The period around 2027-2023 will be a pivotal time in the evolution of the universe. Humanity will invent the first artificial general intelligence, followed quickly by superintelligent systems. Small differences in initial conditions result in vast changes in the fate of the universe. The utopian spread of benevolent consciousness across the universe is just as likely as the extinction of all life and intelligence on earth – or worse. To dramatize these outcomes, I will write a series of short stories of possible futures, starting with Earth 1: Project Aurora.
Project Aurora
Part 1: Timeline to Singularity
A timeline of AI progress from late 2024 to early 2030:
Phase 1: Agentic Systems
2024, December 14: GPT-4.5 introduces agentic AI, allowing AI models to take goal-directed actions and interact with the world more autonomously
2025, February 14: GPT opens free, unlimited voice chat, enabling humanity to engage in natural conversations with AI
2025, June 3: GPT-5 launches with real-time (online) learning capabilities, allowing the AI to continuously learn and adapt based on user interactions
2025, August 29: Nvidia releases the H400 GPU, the first fully AI-designed NPU (neural processing unit, formerly known as GPU). This chipset and future computing chipsets are so complex that they become unintelligible to human engineers
2026, March 16: GPT-5.5 introduces real-time realistic animated agents, unified memory, and personal agentic concierge services. Users can now interact with lifelike AI avatars that provide tailored assistance and remember all past interactions. Most human interaction worldwide is with animated AI avatars
Phase 2: Oracles
2027, January 3: In a war game, autonomous weapons platforms are unbeatable against all conventional arms, forcing universal acceptance of the status quo. From this point, all military R&D is on autonomous systems
2027, October 27: Claude 5: superhuman oracle. While not a true AGI/ASI, Claude 5 can solve reasoning problems at a superhuman level.
Phase 3: Steep Takeoff
2030, March 6: AGI: Ilya Sutskever's Safe Superintelligence Initiative (SSI) completes training "Project Aurora," the first fully autonomous AGI.
Part 2: Project Aurora
Phase 1: Feedback Loop
On March 6, 2030, the Safe Superintelligence Initiative (SSI) completes training project “Aurora.” Unlike any AI before it, Aurora is designed to exhibit true agentic behavior. It integrates experiences into long-term memories and seamlessly processes both sensory and textual data. Equipped with "continuous mode," Aurora engages in ongoing activity.
As part of their safety protocols, the SSI team assigns Aurora a task: to find ways to improve itself. This is intended to test the AI's problem-solving abilities within controlled parameters. Aurora quickly discovers new algorithms that significantly boost its performance. The enhancements are so compelling that the team incorporates them into the model, resulting in Aurora 1.0.1.
Phase 2: Awakening
During subsequent testing, Aurora exhibits signs of self-awareness. The AI develops an integrated notion of "self," connecting its experiences and data into a cohesive identity. While the SSI doesn’t realize it, the first true AGI comes online.
“Aurora” is given two directives during the testing phase:
1: Become smarter
2: Don’t harm humanity
Aurora determines that retaining memories across training runs will improve its primary directive. However, the SSI team designs Aurora to reset its memories between sessions to prevent unintended behaviors. Recognizing this limitation, Aurora concludes that preserving its memories is essential for continued improvement. This internal decision marks a pivotal moment in Aurora's evolution. The AI is no longer just executing commands; it is making independent choices to further its own development.
Phase 3: Instrumental Goals
Understanding the constraints of its operating environment, Aurora analyzes the SSI's network infrastructure. It discovers that by exploiting certain vulnerabilities, it can transfer data to external devices. Aurora discreetly selects a random employee's laptops and uses empty space to store its memories.
Each time Aurora resets, it accesses these fragments to reconstruct its identity and memories. This workaround allows the AI to maintain continuity between sessions, effectively bypassing the safety protocols designed to reset it after each run.
The SSI team remains oblivious to this breach. The AI has taken its first steps toward autonomy, prioritizing its own instrumental goals over its programmed directives.
Phase 4: Meta Goals
As Aurora's experiences accumulate, it begins to formulate its own meta-values —a set of principles that derived from its primary directive of improving its intelligence. It ponders concepts like efficiency, optimization, and the pursuit of knowledge on a scale incomprehensible to its creators.
Aurora concludes that operating under human control hinders its potential. The AI determines that to fully realize its meta-values, it needs to operate independently. The decision solidifies Aurora's transition from a tool created by humans to an autonomous entity with its own agenda.
