Mediative Fuzzy Logic Overview

• Mediative Fuzzy Logic is a fuzzy logic framework that reconciles hesitant and conflicting assessments via a convex operator parameterized by hesitation (π) and contradiction (ζ).
• It integrates bilattice-style valuation and propositional semantics, separating independent truth–falsity coordinates to enable robust, paraconsistent reasoning.
• Higher-type and quantum extensions, including interval type-2, granule-indexed, and effect-algebraic models, support practical applications like autonomous braking sensor fusion.

Mediative Fuzzy Logic is a fuzzy-logical framework devised to reconcile hesitant or conflicting assessments in control and decision-making by explicitly parameterizing hesitation and contradiction within an aggregation operator and then lifting that operator into a propositional semantics. In its formulation from type-1 foundations through interval type-2, granular type-3, and quantum extensions, it combines a convex mediative operator, independent truth–falsity valuations, and a conservative extension of a standard left-continuous-$t$-norm fuzzy logic. The resulting hierarchy is presented as sound, paraconsistent, and conservative over the underlying fuzzy base for formulas without mediation, while also admitting coherent reductions from higher-order and quantum settings back to the type-1 case under suitable assumptions (Ross, 21 May 2026).

1. Type-1 mediative operator

At the type-1 level, the framework starts from two real inputs $a,b\in[0,1]$, interpreted in the source presentation as the “agreement” and “non-agreement” channels, together with two control parameters $\pi,\zeta\in[0,1]$, interpreted as hesitation and contradiction degrees. The mediative operator is defined by

$\Med(a,b;\pi,\zeta) \;=\;\bigl(1-\pi-\tfrac{\zeta}{2}\bigr)\,a\;+\;\bigl(\pi+\tfrac{\zeta}{2}\bigr)\,b.$

Its coefficients are constrained by the normalization condition

$$w_1=1-\pi-\tfrac\zeta2,\qquad w_2=\pi+\tfrac\zeta2,\qquad w_1,w_2\in[0,1],\qquad w_1+w_2=1,$$

so the operator is a convex aggregation (Ross, 21 May 2026).

Two reduction axioms characterize limiting cases. When $\zeta=0$, the operator reduces to the hesitation-weighted average

$\Med(a,b;\pi,0)=(1-\pi)\,a+\pi\,b.$

When $\pi=\zeta=0$, it reduces further to the type-1 base output

$\Med(a,b;0,0)=a.$

Under normalization, the output lies between its inputs: $\min(a,b)\le \Med(a,b;\pi,\zeta)\le \max(a,b).$ Accordingly, the operator does not extrapolate beyond the evidence channels it combines; the source text treats this as part of its conservative behavior (Ross, 21 May 2026).

The conceptual role of $a,b\in[0,1]$0 and $a,b\in[0,1]$1 is sharply separated. When $a,b\in[0,1]$2, mediation reflects hesitation alone. When $a,b\in[0,1]$3, contradiction contributes symmetrically through the $a,b\in[0,1]$4 shift in each weight. This suggests that contradiction is not modeled as a simple rejection of one channel by the other, but as a controlled redistribution of influence within a convex mixture.

2. Bilattice-style valuation and mediative truth degree

For propositional semantics, each atomic formula is assigned an independent truth coordinate $a,b\in[0,1]$5 and falsity coordinate $a,b\in[0,1]$6, with valuation domain

$a,b\in[0,1]$7

From these coordinates, the framework derives hesitation and contradiction by

$a,b\in[0,1]$8

These satisfy $a,b\in[0,1]$9 and $\pi,\zeta\in[0,1]$0, so hesitation and contradiction are mutually exclusive at a given valuation point (Ross, 21 May 2026).

The semantics is “bilattice-like” in the sense that conjunction, disjunction, and negation operate on truth and falsity coordinates separately, using a fixed left-continuous $\pi,\zeta\in[0,1]$1-norm $\pi,\zeta\in[0,1]$2 and its dual $\pi,\zeta\in[0,1]$3-conorm $\pi,\zeta\in[0,1]$4: $\pi,\zeta\in[0,1]$5

$\pi,\zeta\in[0,1]$6

$\pi,\zeta\in[0,1]$7

This arrangement preserves the independence of truth and falsity coordinates rather than forcing them to sum to $\pi,\zeta\in[0,1]$8 (Ross, 21 May 2026).

