The Semantic Arrow of Time, Part I: From Eddington to Ethernet

This paper argues that the perceived arrow of time in computing is a semantic design choice rooted in the Forward-In-Time-Only (FITO) assumption rather than a fundamental thermodynamic law, a category mistake inherited from Newtonian physics that obscures the time-symmetric nature of fundamental reality and constrains distributed systems theory.

The Gist

The Big Idea: Two Different Arrows

Imagine time as an arrow flying through the air. For the last 100 years, physicists have told us this arrow points in one direction because of Entropy (disorder). This is the Thermodynamic Arrow: eggs break but don't un-break; coffee cools down but doesn't spontaneously heat up. This arrow is about physics and energy.

This paper argues that computers have a second arrow, and it's causing all our digital headaches.

This is the Semantic Arrow. It's not about heat or energy; it's about meaning. It asks: "Did the message arrive with the intended meaning intact?"

The author, Paul Borrill, claims that while nature is actually flexible and can sometimes flow both ways (or have no clear order at all), our computers are built on a rigid, outdated rule that forces everything to flow in only one direction: Forward-In-Time-Only (FITO).

---

The Core Problem: The "One-Way Street" Mistake

To understand the paper, imagine you are trying to have a conversation with a friend across a noisy room.

How Nature Actually Works (The Physics View):
In the real world, time is a bit like a dance floor. Sometimes the music stops, sometimes the dancers move in circles, and sometimes two people move at the exact same time without one leading the other. Recent experiments show that in the quantum world, cause and effect can even happen in a "superposition" (both orders at once). Nature doesn't demand a strict "First A, then B" rule.

How Computers Work (The FITO View):
Our computer networks, however, are built like a one-way highway.

• The Assumption: We assume that if you send a letter, it must go from Sender $\to$ Receiver. We assume time moves only forward.
• The Mistake: The author calls this a "Category Mistake." It's like a visitor to Oxford University seeing all the colleges, libraries, and fields, and then asking, "But where is the University?" They are confusing the parts with the whole.
  • In computing, we confuse a logical rule (we decided to order messages this way for convenience) with a law of physics (messages must flow this way because the universe demands it).

Because we built our digital world on this "one-way street" assumption, we created a lot of impossible problems.

---

The Consequences: Why Your Files Get Corrupted

When you force a flexible, two-way world into a rigid, one-way computer system, things break. Here are the analogies for the specific problems mentioned in the paper:

1. The "Timeout and Retry" Trap (The Double-Booking)

• The Scenario: You order a pizza. The phone rings, but you don't answer. You hang up and call back.
• The FITO Problem: The first call did go through, and the pizza shop started cooking. But because you didn't hear the answer, you ordered again. Now you have two pizzas.
• The Computer Version: If a computer doesn't get a "Yes" back quickly, it assumes the message was lost and sends it again. But the first message might have actually arrived! This causes duplicate transactions (charging your card twice) or confused data. We have to build complex "deduplication" systems just to fix this self-made mess.

2. The "Last Writer Wins" Disaster (The Eraser)

• The Scenario: You and your partner are editing a shared Google Doc. You both type at the exact same time.
• The FITO Problem: The computer looks at the clocks. "Oh, your clock says 10:00:01 and hers says 10:00:02. Since hers is later, we keep hers and delete yours."
• The Reality: Maybe your clock is 5 minutes fast! Maybe she was typing a joke and you were typing a serious fact. By blindly trusting the "forward time" stamp, the computer silently deletes your important work. This is Silent Data Corruption. The file looks fine, but the meaning is gone.

3. The "Hallucination" of AI (The Broken Story)

• The Scenario: A storyteller who can only look forward.
• The FITO Problem: Large Language Models (like the one you are talking to right now) work by predicting the next word based on the previous words. They are strictly "Forward-In-Time." They can't look back and say, "Wait, if I say this word now, it contradicts the story I told three paragraphs ago."
• The Result: They generate text that sounds fluent but makes no sense globally. This is the digital equivalent of a "torn write"—the story parses correctly but means the wrong thing.

---

The Solution: A New Way to Build

The paper suggests we stop trying to force computers to obey a "one-way street" rule that doesn't exist in nature. Instead, we should build systems that act more like a conversation.

The "Reflective" Alternative:
Instead of just sending a message and hoping it arrives, imagine a system where:

1. You propose: "I want to change the file."
2. They reflect: "I see your change. Let me check if it fits with my current state."
3. Commit or Abort: Only if both agree does the change become permanent. If there's a conflict, you both hit "Undo" and try again.

This is called Indefinite Causal Order. It allows the system to say, "We don't know who went first yet, so let's hold the decision until we talk it out."

Summary: The Takeaway

• Physics is flexible: The universe allows for time to be symmetric, relative, and sometimes unordered.
• Computing is rigid: We built our digital world on the assumption that time must flow forward in a straight line (FITO).
• The Cost: This rigid assumption causes data corruption, duplicate charges, lost files, and AI hallucinations.
• The Fix: We need to redesign our networks to be bidirectional and reflective. We need to stop treating "time stamps" as the ultimate truth and start treating "mutual agreement" as the truth.

In short: We built our digital cities on a map that says "One Way Only," but the terrain is actually a two-way street. It's time to update the map so our data doesn't get lost in traffic.

Technical Summary

Based on the provided text, here is a detailed technical summary of "The Semantic Arrow of Time: Part I: From Eddington to Ethernet" by Paul Borrill.

1. Problem Statement

The paper identifies a fundamental disconnect between the physics of time and the architectural assumptions of modern distributed computing.

