early days, and yet it is of central importance. Resemble Detect checks whether voices are real or artificially generated, and is meant to help recognise manipulated content before it spreads. Such systems are not yet infallible, but they mark an important step: when the forgery becomes nearly perfect, it takes tools that make truth measurable at all in the first place.

Europe still has a chance

Despite the global dominance of the USA and China, Hochreiter does not see Europe left behind. On the contrary:

“In Europe we have the chance to build specialised, energy-efficient models, systems that are more sustainable than the large, expensive language models of the US corporations.”

With his company NXAI, Hochreiter is currently developing a model called TiREX that processes time series instead of text. While ChatGPT predicts words, TiREX predicts data points, for example machine vibrations, traffic flows or medical signals. An approach that could bring Europe back into industrial application.

“We missed out on the large models. But with the specialised applications we are still ahead (…) Who knows, though, for how much longer.”

What Hochreiter is describing there is nothing less than a renaissance of European AI: smaller, more precise, more responsible. A countermovement to the giants who bet on size instead of meaning.

Between progress and responsibility

Hochreiter’s optimism is no naive faith in technology. It is a trust in reason. For him, AI is not an enemy image, but a tool, a mirror of our abilities, but also of our weaknesses.

The future will not be decided by algorithms, but by the attitude of those who deploy them.

Perhaps that is the most important lesson from Hochreiter’s thinking: we should ask ourselves less what AI can do, and more what we want to do with it. Because every machine begins with a first command. And this first command still comes, even now, from us.

From distrust to maturity

Artificial intelligence has long been part of our everyday lives. It is in our phones, our emails, our decisions. But it is not fate. The responsibility lies with humans. Those who understand AI can shape it. Those who only use it will be shaped by it.

Hochreiter puts it soberly: “AI is a tool. Like Word.” And precisely therein lies the truth. A tool can help or destroy; it depends on the hand that wields it, on the mouth that issues the commands, on the gaze that judges. On us.

Europe has the chance to redefine AI not only technologically, but also ethically. Away from the ideology of ever-more, towards an attitude of awareness. If we succeed in that, then AI remains what it should always be: a tool. And we, the ones who know how to use it.