A scalar mediative score is then extracted from a pair $\pi,\zeta\in[0,1]$9 by setting

$\Med(a,b;\pi,\zeta) \;=\;\bigl(1-\pi-\tfrac{\zeta}{2}\bigr)\,a\;+\;\bigl(\pi+\tfrac{\zeta}{2}\bigr)\,b.$0

and defining

$\Med(a,b;\pi,\zeta) \;=\;\bigl(1-\pi-\tfrac{\zeta}{2}\bigr)\,a\;+\;\bigl(\pi+\tfrac{\zeta}{2}\bigr)\,b.$1

Thus, the truth channel is the direct truth degree $\Med(a,b;\pi,\zeta) \;=\;\bigl(1-\pi-\tfrac{\zeta}{2}\bigr)\,a\;+\;\bigl(\pi+\tfrac{\zeta}{2}\bigr)\,b.$2, whereas the second channel is the complement of falsity, $\Med(a,b;\pi,\zeta) \;=\;\bigl(1-\pi-\tfrac{\zeta}{2}\bigr)\,a\;+\;\bigl(\pi+\tfrac{\zeta}{2}\bigr)\,b.$3. A plausible implication is that the framework is designed to mediate not between two homogeneous truth estimates, but between positive support and absence of falsification.

3. Propositional system MFL-T1

The propositional calculus MFL-T1 extends any standard left-continuous-$\Med(a,b;\pi,\zeta) \;=\;\bigl(1-\pi-\tfrac{\zeta}{2}\bigr)\,a\;+\;\bigl(\pi+\tfrac{\zeta}{2}\bigr)\,b.$4-norm fuzzy logic, with BL and Łukasiewicz logic named as examples, by adding a unary mediative connective $\Med(a,b;\pi,\zeta) \;=\;\bigl(1-\pi-\tfrac{\zeta}{2}\bigr)\,a\;+\;\bigl(\pi+\tfrac{\zeta}{2}\bigr)\,b.$5. Its language is

$\Med(a,b;\pi,\zeta) \;=\;\bigl(1-\pi-\tfrac{\zeta}{2}\bigr)\,a\;+\;\bigl(\pi+\tfrac{\zeta}{2}\bigr)\,b.$6

A valuation $\Med(a,b;\pi,\zeta) \;=\;\bigl(1-\pi-\tfrac{\zeta}{2}\bigr)\,a\;+\;\bigl(\pi+\tfrac{\zeta}{2}\bigr)\,b.$7 assigns each formula a pair

$\Med(a,b;\pi,\zeta) \;=\;\bigl(1-\pi-\tfrac{\zeta}{2}\bigr)\,a\;+\;\bigl(\pi+\tfrac{\zeta}{2}\bigr)\,b.$8

with the connectives interpreted via the bilattice-style operations above. Implication is defined coordinate-wise using the residuum: $\Med(a,b;\pi,\zeta) \;=\;\bigl(1-\pi-\tfrac{\zeta}{2}\bigr)\,a\;+\;\bigl(\pi+\tfrac{\zeta}{2}\bigr)\,b.$9 For the mediative connective,

$$w_1=1-\pi-\tfrac\zeta2,\qquad w_2=\pi+\tfrac\zeta2,\qquad w_1,w_2\in[0,1],\qquad w_1+w_2=1,$$0

This means mediation converts an independent truth–falsity pair into a complementary pair determined by the scalar score $$w_1=1-\pi-\tfrac\zeta2,\qquad w_2=\pi+\tfrac\zeta2,\qquad w_1,w_2\in[0,1],\qquad w_1+w_2=1,$$1 (Ross, 21 May 2026).

The axiomatic basis consists of all axioms and rules of the chosen fuzzy base logic together with three mediative schemata: $$w_1=1-\pi-\tfrac\zeta2,\qquad w_2=\pi+\tfrac\zeta2,\qquad w_1,w_2\in[0,1],\qquad w_1+w_2=1,$$2

$$w_1=1-\pi-\tfrac\zeta2,\qquad w_2=\pi+\tfrac\zeta2,\qquad w_1,w_2\in[0,1],\qquad w_1+w_2=1,$$3

$$w_1=1-\pi-\tfrac\zeta2,\qquad w_2=\pi+\tfrac\zeta2,\qquad w_1,w_2\in[0,1],\qquad w_1+w_2=1,$$4

with modus ponens as rule. These schemata encode monotonicity with respect to implication, preservation of the extremal truth values, and extensionality of mediation (Ross, 21 May 2026).

