Author’s Note
Please note that this paper was generated with the help of AI, based on my instructions and a long series of broad conversations exploring Spatz, fair conversation, objectivity, and AGI.
By Desmond Sherlock – Spatz.ai
Abstract
This paper presents a speculative research hypothesis rather than a claim. It proposes that fair conversation may provide a richer environment for developing general intelligence than the bounded environments traditionally used in many AI systems, such as games.
My proposition is not that current AI research is misguided. Rather, I suggest that an important learning environment may have been underexplored: structured, fair conversation.
For approximately forty years, I have been developing an evolving conversational algorithm called Spatz. Initially conceived as a way to improve human conversations and resolve workplace micro-conflicts, I am also speculating that the same framework could provide a shared learning environment in which both humans and existing AI become progressively more objective (less subjective) together.
If correct, this may represent one complementary pathway toward Artificial General Intelligence (AGI).
The Observation
Many of the landmark achievements in modern AI have emerged from environments with explicit rules and measurable objectives.
Games have proven to be extraordinary research environments because they provide:
- clear rules,
- measurable outcomes,
- repeated feedback,
- opportunities for continual improvement.
The success of systems such as AlphaGo demonstrated that remarkable intelligence can emerge through repeated interaction within these structured environments.
However, I speculate that games also possess an inherent limitation.
They are bounded.
The rules cannot generally be negotiated while playing.
The objectives remain fixed.
The action space, while enormous, is ultimately constrained.
Conversation appears fundamentally different.
Conversation is effectively unbounded.
It encompasses unlimited topics, changing objectives, evolving social norms, uncertainty, ambiguity, disagreement, creativity, negotiation, ethics, teaching, learning, cooperation, competition, misunderstanding, deception, trust, bias, dogmatism, reconciliation and countless other dimensions of human intelligence.
If AGI seeks general intelligence, perhaps we should investigate learning environments that are themselves increasingly general.
A Different Research Question
Rather than asking:
Can AI become more intelligent by mastering games?
I propose asking:
Can humans and AI become more objective by mastering “fair” conversation?
This is an importantly different question.
The objective is not persuasion.
It is not conversion by conversation, but rather that convergence by a fair and objective conversation is the objective.
Spatz
Spatz is an evolving algorithm for fair conversation. (Playbook and Charter)
Its purpose is not to determine who is right.
Its purpose is to improve the quality of the conversational process itself.
The central principle is simple:
Become more objective by objecting to objectionable behaviour.
Importantly, Spatz separates behaviour during conversation from the content.
Participants remain free to disagree with ideas.
The system instead focuses on objectionable conversational behaviour that reduces fairness and objectivity.
The algorithm introduces structured, proportional interventions through progressively stronger conversational signals, together with transparent review when required.
Rather than eliminating disagreement, Spatz attempts to preserve disagreement while improving the behavioural quality of the discussion.
Why Fair Conversation?
Conversation is the medium through which almost every human intellectual activity occurs.
Scientific discovery.
Engineering.
Education.
Medicine.
Politics.
Business.
Families.
Relationships.
Innovation.
Conflict.
Collaboration.
If conversation itself becomes progressively fairer, perhaps every one of these activities can improve.
My speculation is that fairness is not simply a social virtue.
It may also be an important learning mechanism.
Without fairness, conversation easily degenerates into domination, manipulation, dogmatism, intimidation, avoidance or tribalism.
Those behaviours may reduce objectivity.
If they can be systematically identified and reduced, conversation itself may become a richer environment for learning.
Humans and AI Improving Together
Current discussions about AGI often describe humans training AI.
Or AI eventually surpassing humans.
My speculation differs.
I propose a shared learning environment.
Existing AI participates alongside humans.
Humans contribute experience, values, judgement and creativity.
AI contributes consistency, memory, synthesis and analytical capability.
Spatz provides the conversational framework.
Over repeated interactions, both humans and AI learn from one another.
The objective is not merely improving AI.
The objective is improving the quality of reasoning for everyone participating.
A Working Hypothesis
One possible formulation is:
If there is a systematic way—including the use of existing AI—to improve how humans think together through structured, fair conversation, then that same framework may also improve how AI interacts, reasons and learns.
A further speculation follows.
Perhaps AGI will not emerge solely from increasingly capable models.
Perhaps it will emerge from humans and AI continually improving together through fair conversation.
Conversation as the General Environment
Games have demonstrated that intelligence can emerge within bounded environments.
Conversation may represent a more general environment.
Unlike games, conversation continually changes.
Its “rules” evolve.
Its objectives evolve.
Its participants evolve.
Its subject matter is effectively unlimited.
Perhaps conversation is not merely an interface to intelligence.
Perhaps conversation is one of the environments through which increasingly general intelligence develops.
Proposed Research
This proposal is intentionally empirical.
I am not claiming that Spatz leads to AGI.
I am proposing that the hypothesis deserves testing.
Given appropriate funding (for example, approximately two years and sufficient resources), I would propose a pilot involving approximately ten teams of ten participants using existing AI within the Spatz framework.
Possible research questions include:
- Do teams become more objective over time?
- Does objectionable conversational behaviour decrease?
- Does decision quality improve?
- Does AI improve its understanding of conversational behaviour through repeated participation?
- Do humans and AI demonstrably improve together?
The resulting longitudinal dataset could provide evidence supporting, refining or rejecting the hypothesis.
Invitation
I recognise that this proposal is speculative.
That is intentional.
New research programmes necessarily begin with speculation before becoming hypotheses and, eventually, evidence.
I therefore offer these ideas not as conclusions but as questions.
If the assumptions are flawed, I would welcome informed criticism.
If some assumptions appear plausible, I would welcome suggestions for how they might be tested rigorously.
After approximately forty years of developing the Spatz framework, I believe the time has come to expose these ideas to researchers whose expertise extends far beyond my own.
Whether this hypothesis ultimately proves correct or not, I hope it contributes to a broader discussion:
Could fair conversation itself become one of the most important learning environments for both humans and artificial intelligence?
Object123 
Leave a comment