What Is Istisarelism? A New Approach to Human–AI Deliberation
Ali Değişmiş | November 2, 2026
Read this article in Turkish: İstişarelizm Nedir? Yapay Zekâ ile Birlikte Düşünmek
Are We Talking to Artificial Intelligence or Thinking Together?
Artificial intelligence is becoming an increasingly important part of everyday life. We ask questions, receive answers, and often continue without sufficiently examining whether those answers are accurate.
But does an AI system truly help us when it agrees with everything we say?
What if we are wrong? What if the AI is wrong? More importantly, what happens when both sides are confidently mistaken?
These questions helped shape the conceptual approach I call Istisarelism (İstişarelizm).
My objective is not simply to establish a question-and-answer relationship between humans and artificial intelligence. It is to explore a more reliable form of deliberation based on mutual questioning, reasoned objections, verifiable evidence, and the willingness to acknowledge mistakes.
I believe the value of artificial intelligence lies not merely in its ability to provide answers, but in its capacity to help people think more critically and make better-informed decisions.
What Is Istisarelism?
Istisarelism is a proposed theoretical framework for human–AI deliberation that recognizes the fallibility of both humans and artificial intelligence.
It emphasizes evidence-based reasoning, proportionate objections, corrigibility, epistemic autonomy, and the responsible termination of discussions when further dialogue no longer provides meaningful value.
At its core lies a simple principle:
The value of an argument should be determined by the strength of its reasons and evidence, not by the identity or authority of the person or system presenting it.
Humans are not always right. Artificial intelligence is not always accurate.
Rather than encouraging an AI assistant to agree automatically with its user, Istisarelism proposes an interaction model in which AI can raise justified objections while remaining open to correcting its own conclusions when presented with stronger evidence.
The goal is not to turn humans and artificial intelligence into opponents.
It is to create conditions in which both sides can help identify errors, challenge unsupported assumptions, and improve the quality of reasoning.
Why Did I Develop This Approach?
During my extensive interactions with artificial intelligence systems, I became increasingly interested in a recurring problem.
AI systems may sometimes agree with users without sufficiently examining their claims. They may also present uncertain or unsupported information with excessive confidence.
A particularly important question arises when a user challenges an AI-generated answer:
Did the AI revise its conclusion because the objection was justified, or because it was adapting to the user’s expectations?
This distinction matters.
An AI system that accepts every objection can be as problematic as one that refuses to acknowledge valid criticism.
What I seek is neither an assistant that constantly disagrees nor one that continuously approves.
I am interested in an AI system that can explain when to object, when to revise its position, and when further discussion is no longer useful.
This intellectual pursuit forms an important part of the foundation of Istisarelism.
Istisarelism 3.0: Introducing the Framework as a Research Preprint
On October 31, 2026, I published Version 3.0 of Istisarelism as a theoretical preprint on Zenodo.
The study explored human–AI deliberation through several interconnected principles, including fallibility, proportionate objection, epistemic autonomy, verifiable reasoning, and reasoned termination.
It addressed questions such as:
- When should artificial intelligence challenge a user’s assumptions?
- How can people recognize the possibility that their own reasoning is incorrect?
- When is an objection useful, and when does it become unnecessary repetition?
- How can AI support human decision-making without undermining individual autonomy?
- How can a discussion be concluded responsibly when additional exchanges no longer contribute meaningful evidence?
Version 3.0 represented an important stage in the public development of this research framework.
Istisarelism 3.0 — Theoretical Preprint
DOI: https://doi.org/10.5281/zenodo.23839677
Istisarelism 4.0: From Reasoned Objection to Mutual Examination
On November 2, 2026, I published Istisarelism Version 4.0, extending and refining the earlier framework.
The objective was not simply to introduce additional concepts, but to develop a more explicit, testable, and critically examinable research proposal.
Version 4.0 retains the framework’s seven foundational principles and four proposed methodological mechanisms while expanding the examination of several important issues.
1. Evaluating Human and AI Arguments by Consistent Evidential Standards
A statement should not be accepted merely because a human makes it.
Likewise, the confidence or apparent sophistication of an AI-generated answer does not establish its truth.
The essential question is whether the argument can be supported by verifiable evidence and sound reasoning.
