Jingde Cheng

Since the proposal of the Turing Test in 1950 and Searle's Chinese Room argument in 1980, thought experiments have played a central role in philosophical discussions concerning intelligence, understanding, and artificial intelligence. The rapid development of expert systems and, more recently, large language models (LLMs), has significantly changed both the practical capabilities of AI systems and the questions that deserve philosophical investigation. In many real-world applications, the primary concern is no longer whether an AI system genuinely "understands" language in the philosophical sense, but whether it can reliably perform as a human expert whose knowledge and judgment are trusted by human being. Motivated by this changing landscape, this paper proposes a new thought experiment, “the Professor Consultation Room” (PCR for short), that isolates an independent epistemological problem concerning the attribution of expert cognition. In the proposed scenario, students interact exclusively through textual communication with an unseen respondent who is either a distinguished logic professor or an LLM-based expert system possessing all of the professor's logical knowledge but no cognitive abilities beyond the currently recognized capability boundaries of existing LLMs. The central question is whether finite question-answer interactions alone can provide sufficient evidence to determine whether the respondent possesses cognitive abilities exceeding those currently demonstrated by LLMs. Unlike the Turing Test, which evaluates whether a machine can imitate general human conversational behavior, and unlike the Chinese Room, which questions whether symbol manipulation constitutes semantic understanding, the PCR shifts attention to expert cognition and to the epistemological problem of identifying respondent through observable behavior. Rather than attempting to resolve the longstanding debate concerning machine understanding, this thought experiment formulates a new question concerning the behavioral identifiability of human advanced cognitive abilities in the age of LLMs. The primary contribution of the present paper is not to propose a new theory of intelligence, understanding, or cognition. Rather, it introduces a new thought experiment that formulates an independent epistemological problem concerning the publicly observable evidence for attributing expert cognition through finite expert-level interaction. Unlike previous thought experiments, the Professor Consultation Room shifts philosophical attention from the ontology of cognition to the epistemology of attributing expert cognition on the basis of finite expert-level interaction.

Automated theorem finding (ATF), originally advocated by Larry T. Wos as a research direction distinct from automated theorem proving (ATP), aims at discovering previously unknown theorems rather than proving a given conjecture. While the distinction between theorem finding and theorem proving was clearly recognized in the late 1980s, a fundamental question remained largely unexplored: What underlying logic system as the logical foundation is most appropriate for automated theorem finding? Beginning in 1994, Jingde Cheng initiated a long-term research program addressing this question and expanding the wide application of automated theorem finding. Challenging the prevailing assumption that classical mathematical logic is an adequate foundation for all forms of automated reasoning, Cheng argued that the essential characteristics of theorem finding are fundamentally different from those of theorem proving. In particular, due to the large number of implication paradoxes in classical mathematical logic, a massive amount of “garbage theorems” are generated during forward reasoning based on classical mathematical logic, making it fundamentally unsuitable for discovery-oriented reasoning. To overcome the limitations of classical mathematical logic, Cheng proposed a strong relevant logic approach to automated theorem finding and subsequently developed a comprehensive framework based on strong relevant logic including entailment calculus, forward deduction engines, relevant reasoning, predicate abstraction and suggestion, and theorem-interestingness measurement. Over the next two decades, this framework has gradually evolved into a systematic methodology, encompassing entailment calculus based on strong relevant logic, general-purpose forward deduction engines, parallel forward reasoning, relevant (deductive, inductive, abductive) reasoning, automated theorem finding in mathematics, predicate abstraction and suggestion algorithms, theorem-interestingness measurement, theory grid and distributed cooperative discovery, semi-lattice models of formal theories, epistemic operations, epistemic programming paradigm, knowledge appreciation, and general methodology for automatic discovery and prediction. This paper reviews the entire development process of this research program and points out that its most significant principal originality lies in establishing, for the first time in the world, a method for discovery-oriented reasoning supported by strong relevant logic as the underlying logical system. The core conclusion of this research program is that while classical mathematical logic provides a suitable logical foundation for theorem proving, it cannot support theorem finding at all, while strong relevant logic provides a more suitable logical foundation for discovery-oriented reasoning and automated scientific discovery. The ultimate goal of this research program is to achieve automated scientific discovery and automated scientific prediction with human as the agent.

The hot controversy about the role of logic in AI has been repeated so far and probably will continue on as usual. An important fact is that the “logic” as the center of the controversy is classical mathematical logic and/or its various extensions, though there are some “more logical” logic systems. Until recently, what is debated by the researchers working on the fundamentals of AI is, among other things, what role the classical mathematical logic and/or its various extensions plays in knowledge representation and reasoning. As a result, most work on the fundamentals of AI are directly or indirectly based on the classical mathematical logic. It is the time to investigate the role of logic in AI from a more general point of view. What notion plays the most fundamental role in knowledge representation and reasoning? This question is discussed in this paper from the viewpoint of logic. The proposition presented in the paper is that it is the notion of entailment that plays the most fundamental role in knowledge representation and reasoning because any reasoning must invoke it and many important issues in knowledge representation and reasoning have to find a calculus of entailment as the logical basis of their solutions. In order to establish a satisfactory logic calculus of entailment to underlie knowledge representation and reasoning, the paper proposes some new relevant logics and suggests some important research problems.

