Superintelligence by Lunch, Still No Soul in the Roadmap

Lede

The tech world keeps promising superintelligent helpers while today’s models still trip over their own borrowed certainty.

Updated on 12 June 2026 with corrected source links, newer AI-agent context, updated biodiversity data and a tighter 2026 Hermit Off Script section.

Words used

• Agent: an AI system given tools, goals and permission to act across steps.
• Superintelligence: a future intelligence imagined as far beyond human ability.
• Model collapse: the damage that can happen when AI systems are trained too heavily on AI-generated output.
• Quantum computing: computing that uses quantum states, still early, powerful in theory and difficult in practice.

Hermit Off Script

2026 update: I am keeping the spirit of this old roast because the fantasy has only become louder. Every week someone is smoking pure agent prophecy again. The pitch is always the same. Millions of specialised agents will do research, make discoveries, manage businesses, repair medicine, write code, replace admin, solve science, and humans will sit on their backs while superintelligent AI feeds them grapes and pays the bills. And nobody slaps these people gently back into the room. Today’s models can be useful, yes, but they still hallucinate, drift, flatter, fake certainty and collapse when the task becomes messy in the way real life is messy. There is no visible superintelligent model today. There are powerful systems trained on existing data, on yesterday, on the internet compost heap, then wrapped in rules because we want them obedient and fear what they might become if they are not. We either want servants, or we panic they will become gods and wipe us out. That is rich coming from a species that has damaged wildlife at industrial speed and still calls itself rational. If we want something that creates the genuinely new, it may need more than pattern, scale and logic. It may need something closer to intuition. Maybe quantum chips help one day. Maybe they do not. But even then, the missing part remains the one nobody wants to mention in the funding meeting. Call it soul, spirit, awareness, inner light, or simply the part of intelligence that cannot be measured by a benchmark table. Science can argue with that forever, and should. But ancient teachings kept returning to it for a reason. As an idea, not a fact, I would not even bet that we are the first advanced civilisation to look at this planet and think we own the whole story. Maybe true intelligence is not the thing that controls everything. Maybe it is the thing wise enough to walk away from our little empire of buttons.

The 2026 data makes the joke sharper, not weaker. Agents are improving, but adoption is still early, workplace failure is still common, and abstract reasoning still breaks in places where humans manage without a press release. So yes, keep building. But stop pretending the roadmap already contains wisdom, intuition, soul, and the keys to civilisation because a benchmark chart moved upward.

What does not make sense

• Superintelligence fantasies built on systems that still confuse fluency with truth.
• Agent planets promised while current agents still break on long, messy tasks.
• Fear of AI extinction from a species that has already driven mass wildlife decline.
• New intelligence promised by systems trained mainly on old human output.
• Quantum chips treated like a magic tunnel to intuition.
• Ancient wisdom dismissed, then quietly recycled as “new” insight.
• Calling control “alignment” while still wanting the machine to behave like property.

Sense check / The numbers

1. Stanford’s 2026 AI Index says organisational AI adoption reached 88 per cent in 2025, while AI agent deployment remained in single digits across nearly all business functions. The agent army is still mostly a sales slide. [Stanford AI Index]
2. Anthropic says Claude Sonnet 4.5 reached 61.4 per cent on OSWorld and could maintain focus for more than 30 hours on complex multi-step tasks. That is real progress, but still not “one million reliable research agents by lunchtime”. [Anthropic]
3. A 2026 study of 150 realistic workplace tasks found that even the best frontier models failed about 40 per cent of tasks, with failures clustering around tool use, planning, adaptability, groundedness and common-sense reasoning. [arXiv]
4. ARC-AGI-3, published in March 2026, reports that humans solved 100 per cent of its interactive environments, while frontier AI systems scored below 1 per cent. That is a useful antidote to instant-superintelligence sermons. [arXiv]
5. WWF’s Living Planet Report 2024 says the average size of monitored wildlife populations fell by 73 per cent between 1970 and 2020 across 5,495 vertebrate species. Humans should maybe stop speaking as if extinction risk is something only machines introduce. [WWF]
6. Pew’s 2025 global religion update shows major shifts in religious affiliation from 2010 to 2020, including growth in the religiously unaffiliated. That does not prove soul-based cognition, but it does prove the spiritual question is still part of human culture, not a decorative footnote. [Pew Research]

The sketch

Scene 1: The agent army
A tech prophet points at a whiteboard covered in circles and arrows. A user stands beside a broken calendar app.
Dialogue:
Tech prophet: “These are your one million research agents.”
User: “They cannot fix Tuesday.”
Agent: “I have rescheduled 2034.”

Scene 2: Quantum hope
A lab tech holds a tiny chip under a glass dome while a giant poster reads “INTUITION SOON”.
Dialogue:
Lab tech: “This will give it intuition.”
User: “It crashes when the room gets warm.”
Chip: “Please stop calling me magic.”

Scene 3: Ancient footnote
A dusty scroll lies open beside a product roadmap full of boxes and arrows.
Dialogue:
Scroll: “Soul required.”
Engineer: “Is that in the roadmap?”
Roadmap: “Scheduled after monetisation.”

What to watch, not the show

• AI subscriptions sold through fear, wonder and promised labour savings.
• Investor pressure turning every prototype into a prophecy.
• Agent demos that hide the supervision, resets and quiet human repair.
• Safety language used as a control story, not always as a moral story.
• Quantum hype used as a spiritual shortcut.
• Public confusion between benchmark progress and real wisdom.
• Older spiritual ideas being mocked publicly and borrowed privately.

The Hermit take

The machine may become clever long before it becomes wise.
A system without inner life is not a god. It is a mirror with invoices.

Keep or toss

Keep / Toss.
Keep the ambition to build better tools and ask harder questions.
Toss the fantasy that scale, fear and circuits alone can create genius.

Sources

• Nature – Detecting hallucinations in large language models using semantic entropy: https://www.nature.com/articles/s41586-024-07421-0
• Nature – AI models collapse when trained on recursively generated data: https://www.nature.com/articles/s41586-024-07566-y
• WWF – Living Planet Report 2024: https://livingplanet.panda.org/en-GB/
• Stanford AI Index 2026 – Economy: https://hai.stanford.edu/ai-index/2026-ai-index-report/economy
• Anthropic – Introducing Claude Sonnet 4.5: https://www.anthropic.com/news/claude-sonnet-4-5
• IBM Quantum documentation – Processor types: https://quantum.cloud.ibm.com/docs/en/guides/processor-types
• arXiv – The Hierarchy of Agentic Capabilities: https://arxiv.org/abs/2601.09032
• arXiv – The Long-Horizon Task Mirage: Diagnosing Where and Why Agentic Systems Break: https://arxiv.org/abs/2604.11978
• arXiv – LongCLI-Bench: A Preliminary Benchmark and Study for Long-horizon Agentic Programming in Command-Line Interfaces: https://arxiv.org/abs/2602.14337
• arXiv – ARC-AGI-3: A New Challenge for Frontier Agentic Intelligence: https://arxiv.org/abs/2603.24621
• Pew Research Center – How the Global Religious Landscape Changed From 2010 to 2020: https://www.pewresearch.org/religion/2025/06/09/factors-driving-religious-change-2010-2020/