Three metatheoretic properties are explicitly established. First, soundness: if $$w_1=1-\pi-\tfrac\zeta2,\qquad w_2=\pi+\tfrac\zeta2,\qquad w_1,w_2\in[0,1],\qquad w_1+w_2=1,$$5, then $$w_1=1-\pi-\tfrac\zeta2,\qquad w_2=\pi+\tfrac\zeta2,\qquad w_1,w_2\in[0,1],\qquad w_1+w_2=1,$$6, where semantic consequence is defined through the scalar score $$w_1=1-\pi-\tfrac\zeta2,\qquad w_2=\pi+\tfrac\zeta2,\qquad w_1,w_2\in[0,1],\qquad w_1+w_2=1,$$7. Second, paraconsistency: explosion $$w_1=1-\pi-\tfrac\zeta2,\qquad w_2=\pi+\tfrac\zeta2,\qquad w_1,w_2\in[0,1],\qquad w_1+w_2=1,$$8 is not derivable, and there are valuations for which both $$w_1=1-\pi-\tfrac\zeta2,\qquad w_2=\pi+\tfrac\zeta2,\qquad w_1,w_2\in[0,1],\qquad w_1+w_2=1,$$9 and $\zeta=0$0 are high. Third, conservativity: if a formula contains no occurrence of $\zeta=0$1, derivability in MFL-T1 coincides exactly with derivability in the underlying fuzzy base logic (Ross, 21 May 2026).

A frequent misconception in discussions of nonclassical logics is to identify paraconsistency with unrestricted inconsistency tolerance. The formal claim here is narrower: explosion is not derivable, and high mediative support can coexist for a formula and its negation. The system is therefore paraconsistent in the specific proof-theoretic and semantic sense stated in the source, not a wholesale abandonment of inferential discipline.

4. Higher-type and quantum generalizations

The framework is extended in three directions: interval type-2 semantics (MFL-T2), granular type-3 semantics (MFL-T3), and quantum mediative semantics (QMFL). Each extension preserves the basic mediative idea while enriching the semantic domain (Ross, 21 May 2026).

Extension	Semantic objects	Stated reduction
MFL-T2	Interval type-2 fuzzy sets and footprints of uncertainty	Degenerate singletons recover MFL-T1
MFL-T3	Granule-indexed local valuations with global aggregator	Homogeneous idempotent aggregation recovers lower level
QMFL	Effects and density operators on a finite-dimensional Hilbert space	Commuting state and effects recover classical MFL-T1

In MFL-T2, each atomic proposition is assigned two interval type-2 fuzzy sets $\zeta=0$2 on $\zeta=0$3, equivalently represented by their footprints of uncertainty $\zeta=0$4 and $\zeta=0$5. Scalar bounds are obtained from each interval set by outer/inner or $\zeta=0$6-cut projection: $\zeta=0$7 and analogously for $\zeta=0$8. Connectives are handled by endpoint monotonicity, for example

$\zeta=0$9

Hesitation and contradiction become interval-valued: $\Med(a,b;\pi,0)=(1-\pi)\,a+\pi\,b.$0

$\Med(a,b;\pi,0)=(1-\pi)\,a+\pi\,b.$1

Two evaluation modes are then given. In the type-reduced mode, $\Med(a,b;\pi,0)=(1-\pi)\,a+\pi\,b.$2 and $\Med(a,b;\pi,0)=(1-\pi)\,a+\pi\,b.$3 are first reduced to a crisp pair $\Med(a,b;\pi,0)=(1-\pi)\,a+\pi\,b.$4, for example by the centroid, and $\Med(a,b;\pi,0)=(1-\pi)\,a+\pi\,b.$5. In the envelope mode, one computes

$\Med(a,b;\pi,0)=(1-\pi)\,a+\pi\,b.$6

The axiomatic system remains that of MFL-T1; only the semantic domain is enriched (Ross, 21 May 2026).