Can Artificial Intelligence (AI) ever fully surpass human intelligence? This question has become increasingly prominent following the remarkable achievements of systems such as AlphaGo, AlphaGo Zero, and large language models (LLMs). Many researchers and commentators argue that continuing advances in computational capability will eventually lead to artificial general intelligence (AGI) and perhaps even super-intelligence. This paper argues that contemporary discussions of AI superiority focus excessively on intelligence while largely neglecting three more fundamental issues: agency, evaluation criteria, and concept/problem formation. The paper distinguishes intelligence from agency and proposes the Agency Priority Thesis, according to which intelligence presupposes agency because every meaningful problem- solving activity requires an agent that establishes goals (problems) and evaluation criteria. Building upon this framework, the paper introduces four central theses: (1) the Agency Priority Thesis, according to which intelligence presupposes agency; (2) the Evaluation Criteria Thesis, according to which problem solving presupposes evaluation criteria; (3) the Evaluator Thesis, according to which evaluation presupposes an evaluator; and (4) the Problem-Posing Thesis, according to which the formulation of genuinely new problems is a higher-order intellectual capability than the solution of predefined problems. Based on these theses, three necessary conditions for full AI superiority are formulated and examined. The paper argues that computational systems cannot independently establish goals (problems), generate evaluation criteria, or create genuinely new conceptual frameworks. Consequently, none of the necessary conditions for full AI superiority can be satisfied. The paper concludes that AI systems may greatly extend and amplify human cognitive capabilities, but they cannot fully surpass human intelligence. The fundamental limitation is not computational power but the absence of autonomous agency.

In order to provide scientists with a computational methodology and some computational tools to program their epistemic processes in scientific discovery, we are establishing a novel programming paradigm, named ‘Epistemic Programming’, which regards conditionals as the subject of computing, takes primary epistemic operations as basic operations of computing, and regards epistemic processes as the subject of programming. This paper presents our fundamental observations and assumptions on scientific discovery processes and their automation, research problems on modeling, automating, and programming epistemic processes, and an outline of our research project of Epistemic Programming.

In many applications in computer science and artificial intelligence, in order to represent, specify, verify, and reason about various objects and relationships among them, we often need a right fundamental logic system to provide us with a criterion of logical validity for reasoning as well as a formal representation and specification language. Although different applications may require different logic systems, the fundamental logics must be able to underlie truth-preserving and relevant reasoning in the sense of conditional, ampliative reasoning, paracomplete reasoning, and paraconsistent reasoning. Based on our experiences, this paper shows that strong relevant logic can be used as the universal basis to construct various applied logics to satisfy the requirements. The paper discusses why any of the classical mathematical logic, its various classical conservative extensions, and its non-classical alternatives is not a suitable candidate for the universal basis to construct various applied logics, shows that strong relevant logic is a more hopeful candidate for the purpose, and presents our experiences on constructions of temporal relevant logics, deontic relevant logics, spatial relevant logics, and spatial-temporal relevant logics.

Scientific discovery has traditionally been regarded as one of the most sophisticated manifestations of human intelligence. While computer science has developed numerous programming paradigms for data processing, symbolic manipulation, numerical computation, and logical deduction, comparatively little attention has been devoted to the problem of programming the processes through which new knowledge is created. In the mid-1990s, Jingde Cheng initiated a research program known as “Epistemic Programming,” which explicitly proposed a programming paradigm for scientific discovery and knowledge creation. Unlike traditional programming paradigms, Epistemic Programming regards conditionals as the subject of computing, takes primary epistemic operations as basic operations of computation, and regards epistemic processes as the subject of programming. Over the subsequent two decades, the research evolved into a broader framework encompassing logical foundations, computational models of scientific discovery, programming methodologies, programming languages, truth maintenance mechanisms, and automated theorem finding systems. This paper reviews the origins, theoretical foundations, major developments, and intellectual significance of Epistemic Programming. It argues that the framework represents one of the earliest systematic attempts to establish a computational theory of scientific discovery and knowledge creation. Particular attention is paid to its conception of scientific discovery as an epistemic process, its use of strong relevant logic as a logical foundation, its reinterpretation of computation as epistemic transformation, and its relevance to contemporary research on AI scientists and autonomous scientific discovery. The paper further argues that Epistemic Programming may be viewed as a significant conceptual shift in the history of programming paradigms, extending the focus of programming from computation and deduction toward knowledge creation itself.

The problem of automated theorem finding is one of the 33 basic research problems in automated reasoning which was originally proposed by Wos in 1988, and it is still an open problem. To solve the problem, an approach of forward deduction based on the strong relevant logics was proposed. Following the approach, this paper presents a systematic methodology for automated theorem finding. To show the effectiveness of our methodology, the paper presents two case studies, one is automated theorem finding in NBG set theory and the other is automated theorem finding in Peano’s arithmetic. Some known theorems have been found in our case studies.

