--- title: "From AI to SI: The Next Big Thing Is Intelligence" slug: from-ai-to-si description: "Why the conversation is moving from AI to superintelligence: what SI means, how it differs from AI and AGI, self-improvement, and why safety matters." date: 2026-10-28 last_reviewed: 2026-11-02 status: published type: story author: [liyam-flexer] image: from-ai-to-si image_alt: "Cover graphic: From AI to SI: The Next Big Thing Is Intelligence, over a dark blue network of connected points." tags: [superintelligence, si, asi, agi, ai safety] --- # From AI to SI: The Next Big Thing Is Intelligence **In short:** AI is the technology we use today; superintelligence (SI) is a hypothetical level of intelligence that would outperform the best humans across virtually every intellectual field. It has not been built, nobody knows how or when it might arrive, and it is not a guaranteed next stage after AI. The shift from AI to SI is a shift in the question we ask: from what AI can do for us now to what intelligence itself could become, and how to keep it safe. Every generation gets its defining technology. Electricity changed how we live. The internet changed how we connect. Smartphones put the world in our pockets. Now AI is changing how we work, create and think. But what if AI is just the beginning? What if the real story isn't about machines that help us do more, but about intelligence that can take us somewhere we've never been? This is the shift from AI to SI: from artificial intelligence to superintelligence. ## AI is the beginning. SI is the bigger question. AI is already impressive. It can write, code, create images, analyse information and help turn ideas into reality. Things that once took hours can happen in seconds. But today's AI still has limits. It can get things wrong, struggle with unfamiliar problems and produce answers that sound brilliant without being correct. See [Why AI Hallucinates](./why-ai-hallucinates.md). Superintelligence takes the idea much further. Nick Bostrom's 1998 paper defines it as "an intellect that is much smarter than the best human brains in practically every field". Think science, invention, strategy, mathematics, creativity, and problems we haven't yet figured out how to ask. Not a better chatbot, but a whole new level of intellectual capability. We haven't built it. We don't know exactly how it would work. Nobody can honestly say when it will arrive. But the possibility changes the conversation. For the full definition, see [What Is Artificial Superintelligence (ASI)?](./what-is-artificial-superintelligence.md) and [Artificial superintelligence](../vol06_asi/ASI-001_artificial_superintelligence.md). ## The shift is bigger than two letters AI has become the name of a technology. SI is a way to think about its possible destination. The difference: - **AI:** making machines capable of intelligent tasks. See [Artificial intelligence](../vol01_foundations/FOUND-001_artificial_intelligence.md). - **AGI:** building AI that can handle a broad range of intellectual tasks at a general level comparable to humans. See [What Is AGI? Definitions Compared](./what-is-agi-definitions-compared.md). - **SI:** exploring the possibility of intelligence that goes beyond the strongest human capabilities across almost every intellectual field. These are not guaranteed stages in a predictable sequence. They are different ideas that help us talk about where intelligence stands today and where it might go. Morris et al. (2023) go further and propose graded levels of AGI instead of a single threshold. The important shift is to stop looking only at what AI can do for us today, and to ask what intelligence itself could become. ## Imagine what becomes possible Forget the usual demo, another chatbot writing another email. Think bigger. **Science without the same old limits.** An intelligence that helps researchers discover new medicines, understand complex diseases or answer questions that have challenged generations of scientists. See [AI in science](../vol07_applications/APP-SCI-001_ai_in_science.md). **Ideas that change everything.** Exploring thousands of possible designs, testing new approaches and finding solutions no human expert had considered. **A new kind of creativity.** Intelligence that connects ideas across science, art, engineering and philosophy to produce something useful and unexpected. **Problems we finally know how to solve.** A new way to tackle challenges in energy, education, transportation or the environment. None of this is a promise. It is a glimpse of what superintelligence research asks us to consider. The exciting part isn't that we know what comes next. It's that we don't. ## It's not about making machines more human We often judge AI by how human it seems: can it talk, write and answer like us? But intelligence doesn't have to copy the human brain to become powerful. A calculator doesn't think like a mathematician. A telescope doesn't see like a human eye. Both extend what people can do. Superintelligence could work in ways fundamentally different from human thinking, combining large-scale computing, advanced reasoning, learning and the ability to explore problems at a scale people cannot match. The question isn't whether a machine will think exactly like us. It's whether it could find