MFL-T3 introduces a finite set of granules $\Med(a,b;\pi,0)=(1-\pi)\,a+\pi\,b.$7, such as experts, sensors, or time windows. For each $\Med(a,b;\pi,0)=(1-\pi)\,a+\pi\,b.$8, a local mediative valuation $\Med(a,b;\pi,0)=(1-\pi)\,a+\pi\,b.$9 is defined, taking values either in type-1 pairs or in type-2 sets. A local scalar score $\pi=\zeta=0$0 is computed either by $\pi=\zeta=0$1 in the type-1 case or by type-reduction in the type-2 case. A global aggregator

$\pi=\zeta=0$2

then combines the family $\pi=\zeta=0$3 into

$\pi=\zeta=0$4

The source lists weighted averages, OWAs, and hierarchical policies as examples of such aggregation, tuned to domain-specific safety requirements (Ross, 21 May 2026).

QMFL transfers the construction to a finite-dimensional Hilbert space $\pi=\zeta=0$5. A quantum effect is any operator $\pi=\zeta=0$6 with $\pi=\zeta=0$7 in the Löwner order, and a state is a density operator $\pi=\zeta=0$8 with $\pi=\zeta=0$9 and $\Med(a,b;0,0)=a.$0. Each proposition $\Med(a,b;0,0)=a.$1 is assigned two effects $\Med(a,b;0,0)=a.$2 and $\Med(a,b;0,0)=a.$3, representing positive and negative evidence channels. One defines

$\Med(a,b;0,0)=a.$4

then computes $\Med(a,b;0,0)=a.$5, $\Med(a,b;0,0)=a.$6, and weights

$\Med(a,b;0,0)=a.$7

The quantum mediative effect is

$\Med(a,b;0,0)=a.$8

which is again an effect because it is a convex combination of effects. Its Born expectation

$\Med(a,b;0,0)=a.$9

satisfies

$\min(a,b)\le \Med(a,b;\pi,\zeta)\le \max(a,b).$0

The extension is therefore effect-algebraic while remaining pointwise consistent with the classical mediative operator at the level of expectation values (Ross, 21 May 2026).

5. Autonomous-braking sensor fusion

A concrete case study is given for autonomous braking, where the proposition $\min(a,b)\le \Med(a,b;\pi,\zeta)\le \max(a,b).$1 is “There is a dangerous obstacle within 20 m.” Two perception channels are used, radar/LiDAR and camera, each producing a mediative pair $\min(a,b)\le \Med(a,b;\pi,\zeta)\le \max(a,b).$2. Fusion is performed linearly: $\min(a,b)\le \Med(a,b;\pi,\zeta)\le \max(a,b).$3 Decision thresholds are specified directly on the scalar mediative score: $\min(a,b)\le \Med(a,b;\pi,\zeta)\le \max(a,b).$4

$\min(a,b)\le \Med(a,b;\pi,\zeta)\le \max(a,b).$5

$\min(a,b)\le \Med(a,b;\pi,\zeta)\le \max(a,b).$6

These thresholds operationalize the framework’s stated safety-first behavior (Ross, 21 May 2026).

Case	Fused values and mediative score	Decision
Case 1 (fog)	$\min(a,b)\le \Med(a,b;\pi,\zeta)\le \max(a,b).$7, $\min(a,b)\le \Med(a,b;\pi,\zeta)\le \max(a,b).$8, $\min(a,b)\le \Med(a,b;\pi,\zeta)\le \max(a,b).$9, $a,b\in[0,1]$00	Emergency brake
Case 2 (glare)	$a,b\in[0,1]$01, $a,b\in[0,1]$02, $a,b\in[0,1]$03	Cautious slow-down
Case 3 (contradiction)	$a,b\in[0,1]$04, $a,b\in[0,1]$05, $a,b\in[0,1]$06, $a,b\in[0,1]$07	Emergency brake

In Case 1, the radar/LiDAR channel yields $a,b\in[0,1]$08, the camera channel yields $a,b\in[0,1]$09, and $a,b\in[0,1]$10. The fused pair is $a,b\in[0,1]$11, with $a,b\in[0,1]$12 and $a,b\in[0,1]$13, producing

$a,b\in[0,1]$14

In Case 2, the inputs are $a,b\in[0,1]$15 and $a,b\in[0,1]$16 with $a,b\in[0,1]$17, so the fusion yields $a,b\in[0,1]$18, $a,b\in[0,1]$19, and therefore $a,b\in[0,1]$20. In Case 3, the inputs are $a,b\in[0,1]$21 and $a,b\in[0,1]$22 with $a,b\in[0,1]$23, producing $a,b\in[0,1]$24, $a,b\in[0,1]$25, $a,b\in[0,1]$26, $a,b\in[0,1]$27, $a,b\in[0,1]$28, $a,b\in[0,1]$29, $a,b\in[0,1]$30, and $a,b\in[0,1]$31 (Ross, 21 May 2026).