This paper proposes a new relevant logic B+⊓⊔, which is obtained by adding two binary connectives, intensional conjunction ⊓ and intensional disjunction ⊔, to Meyer–Routley minimal positive relevant logic B+, where ⊓ and ⊔ are weaker than fusion ◦ and fission +, respectively. We give Kripke-style semantics for B+⊓⊔, with →, ⊓ and ⊔ modelled by ternary relations. We prove the soundness and completeness of the proposed semantics. A number of axiomatic extensions of B+⊓⊔, including negation-extensions, are also considered, together with the corresponding semantic conditions required for soundness and completeness to be maintained.

Traditional information systems are passive, i.e., data or knowledge is created, retrieved, modified, updated, and deleted only in response to operations issued by users or application programs, and the systems only can execute queries or transactions explicitly submitted by users or application programs but have no ability to do something actively by themselves. Unlike a traditional information system serving just as a storehouse of data or knowledge and working passively according to queries or transactions explicitly issued by users and application programs, an autonomous evolutionary information system serves as an autonomous and evolutionary partner of its users that discovers new knowledge from its database or knowledge-base autonomously, cooperates with its users in solving problems actively by providing the users with advices, and has a certain mechanism to improve its own state of "knowing" and ability of "working". This paper seminally defines what is an autonomous evolutionary information system, explain why autonomous evolutionary information systems are needed, and presents some new issues, fundamental considerations, and research directions in design and development of autonomous evolutionary information systems.

In order to provide scientists with a computational methodology and some computational tools to program their epistemic processes in scientific discovery, we are establishing a novel programming paradigm, named ‘Epistemic Programming’, which regards conditionals as the subject of computing, takes primary epistemic operations as basic operations of computing, and regards epistemic processes as the subject of programming. This paper seminally presents our fundamental observations and assumptions on scientific discovery processes and their automation, research problems on modeling, automating, and programming epistemic processes, and an outline of our research project of Epistemic Programming.

Recent advances in large language models (LLMs) have led to widespread claims that these systems exhibit increasingly sophisticated “reasoning” abilities. Contemporary LLM research focuses primarily on the behavioral success of reasoning-related tasks while paying comparatively little attention to the logical correctness/validity of the reasoning processes involved. Evaluations of mathematical problem solving, planning, scientific question answering, and chain-of-thought prompting have reinforced the perception that reasoning emerges from scale. However, from the viewpoint of Philosophy of Logic, two important conceptual issues remain insufficiently examined: First, the nature of “reasoning”, what exactly is “reasoning”? Second, the criteria for “reasoning correctness”, what exactly is “correct reasoning” and how evaluate “reasoning correctness”? Drawing on insights from Logic and Philosophy of Logic, this paper argues the above two important conceptual issues. We provide a definition of “reasoning” and, based on this definition, clearly distinguish between “reasoning correctness” and “reasoning appearance.” We point out that generating seemingly reasonable reasoning trajectories and intermediate steps, and even correct results do not prove the logical correctness/validity of the underlying reasoning process. We further argue that correctness should be treated as an independent dimension in the evaluation of any machine reasoning. Future progress may require explicit inferential evaluation criteria in addition to benchmark accuracy. The paper proposes a new correctness-oriented evaluation methodology for machine reasoning.

A novel programming paradigm, named ‘Epistemic Programming’, is proposed in order to provide scientists with a computational methodology and computational tools to program their epistemic processes in scientific discovery. The programming paradigm regards conditionals as the subject of computing, takes primary epistemic operations as basic operations of computing, and regards epistemic processes as the subject of programming. Some fundamental observations and assumptions on scientific discovery processes and their automation are presented as the fundamentals of this new research direction. Some research problems in this direction are set up. It is discussed why classical mathematical logic and its various classical conservative extensions cannot satisfactorily underlie epistemic processes. As the logic foundation to underlie epistemic programming, a strong relevant logic model of epistemic processes in scientific discovery is presented, and epistemic programs are defined based on the model.

Modeling epistemic processes in scientific discovery satisfactorily is an indispensable step to automating scientific discovery processes. This paper presents some significant fundamental observations and assumptions on scientific discovery processes and their automation at first. Based on the observations and assumptions, the paper shows why classical mathematical logic, its various classical conservative extensions, and traditional (weak) relevant logics cannot satisfactorily underlie epistemic processes in scientific discovery, and then presents a strong relevant logic model of epistemic processes in scientific discovery. A novel program paradigm, named ‘Epistemic Programming’, based on the strong relevant logic model of epistemic processes is also presented.

Recently, with the application progress of AIGC tools based on large language models (LLMs), led by ChatGPT, many AI experts and more non-professionals are trumpeting the “reasoning ability” of the LLMs. The present author considers that the so-called “reasoning ability” of LLMs are just illusions of those people who with vague concepts. In fact, the LLMs can never have the true reasoning ability. This paper intents to explain that, because the essential limitations of their working principle, the LLMs can never have the ability of true correct reasoning.