AI already surpasses human average capabilities in poetry, fiction, music, coding, and translation, making the possibility of technology exceeding human cognition no longer science fiction. But the era of "Superintelligence" — surpassing humans in all intellectual abilities — is not yet reality. When will it come, and in what form?
Nick Bostrom's definition: "An intellect that is much smarter than the best human brains in practically every field, including scientific creativity, general wisdom and social skills." Current state (as of May 2025): GPT-4.1 (MMLU 80.1%, SWE-bench 54.6%), Claude Opus 4 (SWE-bench 72.5%), Gemini 2.5 Pro (excels in GPQA and AIME 2025), Grok 3 (AIME 2025 93.3%) — performing at or exceeding human expert levels in specific domains, but still classified as early-stage AGI, not true superintelligence.
Timeline predictions: OpenAI (AGI around 2030, launched "Superalignment" project July 2023 with $100M investment); DeepMind CEO Demis Hassabis (human-level general intelligence within 10-20 years, mid-2030s); Nick Bostrom (superintelligence possible before 2050); Yann LeCun (current language models lack true understanding, decades away); AI Index Report 2024 (AGI possible in early 2030s, superintelligence likely after 2045). Three barriers to superintelligence: (1) Computational resource limits and scalability (trillions of parameters, energy-efficient architecture needed); (2) Self-awareness and goal-setting ability (humans can ask "why solve this?" but AI cannot redefine its own purpose); (3) Understanding and intervention in the physical world. Three scenarios: optimistic (AI as cooperative partner solving climate, medical, science challenges by 2040s); neutral (partial implementation in limited domains 2035-2045); pessimistic (monopolized by specific corporations/nations or losing control — most feared, possible after 2040). International strategies: US (NAIRR program, DARPA AI R&D); China (Next-Generation AI Development Plan, targeting world leadership by 2030 with DeepSeek open-source); UK (first AI Safety Summit November 2023, Bletchley Declaration, AI Safety Institute). "Superintelligence is not a technology problem but a 'decision' problem."
