Synthetic Philosophy and Deductive Engineering
Synthetic Philsophy and SPaDE are both projects which I have initiated and progressed in recent years. The former was intended to be purely philosophical, aimed at writing a small book presenting an approach to epistmology inspired by the relevance of epistemology to knowledge engineering, and artificial intelligence more broadly. Philosophy as groundwork for the engineering of cognitive systems.
This represented a retreat from the continuing software development aspirations of this retired software engineer, who thought little of his chances of turning his aspirations into code before time out. For me a constructive interest in the theory of knowledge had long been inextricably coupled to the earliest stages in the design of intelligent software, and it was that coupling which motivated the particular kind of epistemology which I sought to articulate. One in which knowledge is a means to an end, and its nature and metatheory (epistemology) should be designed with that end in mind. A design taking advantage of the advances in the foundations of mathematics and logic which our culture had achieved over the millenia between Aristotle’s first treatises on logic (and demonstrative science)and the present day, and which had been so influential in the development of science and technology.
Let me say a little more about my conception of synthetic epistemology before coming to SPaDE. My own closest philosophical sympathies eventually aligned with Rudolf Carnap in a turn of positivism influenced most conspicously by Frege and Russell in which then recent advances by philosophers in the foundations of mathematics could be seen as transformative for science. This was the most prominent thread in the `Logical Positivism’ which emerged from Schlick’s circle in Vienna. It made philosophy and the formal foundations of mathematics a handmaid to science, discarding those aspects of philosophy which resisted comprehension in such terms, notably metaphysics.
Carnap’s persistent rejection of metaphysics should be seen as tempered by a particular conception of what metaphysics is, and of some of the things which are often thought of as metaphysics which he embraced. His main critiques of metaphysics were, firstly that it is meaningless, and secondly that it is synthetic a priori. He nevertheless engaged with ontology, admitting the lavish abstract ontology of set theory in (some of) his foundation systems. This was a coherent position, because he considered ontology to be in effect part of the definition of a language, and hence a convention in the context of which ontological claims in the language are rendered meaningful.
In this kind of conventionalism, extended perhaps beyond Carnap’s own position, may be found a key to understanding the nature of synthetic epistemology. My own usage of the expression is not connected with the prior use as melding together more than one approach to epistemology, nor does it have any connection with the use of the term “synthetic” as a kind of proof (forward) in classical Greece, or with Kant’s use of the term to propositions which express a substantive claim about the world. The intent is to distinguish epistemology engaged in as a stage in the design of intelligent systems, from one which studies the nature of knowledge as it has been and seeks to understand its essential nature.
It flows from a conventionalism about language which not only admits that languages are systems of conventions (whether or not explicit and intentional) and admits not only that making deliberate and explicit choices is admissible, but also that this also extends to the vocabulary of epistemology and admits choices about what knowledge is and how it is to be represented and reasoned about. Such choices are implicit in the design of intelligent systems, even though the dominant approach at present involves little or no explicit attention to the epistemological choices being made.
Engaging in this kind of synthetic epistemology depends upon some sense of purpose beyond curiousity about the nature of knowledge. Consequently, synthetic epistemology as I conceived it was always a stage in the design of intelligent systems, if one of sufficient philosophical interest and breadth of relevance that it might usefully be separated from the architecture, design and implementation which follow en route to achieving machine intelligence. But with this mental attitude toward epistemology, progressing it independently of the intended sequel was beyond me.
There is a model for software development with a sequence of stages, requirements analysis, design at various levels, implementation and testing, which was related naturally to the logical dependencies between the different aspects of the process, but which is generally understood if adhered to in rigid temporal sequence to be a recipe for failure. The reason is that the adequacy of the results of one stage cannot be reliably assessed until the later stages for which it sets the rules are undertaken. A more iterative process often including prototype implementations, allows tentative design decisions to be tested and refined before they are locked in by the later stages of the process.
It was the rapid advances in generative AI following the release of ChatGPT in late 2022 which made it plausible that I could, with what would soon be called agentic AI, undertake the kind of software development which might flow naturally from the kind of epistemological synthesis I was working on. My Synthetic Epistemology project was mothballed, and SPaDE was born.
SPaDE explicitly straddles philosophy and engineering.