• The Core Issue: Distributed systems theory (including protocols, consensus algorithms, and file synchronization) relies on the Forward-In-Time-Only (FITO) assumption. This assumption posits that causation is irreversible, acyclic, and globally monotonic (events have a single, well-defined temporal order).
• The Consequence: This assumption is treated as an ontological law of nature, leading to "impossibility theorems" (e.g., FLP, CAP, Two Generals) that are presented as fundamental limits of computing.
• The Reality: Fundamental physics is time-symmetric at the microscopic level. The thermodynamic arrow of time (entropy increase) is an emergent property of boundary conditions, not a fundamental law. Furthermore, recent quantum experiments (Indefinite Causal Order) demonstrate that nature does not require a globally fixed causal ordering.
• The Resulting Failure: By treating an epistemic convention (logical message ordering) as a physical constraint, computing systems suffer from semantic corruption (e.g., torn writes, silent data loss, hallucinations in LLMs) rather than just thermodynamic inefficiency.

2. Methodology

The paper employs a cross-disciplinary analytical approach, tracing a philosophical and physical argument from the philosophy of physics to computer networking:

• Philosophical & Physical Review: The author synthesizes arguments from:
  • Eddington & Thermodynamics: Establishing that the thermodynamic arrow is emergent, not fundamental.
  • Huw Price (Time-Symmetric Ontology): Arguing for an "Archimedean viewpoint" where temporal asymmetry is not presupposed.
  • Lee Smolin (Temporal Naturalism): Asserting the reality of the present moment and the evolution of laws.
  • Carlo Rovelli (Relational Time): Viewing time as a consequence of interaction rather than a background parameter.
  • Indefinite Causal Order (ICO): Citing experimental evidence (Quantum Switch) that nature permits correlations without fixed causal ordering.
• Category Mistake Analysis: Applying Gilbert Ryle's concept of the "category mistake" to computing. The paper argues that computing mistakenly treats logical ordering (an epistemic tool for reasoning) as physical causality (an ontic constraint).
• Genealogical Tracing: Tracing the FITO assumption through three historical stages:
  1. Shannon (1948): Unidirectional channel modeling.
  2. Lamport (1978): The "happened-before" relation enforcing a global Directed Acyclic Graph (DAG).
  3. Impossibility Theorems (1980s–2000s): Deriving limits (FLP, CAP) based on the FITO premise.
• Leibnizian Application: Applying the Principle of the Identity of Indiscernibles to show that assigning causal order to empirically indistinguishable events constitutes "surplus structure."

3. Key Contributions

• Definition of the Semantic Arrow of Time:
  The paper defines a new concept distinct from the thermodynamic arrow. The Semantic Arrow is the direction in which the intended interpretation of shared state is preserved.
  • Properties: It is local (not global), emerges from interaction (not clocks), is reversible until commitment, and its violation causes semantic corruption (not heat death).
• Identification of the FITO Assumption:
  The paper explicitly names and critiques the Forward-In-Time-Only (FITO) assumption as the root cause of distributed system limitations. It argues this is a design choice, not a physical law.
• Reframing Impossibility Theorems:
  The paper posits that theorems like FLP (impossibility of consensus in asynchronous systems) and CAP are not laws of physics but theorems specific to FITO systems. If the FITO assumption is dropped, these impossibilities dissolve, opening a new design space.
• Diagnosis of Systemic Failures:
  The paper links specific computing failures to the FITO assumption:
  • Timeout-and-Retry (TAR): Causes duplicate effects because it assumes forward-only recovery.
  • Clock Synchronization (NTP/Spanner): Imposes a false global "now" on relativistic systems.
  • Last-Writer-Wins (LWW): Resolves conflicts based on timestamps rather than semantic intent, leading to silent data loss.
  • Autoregressive Prediction (LLMs): Generates hallucinations because the model cannot "look back" to verify consistency, mirroring distributed system "torn writes."

4. Results & Proposed Alternative

While Part I is primarily theoretical, it establishes the foundation for a constructive alternative presented in the subsequent papers of the series:

• The Constructive Alternative Framework:
  1. Interaction-Derived Causality: Causal order is discovered through interaction, not assumed a priori.
  2. Reflective Acknowledgment: Causal order requires a round-trip (message + reflection) rather than one-way transmission.
  3. Reversibility: State transitions remain tentative until mutual commitment, allowing for abortion and reversal.
  4. Indefinite Logical Timestamps: A four-valued causal structure that admits indefinite ordering between concurrent events, resolving only after symmetric exchange.
  5. Mutual Information Conservation: The new consistency primitive is the conservation of mutual information (The Leibniz Bridge), replacing temporal precedence.

5. Significance

• Paradigm Shift: The paper challenges the foundational axioms of distributed systems theory, suggesting that 40 years of research has been constrained by a category mistake inherited from Shannon's telephone channel model.
• Unification of Physics and Computing: It bridges the gap between modern quantum gravity/thermodynamics (which allow for indefinite causal order) and computer science, arguing that computing architectures should evolve to match the time-symmetric nature of the universe.
• Practical Implications: By recognizing that "impossibility" is a result of the FITO design choice, the paper suggests that current limitations in consistency, availability, and partition tolerance (CAP) are solvable through new protocols that utilize bilateral, reflective, and reversible interactions.
• Broader Impact: The framework extends beyond networking to explain failures in file synchronization (e.g., iCloud), email consistency, and even the hallucination mechanisms in Large Language Models, offering a unified theory of "semantic corruption."

In summary, the paper argues that computing must abandon the Newtonian/absolute view of time embedded in its protocols and adopt a relational, time-symmetric architecture to eliminate semantic corruption and overcome current theoretical limits.