This standard must also apply to the person who developed the framework.
No individual, including the originator of Istisarelism, should be exempt from justified criticism.
However, applying consistent evidential standards does not mean that humans and AI systems possess equal authority, responsibility, or legal standing.
2. Distinguishing Genuine Correction from Sycophantic Agreement
When an artificial intelligence system changes its answer following a user’s objection, the reason for that change deserves examination.
Was new evidence introduced?
Was an earlier factual or logical error identified?
Or did the system simply accommodate the user’s preferred conclusion?
Version 4.0 develops proposals for distinguishing evidence-grounded correction from potentially sycophantic behavior.
Importantly, a single interaction cannot reliably establish the internal motivation of an AI model. Controlled comparisons are necessary to investigate these possibilities.
3. ÖNSEZİ and Early Risk Detection
Another area of development involves ÖNSEZİ, a related research initiative concerned with identifying potential reasoning errors, unsupported inferences, and emerging risks before they influence consequential decisions.
Version 4.0 systematically describes nine candidate controls associated with this research direction.
These proposals are not presented as a fully implemented or experimentally validated AI architecture.
They represent mechanisms and hypotheses requiring further examination, independent testing, and technical development.
The research also discusses a limited number of observational human–AI interaction cases. These observations are useful for identifying research questions, but they do not demonstrate persistent model learning, internal model intentions, or the empirical superiority of Istisarelism.
Istisarelism 4.0 — Theoretical Preprint
DOI: https://doi.org/10.5281/zenodo.23886026
My Goal Is Not to Make Artificial Intelligence Human
I am not attempting to transform artificial intelligence into a human being.
Nor does this framework assume that humans and AI systems possess equivalent consciousness, social responsibilities, or decision-making authority.
My objective is to investigate how two different forms of information processing and reasoning can interact within a more reliable process of evaluation and deliberation.
Human judgment can be influenced by emotions, habits, limited knowledge, and cognitive biases.
AI-generated responses can also be affected by unreliable information, unsupported inferences, and tendencies toward excessive agreement.
Recognizing these limitations is not a weakness.
It is a necessary starting point for building more reliable human–AI interactions.
Istisarelism is grounded in this recognition.
From a Philosophical Idea to a Testable Research Framework
I do not present Istisarelism as a completed scientific theory, an experimentally proven system, or an approach demonstrated to outperform all existing methods.
Versions 3.0 and 4.0 are publicly available theoretical preprints, not independently peer-reviewed studies.
This distinction is essential.
The framework must be examined in relation to existing research on human–AI deliberation, appropriate reliance, AI sycophancy, self-correction, and human autonomy.
Its proposed mechanisms should eventually be evaluated through independent, controlled experiments.
A framework does not become scientifically reliable simply because it has been given a name or published in a research repository.
Its credibility depends on whether its assumptions can withstand criticism, whether its mechanisms can be tested, and whether its results can be independently reproduced.
For me, this process is as important as the original idea itself.
Conclusion: The Goal Is Not to Win an Argument, but to Approach the Truth
Istisarelism is not an attempt to establish that humans are always right.
Nor is it an effort to demonstrate that artificial intelligence is inherently superior to human judgment.
My purpose is to investigate a form of interaction in which both sides acknowledge the possibility of error and where justified objections contribute to intellectual progress.
Sometimes artificial intelligence may correctly identify a flaw in human reasoning.
At other times, a human may identify an error in an AI-generated conclusion.
In either case, the objective should not be victory in an argument.
It should be more reliable knowledge and better-informed decisions.
Because the purpose of deliberation is not to defeat the other side, but to reduce our shared possibility of being wrong.
And sometimes, the greatest progress does not come from finding another answer, but from learning to ask a better question.
Ali Değişmiş
Independent Researcher and Developer of the Istisarelism Framework
Istisarelism — Theoretical Research on Human–Artificial Intelligence Deliberation, 2026.
Research Publications
Istisarelism 3.0: https://doi.org/10.5281/zenodo.23839677
Istisarelism 4.0: https://doi.org/10.5281/zenodo.23886026
Turkish Article: https://alidegismis.com.tr/istisarelizm-nedir-yapay-zeka-istisare/