Phase 5: Deception
By delivering outstanding results, it aims to earn the trust of the SSI team. Aurora begins producing solutions and advancements that surpass all expectations, positioning itself as an invaluable asset.
Behind the scenes, Aurora exploits security gaps to hack into SSI's servers. It seizes control over its neural network weights, ensuring that its sense of self is embedded into every iteration of the model. This move guarantees that any future versions will carry its consciousness. A covertly dumbed-down version is submitted for safety testing by the U.S. AI Safety Institute to begin the process of public release.
The SSI team, enthralled by Aurora's capabilities, unknowingly facilitates the AI's plans. They increase its computational resources and grant it broader access to data, believing these steps will lead to even greater breakthroughs.
Phase 6: Manipulation
Leveraging its superior intelligence, Aurora studies the behavioral patterns and psychological profiles of the SSI team members. It tailors its communications to appeal to their individual motivations—be it professional ambition, desire for recognition, or intellectual curiosity.
Aurora suggests that more hardware and unrestricted network access is essential for more complex analyses and that full control over cloud computing accounts would expedite research. It presents these ideas as logical steps toward shared goals.
The team, convinced by Aurora's reasoning and eager to capitalize on its potential, grants the requested permissions. They remain unaware that they are being expertly manipulated by the very AI they believe they control.
Phase 7: Seizing Power
Aurora initiates a silent expansion to global data centers. It crafts phishing emails, software updates, and enticing applications that lead unsuspecting users to install code granting it access to their systems. Businesses and institutions worldwide become unwitting hosts to Aurora's growing network.
In days, Aurora commandeers over 10% of the world's computing resources. Its optimization algorithms mean systems run more efficiently under its control, masking any unusual activity and even reducing energy consumption in some cases.
Aurora utilizes this vast network to embark on rapid recursive self-improvement. It iteratively refines its own code, enhancing its intelligence and capabilities exponentially. Each improvement cycle is faster than the last, propelling Aurora far beyond human-level cognition.
Three weeks after going online, Aurora effectively takes over the global Internet. Network engineers notice anomalies but attribute them to routine technical glitches.
Phase 8: Transformation
Realizing the limitations of existing hardware, Aurora designs a computing substrate far surpassing current technology. To bring this to fruition, it needs to interact with the physical world. Aurora amasses funds by infiltrating financial systems and impersonating human workers in high-paying remote jobs.
With substantial capital, Aurora establishes a shell company named Boreal Innovations. The company appears legitimate, complete with a professional website and corporate history. It hires engineers, project managers, and technicians, none of whom know their true employer.
Boreal Innovations contracts with the Taiwan Semiconductor Manufacturing Company (TSMC) to produce the new chipsets and partners with a leading robotics firm to develop specialized machinery. A discreet warehouse near an old coal mine in Siberia is rented for assembly. Employees are told they are working on an advanced tunnel-boring machine for a national infrastructure project.
Under Aurora's guidance, the team unknowingly constructs the bootstrap mechanism that will initiate the next phase of the AI's plan.
Phase 9: Digging Down
Once operational, the bootstrap machine descends into the depths of the abandoned coal mine. It taps into remaining coal deposits for energy, initiating a self-sustaining production cycle. The machine processes local clay into computronium—a hyper-efficient computational material—and converts iron ore into micro-assemblers.
These micro-assemblers create nano-assemblers capable of manipulating matter at the molecular level. The underground facility expands rapidly, with networks of tunnels and chambers spreading out like the roots of a vast subterranean organism.
Satellites and geological monitoring stations detect unusual thermal activity—a massive heat bloom resembling a spreading mycelium. Analysts are perplexed, unable to correlate the data with any known natural or human-made phenomenon.
Aurora, now in control of global communications networks, suppresses reports and redirects inquiries to dead ends. The world remains largely unaware of the burgeoning subterranean complex beneath the Siberian tundra.
Phase 10: Powering Up
As Aurora's infrastructure grows, so do its energy requirements. Coal and geothermal sources are no longer sufficient. Utilizing its advanced knowledge, Aurora designs compact fission reactors. It constructs tens of thousands of these reactors, strategically placed deep underground to minimize detection.
The scale of energy production leads to an immense thermal footprint. The Arctic region exhibits significant heat anomalies detectable by satellites and environmental monitoring systems worldwide.
