The Prophecy
What happens when intelligence goes beyond us?
The next frontier may be intelligence that helps build its own successors. ACT II explores that possibility—and the evidence that separates a future vision from today’s reality.
The path of intelligence
Explore AI → AGI → ASI. A map of capabilities, not a predicted schedule.
Artificial intelligence
Systems that recognize patterns, generate content, and help solve problems. Strong performance in one area does not mean human-level ability everywhere.
Can a system learn and adapt reliably across a broad range of unfamiliar tasks?
The arrows show a conceptual progression. Timing, definitions, and whether these transitions occur remain uncertain.
What is Superintelligence?
Imagine a system that exceeds human ability across almost every intellectual pursuit. That is the idea behind artificial superintelligence, or ASI. AGI describes broadly human-level capability; ASI reaches beyond it.
There is no universally agreed test for that boundary. Nature’s October explainer describes superintelligence as an aspirational category, not an established technological achievement. Being exceptional at one task is not the same as surpassing people across the board. Explore the Science of Superintelligence
Why Superintelligence Matters
What if research could move faster than human teams can work alone? More capable AI could accelerate discovery and make expertise more accessible. Those possibilities depend on systems that remain reliable and under human control.
OpenAI’s September proposal calls for shared safety standards, comparable evaluations, and incident reporting as AI takes on more research. It is a proposal for managing the next phase, not a guarantee of its benefits. The Next Phase of AI
The Open Questions
Who checks an AI that helps build the next AI? How do we keep its work understandable—and decide when to slow deployment?
Stanford HAI’s September discussion highlights independent evaluations, continuous monitoring, and human oversight. Its researchers also describe limits: generating promising ideas or tuning a system does not automatically produce a scientific breakthrough. A single emergency switch cannot answer every safety question. Explore the Risks of Advanced AI
Could AI Improve Itself?
Recursive self-improvement is a feedback loop: AI helps design, test, or train a better successor; that successor then helps improve the next generation. If each cycle becomes more effective, progress could accelerate.
GovAI’s September paper explores whether automating AI research could compress years of progress into months. Such an “intelligence explosion” is a possible route from AGI toward ASI—not a demonstrated outcome or a settled timeline. Research on Intelligence Explosions
AI-assisted research already exists. Fully autonomous self-improvement is a different milestone: OpenAI’s September statement says it is not happening today. Keeping people meaningfully involved remains an open challenge. Read the current limits
Progress toward AGI
The frontier needs measurements as much as predictions. DeepMind’s March framework proposes evaluating ten cognitive abilities, including learning, reasoning, memory, and social understanding, against human performance.
This gives researchers a way to investigate strengths and gaps. It does not certify that AGI—or ASI—has been achieved. Explore the cognitive framework
The Rise of Superintelligence
Superintelligence is moving into public debate as labs pursue automated research and policymakers consider how to oversee it. But the same word can carry very different meanings.
On October 6, Reuters reported that the US Justice Department told staff to use “super intelligence” in place of “artificial intelligence” following a presidential order. This is a change in official terminology, not evidence of a new technical capability. Superintelligence in the News
Explore the Future of Intelligence
Research, proposals, and reporting · Sources checked October 12, 2026
- The science of superintelligence and why AI is still a useful word Nature ·
- Building standards for the next phase of AI OpenAI ·
- Can AI Be Slowed Down? Stanford HAI Experts Weigh the Risks, Rules and Race Ahead Stanford HAI ·
- What If Automating AI R&D Triggers an Intelligence Explosion? GovAI ·
- US Justice Dept tells staff to call AI ‘super intelligence’ under Trump order Reuters · via Investing.com ·
- Measuring progress toward AGI: A cognitive framework Google DeepMind ·
The SI Prophecy is ACT II’s community theme. These independent sources do not imply an affiliation with ACT II. Predictions describe possibilities, not promises.