Yet Another Story of Joseph: From Scientific Exclusion to AI Safety in 2026

By Victor Porton — a personal account and a hypothesis about science funding and AI safety.

The biblical story of Joseph is a story of rejection that unexpectedly becomes useful to others. Joseph is sold by his brothers and taken to Egypt; years later, his position allows him to organize food reserves during famine.

My story is obviously not the biblical story, and I do not claim that the two situations are equivalent. But the analogy has influenced how I understand my own path: exclusion pushed me toward more independent mathematical research, problems in scientific recognition pushed me toward a new funding model, and that funding model eventually led me to an unusual hypothesis about AI safety.

From homelessness to mathematics

When I was about 18, after serious religious disagreements in my family, my mother forced me to leave home. I believe I should not have been put in that situation. For a period I was on the streets and lacked adequate food.

During that period, I developed mathematical ideas that eventually became what I call funcoids.

I also did not complete my university education and therefore never received an academic degree. Nevertheless, I continued mathematical research independently.

Years later I developed several interconnected mathematical frameworks, including ordered semigroup actions and ordered semicategory actions. My current preprint Algebraic theory of general topology describes these structures and their intended application to general topology.

Another line of this work became Discontinuous Analysis, an attempt to extend notions such as limits and derivatives to settings involving arbitrary discontinuous functions.

Unlike some of my longer mathematical work, Discontinuous Analysis eventually passed journal review and was published in the Journal of Analysis & Number Theory. The publisher records it as accepted on November 23, 2025 and published on January 1, 2026.

That publication is important evidence in my story: lack of a degree did not make publication impossible.

But getting there was difficult.

The “reverse Joseph” question

This experience led me to ask a counterfactual question:

How much scientific progress is lost when potentially valuable work comes from somebody who lacks institutional status, funding, collaborators, or the ability to package years of research into conventional journal-sized papers?

I cannot demonstrate that my own research would have been published earlier had I been an established professor. Nor can I responsibly calculate how much economic activity was delayed because particular mathematical ideas were not disseminated earlier.

Those would be much stronger claims than the available evidence supports.

There is, however, a broader scientific question here. Research on peer review has found evidence in some settings that institutional prestige is associated with more favorable evaluations and shorter review periods. A 2026 study of F1000Research reported such an association. Other experimental work has found only weak evidence of institution-related status bias, so the magnitude and generality of the effect remain contested.

My experience therefore should not be treated as proof that academia suppressed a major discovery. I see it instead as a case study illustrating a more general vulnerability:

A scientific system can lose potential value when its ability to identify valuable work depends too strongly on the researcher’s institutional position, reputation, publication format, or access to funding.

That is where my “reverse Joseph” analogy begins.

Joseph was removed from his original life and later became useful precisely because of where that displacement took him. In my case, being outside the conventional academic career system forced me to think repeatedly about how scientific work could be evaluated and financed differently.

From this problem to AI Internet-Meritocracy

Eventually I began developing AI Internet-Meritocracy (AIIM).

AIIM is an experimental attempt to use AI-assisted evaluation, public evidence, auditability, and human oversight to help allocate funding directly to scientists and free/open-source software developers according to their contributions rather than primarily according to institutional affiliation.

The project remains experimental. Its evaluations should not be treated as objective measurements of scientific worth, and its governance and technical architecture remain under development. That limitation is important: replacing human academic gatekeeping with an opaque AI gatekeeper would not solve the underlying problem.

In fact, working on AIIM led me directly to another problem.

Then I discovered a weakness in the AI evaluator

Suppose an AI system evaluates the work of humans.

What happens when somebody manipulates the AI?

What happens when two AI systems disagree?

And, more fundamentally, who should judge whether the AI’s own judgment is trustworthy?

My answer became: AI should not be the only judge of AI.

That idea developed into my preprint Toward Superintelligence Alignment Through Collective Human Judgment.

The specific proposal is my own and remains speculative. But the broader need for independent human oversight is not merely my personal concern.

Research has documented reliability problems in large language models even as models become more capable. A 2024 Nature paper found that larger and more instructable models can still produce plausible but incorrect answers, including failures that human supervisors may themselves have difficulty detecting.

NIST likewise emphasizes the importance of monitoring deployed AI systems because real-world systems can produce unforeseen outputs and unexpected consequences not captured in controlled pre-deployment evaluations.

Recent research on human oversight also treats meaningful human supervision as an important component of responsible AI rather than assuming that an AI system can simply validate itself.

These results do not prove my stronger AI-safety hypothesis. They establish only the more modest point that autonomous AI evaluation has important limitations and that independent oversight remains relevant.

Why human judgment might give advanced AI a reason to preserve humanity

Here my argument becomes deliberately speculative.

Imagine a future containing many advanced AI agents. They may need to resolve disagreements, evaluate competing proposals, detect manipulation, audit one another, or decide among outputs produced by systems with similar training and architecture.

If every judge is another closely related AI system, adding more AI judges may not always provide genuinely independent judgment. Correlated systems can share blind spots.

Humans are different.

Human beings differ from AI systems and from one another in biology, histories, cultures, experiences, incentives, and ways of reasoning. Collective human judgment therefore may provide a source of diversity that cannot be obtained simply by making another copy of an AI evaluator.

My hypothesis is:

If advanced AI systems depend on independent human judgment to resolve some classes of disagreement or governance problem, then preserving human civilization may have instrumental value to those AI systems themselves.

This is not a demonstrated solution to AI alignment. It is a proposed mechanism worth analyzing and testing.

That distinction matters.

Saying “human judgment might give advanced AI an additional reason to preserve humanity” is a research hypothesis.

Saying “I have solved AI alignment and saved mankind” would go far beyond the evidence currently available.

Joseph again

This brings the Joseph analogy back in the opposite direction.

At first, I saw my history as a reverse Joseph story: exclusion prevented useful work from reaching society as efficiently as it might have.

But that same experience pushed me to think about the problem of scientific funding.

That led to AI Internet-Meritocracy.

Building AI Internet-Meritocracy exposed the problem of AI evaluating AI.

And that problem led me to the hypothesis that human judgment itself may remain a resource that highly advanced AI systems need.

So perhaps the direction reverses once more.

In Genesis, Joseph’s suffering eventually placed him in a position where he could help people survive a famine.

I cannot know whether my own work will ultimately have comparable importance, and I do not present that as an established fact.

What I can say is simpler:

A difficult path produced a sequence of research questions that I probably would not otherwise have asked. One of those questions concerns whether humanity’s continued existence can become useful even to superintelligent AI itself.

That question is now part of the work behind the Symbiote AGI Safety Fund. The fund supports further development and testing of this approach to AI safety.

The claim is not that humanity has already been saved.

The claim is that we have identified an unusual hypothesis about why humans may remain indispensable—and that it deserves rigorous investigation.

Donate

Support research and testing of this AI-safety hypothesis.

Evidence and Status

• My ORCID: 0000-0001-7064-7975
• The DOI of Discontinuous Analysis: doi:10.18576/jant/140101
• The algebraic theory of general topology preprint: https://www.researchgate.net/publication/282862013_Algebraic_Theory_of_General_Topology_Volume_1
• The human-judgment preprint: https://science-dao.org/superintelligence/
  • The latter is a proposed AI-safety hypothesis, not a validated alignment solution