ways of thinking that take us beyond what we can do alone. See [Defining and measuring superintelligence](../vol06_asi/ASI-DEF-001_defining_and_measuring_superintelligence.md) and [Pathways to superintelligence](../vol06_asi/ASI-PATH-001_pathways_to_superintelligence.md). ## The big question: can intelligence improve itself? One of the most discussed possibilities is that AI could help build better AI. A system helps researchers improve a model; that model helps develop the next one; the next finds better approaches still. Could that create a cycle of increasingly capable intelligence? Researchers study related ideas, including automated AI development and recursive self-improvement, but the limits, speed and feasibility of such a process are unknown. It is not a guaranteed intelligence explosion. It is one of the big questions worth investigating. See [Automated AI research](../vol06_asi/ASI-AIRD-001_automated_ai_research.md), [Intelligence explosion](../vol06_asi/ASI-EXP-001_intelligence_explosion.md) and [Intelligence Explosion and the Singularity](./intelligence-explosion-and-singularity.md). And it raises a second question: if intelligence becomes more powerful, how do we make sure it stays reliable, controllable and beneficial? ## The future needs more than smarter machines More intelligence doesn't automatically mean a better world. Powerful AI could speed up discovery and open new opportunities. It could also be misused, make consequential mistakes, concentrate power or behave in ways that are hard to predict. That is why safety matters as much as capability. Amodei et al. (2016) is an early catalogue of concrete safety problems. The goal is not to build the most powerful system possible and hope for the best. It is to understand what these systems can do, set appropriate safeguards and make sure their development serves people. The future of intelligence is a technical challenge, and a human one. See [The AI Alignment Problem, Explained](./ai-alignment-problem-explained.md), [AI alignment](../vol08_safety_governance/SAFE-ALIGN-001_ai_alignment.md) and [Existential risk](../vol08_safety_governance/SAFE-XRISK-001_existential_risk.md). ## So, why from AI to SI? Because the conversation is getting bigger. AI asks how machines can perform intelligent tasks. AGI asks whether machines can achieve broad, general-purpose intelligence. SI asks what happens if intelligence goes beyond the best of humanity. That last question reaches far beyond software: it touches science, creativity, discovery, economics, safety and our understanding of intelligence itself. We don't know whether superintelligence will become real, when it might happen or what form it might take. But we don't need to pretend the future is settled to explore what is possible. For how experts' forecasts differ, see [Expert Surveys on AGI Timelines](./expert-surveys-on-agi-timelines.md). AI is changing what we can do. Superintelligence challenges us to imagine what we might be able to discover. The next chapter of intelligence is still being written, and the interesting question is what we will do with it. ## Frequently asked questions **What does "from AI to SI" mean?** It describes a shift in focus from today's AI, which performs intelligent tasks, to superintelligence (SI), a hypothetical intelligence that would outperform the best humans across almost every intellectual field. It is a way of framing a question, not a prediction that SI will follow AI. **Is superintelligence the same as AGI?** No. AGI refers to AI with broad, human-comparable ability across many tasks. SI refers to intelligence beyond the strongest human capabilities across almost every intellectual field. Both terms lack a single agreed technical definition. **Does superintelligence exist yet?** No. It has not been built, and nobody can reliably say when or whether it will be. **Does SI have to think like a human?** No. Intelligence does not need to copy the human brain to be powerful; a superintelligent system could work in ways very different from human thinking. **Would AI improving itself lead to superintelligence?** That is an open research question. Automated AI development and recursive self-improvement are being studied, but their limits, speed and feasibility are unknown, and an intelligence explosion is not guaranteed. **Why does safety matter for SI?** More capable systems could be misused, make consequential mistakes or behave unpredictably. Safety research aims to keep advanced systems reliable, controllable and beneficial. ## Sources - Bostrom, N.: [How Long Before Superintelligence?](https://nickbostrom.com/superintelligence) (1998; reprinted 2006) - Morris, Sohl-Dickstein, Fiedel, Warkentin, Dafoe, Faust, Farabet, Legg: [Levels of AGI for Operationalizing Progress on the Path to AGI](https://arxiv.org/abs/2311.02462) (2023) - Amodei, Olah, Steinhardt, Christiano, Schulman, Mané: [Concrete Problems in AI Safety](https://arxiv.org/abs/1606.06565) (2016) *Status: draft, 28 October 2026. An opinion and explainer piece by Liyam Flexer. The definition is Bostrom's (1998) wording; the two arXiv references were checked against their records but not re-read in full. Statements about what superintelligence might do are possibilities, not findings, and nothing here is a forecast.*