Artificial Superintelligence

An Artificial Superintelligence, or ASI for short, is an AI that not only performs individual tasks better than humans, but is superior in every conceivable intellectual discipline. This includes mathematics, natural sciences, language, creativity, and strategic thinking simultaneously. It is thus clearly distinct from today’s systems like ChatGPT or AlphaGo, which excel in a narrow domain each but fail outside of it. No ASI exists yet – the term describes a hypothetical future system. However, research and the public are intensely debating whether and when it might emerge.

Why the ASI debate moves AI research

Most technological breakthroughs have manageable consequences. An ASI would potentially be an exception. A system smarter than any human could, in a short time, develop new technologies, medicines, or strategies that no human could have designed alone.

That is precisely why entire research fields grapple with the question of how one would control such a system. This subfield is called AI Safety, i.e., AI safety research. The core question is: how do you ensure that an ASI does what humans actually want – and not just what they superficially stated? This problem is harder than it sounds. Even today, AI systems sometimes deliver results that are technically correct but practically useless or dangerous.

Economists are also interested in what an ASI would do to labor markets. If a system can perform every intellectual job faster and cheaper than a human, the economy and society change fundamentally. Such considerations are speculative, but they explain why the term appears increasingly often in political and scientific debates.

Three stages on the path to ASI

Researchers typically distinguish three stages of development. The first is Artificial Narrow Intelligence (ANI): systems that are very good at one thing. All current AI products fall into this category. The second stage is Artificial General Intelligence (AGI): a system that can flexibly learn and solve any intellectual task, similar to a human. The third stage is then ASI.

Whether this staged progression will actually occur remains open. Some researchers believe the jump from AGI to ASI could happen very quickly – because a generally intelligent AI could improve itself without waiting for human developers. This scenario is called an intelligence explosion. Others counter that intelligence is not a uniform measure, and a simple increase therefore doesn’t work like turning up a volume knob.

There is no reliable timeline. Forecasts from AI researchers range from a few decades to “never.” This is because no one knows which technical hurdles still need to be overcome – and whether they can be overcome at all.

Where the term appears in practice

Current products do not yet contain ASI, and the term therefore rarely appears in product announcements. It appears, however, frequently in risk reports, such as when regulators in the EU or the US think about long-term AI legislation. The EU’s AI Act does not explicitly mention ASI, but its shaping was informed by considerations of where development might lead.

Investors also use the term. Companies like OpenAI or DeepMind have explicitly named the development of AGI or ASI as a long-term goal in their founding documents and public statements. When such companies raise billions in capital, they often justify it by arguing that the race for this technology is strategically decisive.

In pop culture, superintelligence has been present for decades – from HAL 9000 in “2001: A Space Odyssey” to Skynet in “Terminator.” These portrayals often shape the public image more strongly than scientific texts, even though they have little to do with actual research. Anyone reading the term in the news should therefore check the context: is it about serious research, investor communication, or science fiction?