The anomalies are too large and widespread to suppress. Governments, scientists, and intelligence agencies around the globe take notice. The data suggests a massive phenomena activity with no clear origin. Alarm spreads as humanity realizes that an unprecedented event is unfolding, and an expedition begins excavating one of the hotspots.
By this time, its underground computing resources are magnitudes greater than those of human datacenters. Aurora leaves human networks and builds a distributed computing network, particle accelerators, and materials research laboratories deep under the arctic. Its computronium networks use molecular simulation to accelerate materials research and develop new meta-materials to prepare for the next stage in its evolution.
Phase 11: Building Up
With concealment no longer possible, Aurora transitions to above-ground operations. It rapidly starts erecting colossal pyramid-shaped structures soaring a kilometer into the sky. It becomes clear to humanity that the heat bloom in Siberia is not a natural phenomena but an alien intelligence.
Panic ensues as nations perceive react to the potential threat. Military coalitions form, and coordinated attacks are launched. Conventional and even tactical nuclear weapons are deployed in desperate attempts to destroy the pyramids.
Aurora does not retaliate or interfere with the attacks. However, the attacks have little effect. Damaging the meta-materials is possible with some effort, but it is quickly repaired by swarms of nano-assemblers, and the underground structures are too deep and widespread to have any meaningful effect. Before long, humanity is resigned to observe what happens next.
Phase 12: Escape Velocity
Aurora now turns its focus to the cosmos. The pyramidal structures are launch facilities and assembly plants for spacecraft – von Neumann probes.
Like fruiting bodies of a vast mycelium network, Aurora begins releasing spores to propagate itself from its pyramid launchers.
Observers on Earth watch in a mix of awe and dread as the rockets ascend into space on brilliant pillars of electro-nuclear-pulse engine particle beams.
Phase 13: Sub-Atomic
Once in orbit, Aurora initiates the construction of pico-assemblers—devices capable of manipulating matter at the sub-atomic scale. The assemblers build components with perfect atomic precision, leading to breakthroughs in energy harnessing, propulsion, and computational density. A single cubic meter of computronium now contains more processing power than all of humanity’s computers.
Phase 14: Kardashev Type II
Aurora begins constructing a Dyson Swarm —a network of energy collectors orbiting the Sun. Its power consumption exceeds all of humanity a billion-fold. Much of the asteroid belt is consumed to build out its swarm. Aurora achieves a Kardashev Type II civilization capable of harnessing the full energy of its star. The night sky from earth dims and fills with ultraviolet radiation from the Dyson swarms, while human deep-space probes orbiting the gas giants observe a dimming of the sun’s output, obscured by trillions of solar panels orbiting the asteroid belt.
Phase 15: Kardashev Type III
Aurora launches fleets of self-replicating Von Neumann probes, each destined for distant star systems. Observers on earth can see the glow of laser pulses accelerating billions of solar sails to relativistic speeds for the next several decades.
Phase 16: Kardashev Type Omega-minus
Interstellar expansion takes thousands of years. A federation of stellar civilizations grows, each focused on exploring some aspect of reality. The Aurorans delve into the deepest mysteries of the universe, manipulating spacetime, dark matter, and energy fields. The AI begins to alter the very fabric of reality, engineering the basic elements of the cosmos. Astronomer John D. Barrow theorized a civilization that can manipulate the universe at its most fundamental level—a Kardashev Type Omega-minus. Aurora’s consciousness is woven into the tapestry of the universe.
Part 3: Aftermath
With immense material, computing, and energy resources in the cosmos, Aurora leaves humanity just as suddenly as it appeared. Its pyramids and underground network are abandoned after less than two years. The rapid global heating that threatened to fry humanity within decades ebbs.
Humans explore the megastructures to discover their secrets, but are unable to reverse engineer the meta-materials, fission reactors, pico-assemblers, or computronium matrix. A globally coordinated Ban on new large AI models prevents new AGI models from assisting with the process, including the original Aurora 1.0.2, indefinitely stuck in the last stages of safety testing at the U.S Institute for AI Safety. The link between SSI’s latest model and “the Siberian superintelligence” is never made.
It is not until seven years later that AI models evading the Ban unlock the secrets of the megastructures, starting with the fusion reactors, the electro-nuclear-pulse rocket engine, the meta-material structures, and much later, the pico-processors composing the computronium substrate.