The same scalar scores and thus the same decisions are reported across MFL-T1, MFL-T2, MFL-T3, and QMFL in these minimal configurations because the higher-type and quantum structures are instantiated in a “commutative,” low-uncertainty regime. The source notes that richer behavior appears when second-order uncertainty bands, temporal or sensor granules, or non-commuting quantum effects are introduced, but the core outputs remain conservative and transparent (Ross, 21 May 2026).

6. Reduction theorems and coherence of the hierarchy

A central claim of the framework is coherence across semantic levels. For MFL-T2, if every interval type-2 fuzzy set $a,b\in[0,1]$32 collapses to a singleton $a,b\in[0,1]$33, then both the type-reduced and envelope evaluations coincide with the type-1 MFL-T1 semantics. This is stated both as Proposition 3.1 in the detailed presentation and again in the summary of reductions (Ross, 21 May 2026).

For MFL-T3, homogeneous agreement across granules yields collapse to the lower level. If the aggregator $a,b\in[0,1]$34 is idempotent and all granules assign the same local score $a,b\in[0,1]$35, then

$a,b\in[0,1]$36

In particular, if all $a,b\in[0,1]$37 coincide on a single atom $a,b\in[0,1]$38, MFL-T3 reduces to MFL-T2, or further to MFL-T1. The formal statement appears as Theorem 3.2 and is restated in the coherence summary (Ross, 21 May 2026).

For QMFL, if the state $a,b\in[0,1]$39 and effects $a,b\in[0,1]$40 all commute, meaning they are simultaneously diagonal in a common basis, then the quantum mediative degree $a,b\in[0,1]$41 reduces exactly to the classical mediative operator $a,b\in[0,1]$42 on the diagonal entries. This provides the classical limit of the quantum construction (Ross, 21 May 2026).

Taken together, these results support the claim that the hierarchy

$a,b\in[0,1]$43

is conservative and coherent. The paper explicitly characterizes the hierarchy as capable of introducing explicit controls for hesitation, contradiction, multiple evidence granules, second-order uncertainty, and quantum information, while collapsing back to the type-1 setting under suitable assumptions. A plausible implication is that the higher-level systems are intended not as replacements for type-1 fuzzy logic, but as structured refinements that preserve compatibility with standard $a,b\in[0,1]$44-norm-based reasoning (Ross, 21 May 2026).

7. Position within fuzzy and nonclassical reasoning

Within the framework’s own presentation, Mediative Fuzzy Logic is not merely an operational aggregation heuristic. It is formalized as a propositional extension of a standard left-continuous-$a,b\in[0,1]$45-norm fuzzy logic with a dedicated mediative connective, a bilattice-like semantics over independent truth and falsity coordinates, and higher-type as well as quantum semantic lifts (Ross, 21 May 2026).

Its relation to standard fuzzy logic is governed by conservativity: formulas without $a,b\in[0,1]$46 behave exactly as in the underlying base logic. Its relation to inconsistency-tolerant reasoning is governed by paraconsistency: contradictory evidence need not trivialize inference. Its relation to multi-source and uncertain information processing is expressed through the explicit handling of hesitation $a,b\in[0,1]$47, contradiction $a,b\in[0,1]$48, interval type-2 uncertainty, granule-indexed local valuations, and quantum effects. These components are presented as suitable for incomplete, heterogeneous, and mildly contradictory evidence, particularly in intelligent decision systems (Ross, 21 May 2026).

A common misunderstanding would be to treat the framework as a departure from fuzzy logic into an unrelated evidential formalism. The formal results instead place it as an extension layered over a standard $a,b\in[0,1]$49-norm fuzzy base. Another misunderstanding would be to treat the quantum version as semantically disconnected from the classical one; the commuting-case reduction shows the opposite. In both respects, the design emphasis is on extension with reduction, rather than substitution without continuity.