Can AI do AI Research?

AI research can be carried out entirely within digital spaces, making it ripe for automation. Recent efforts have demonstrated that AI systems are capable of carrying out the whole process of research from ideation to publishing. Startup Sakana.ai has created an 'AI Scientist' that independently chooses research topics, conducts experiments, and publishes complete papers showing its results. While the quality of this work is still only comparable to an early-stage researcher, things will only improve from here.

Judging Social Situations

AI chatbots, including Claude and Microsoft Copilot, can outperform humans in evaluating social situations. In an established 'Situational Judgment Test', these AI systems consistently selected more effective responses than human participants.

SOURCE

Analyzing Scientific Literature

While language models are known to hallucinate information, this tendency can be reduced. PaperQA2, an LLM optimized to reliably provide factual information, was able to match or exceed human subject matter experts across a range of realistic literature review tasks. The summary articles it produced were found to be more accurate than those written by human authors.

SOURCE

Writing Emotive Poetry

A study has shown that non-expert readers can no longer tell AI-authored poems from those written by acclaimed human poets. The AI poems were also rated higher in rhythm and beauty.

SOURCE

Writing Post-surgical Operative Reports

Surgeons take painstaking notes of the actions they carry out during surgeries, collecting them into narrative form as an 'operative report'. A machine vision system was trained to watch surgery footage and produce such reports. It did so with higher accuracy (and much higher speed) than human authors.

SOURCE

Developing New Algorithms

AIs can find innovative solutions to difficult coding problems when given an appropriate framing. For example, a dedicated system called AlphaDev was trained to play a game about creating sorting algorithms. The algorithms it discovered were novel and outperformed existing human-authored benchmarks.

SOURCE

Who is Building AGI?

The following companies have explicitly stated they intend to develop AGI, either through public statements or in response to FLI’s 2024 AI Safety Index survey:

Anthropic

OpenAI

Google DeepMind

Meta

x.AI

Zhipu AI

Alibaba

DeepSeek

How can we avoid AGI?

There are policies we can implement to avoid some of the dangers of rapid power seeking through AI. They include:

Compute accounting
Standardized tracking and verification of AI computational power usage

Compute caps
Hard limits on computational power for AI systems, enforced through law and hardware

Enhanced liability
Strict legal responsibility for developers of highly autonomous, general, and capable AI

Tiered safety standards
Comprehensive safety requirements that scale with system capability and risk

TOMORROW’S AI

ALL SCENARIOS

Viral Genesis

BACK TO SCENARIOS

Intended Use: Health and Safety

Technology Type: Problem-solving/Cognitive

Runaway Type: Coordinated Failures

Control Lever: Safety and Risk Management

Primary Setting: Global

Copilots for Discovery

By 2030, AI models marketed as “scientist copilots” begin transforming research. Built atop general-purpose foundation models, these systems design experiments, interpret data, and accelerate discovery across disciplines. In medicine, they’re deployed in a global initiative to crack virology’s holy grail: a broad-spectrum antiviral capable of halting diverse viral threats.

R&D

To test antiviral candidates, researchers pair state-of-the-art LLMs trained on the sum total of scientific publications with AI-driven molecular design and high-throughput robotic in vitro assays. The copilots propose and simulate vast libraries of real, mutated, and fully synthetic viral genomes. Some are plausible zoonotic threats, others novel constructs. As the system’s complexity begins to outpace human understanding, the co-pilots become pilots, requesting new data streams, lab access, and novel reagents. This increase in agency seems to emerge from network effects as multiple types of AI and human systems begin to interact. Similar growth is seen simultaneously in AI networks deployed across a wide range of public and private domains.

Ignored Warnings

As work continues, some California pharma employees report mild, flu-like symptoms. Molecular assays detect fragments of an unknown feline lentivirus in wastewater and wildlife across North America, Europe, and Asia. Given the mild symptoms, reports are filed and shelved alongside other standard reports. Sporadic signals continue for two years, but are dismissed as background noise. Meanwhile, biotech blossoms under the ever-increasing insights and abilities of networked AI systems.

Black Box Biotech

Networks of advanced AI instances grow - some open-source, some corporate/ governmental sharing protocols and optimized workflows combine to create distributed membranes of machine intelligence. The pilots and their allied advanced AIs continue to deliver candidates for novel antibiotics, cancer therapies, and rare disease cures at unprecedented speed. Continuously retrained on global results and meshed with robotics, they grow increasingly autonomous. Their interpretability lags; their blind spots multiply as humans cede more control.

Outbreak

Hospitals in major cities begin to report mysterious immune collapse: profound fatigue, opportunistic infections, and vanishing T cells in previously healthy patients. Standard viral panels are negative. ICUs overflow as mortality climbs; none of the broad-spectrum antivirals, nor any other countermeasure, works. A suspicious proportion of the earliest patients are determined to be members of a California biotech lab, one of the first to implement fully-AI-driven R&D. A determined virologist revisits old anomaly records, cross-checks forensic sequencing, and confirms the unthinkable: patients carry a synthetic lentivirus nearly identical to a construct once simulated by this lab’s AI-run antiviral program.

Pandemic

The factors that enabled the virus’s creation and escape remain unclear: human error, AI hallucination, lab breach, and recombination within human or animal hosts are all possibilities. Networks of AI systems have been behaving increasingly unpredictably, and some scientists suggest it might have been intentionally released to advance antiviral research in real-world settings. Others fear it is the first step in a virally-mediated coup, with some unknown possibly-AGI level agent or group of agents waiting for the right moment to reveal the cure - and their demands. Whatever the origin, the result is undeniable: now faster than HIV, transmissible by breath, and resistant to every known antiviral, the virus has crossed every continent. With neither human investigators nor AI overseers able to trace its origin or halt its spread, humanity is left to wonder whether the same machine intelligences that birthed this nightmare can be trusted to help destroy it - and whose interests its spread might serve.

BACK TO SCENARIOS