Start Here
DOGMA in One Hour
DOGMA is not DNA stored in a computer, not a wet-lab molecular computer, not a renamed transformer, and not AGI. It is an experimental architecture and evaluation program. “DNA-inspired” describes a source of hypotheses; it does not prove that the hypotheses work.
DOGMA Pipeline
Six local stages replace one monolithic attention operation
Advance through the canonical non-transformer computation from sequence to realized output.
Genome
Encode
HelixHash creates structural sequence representations.
G = H(sequence)
The Four Layers
DOGMA becomes easier when four different subjects are kept separate.
- 1.
- Molecular biology studies how cells store and express information. DNA is transcribed into RNA; RNA can be translated into protein. Regulation determines what is expressed, where, and when.
- 2.
- Molecular computing constructs physical computations from molecules. Adleman encoded a graph problem in DNA [Adleman, 1994]; strand-displacement systems later implemented logic and neural-like circuits [Seelig et al., 2006, Qian and Winfree, 2011].
- 3.
- Genomic machine learning learns statistical structure from biological sequences. Transformers and state-space models are useful baselines here, but they are software models of sequence data, not molecular computers.
- 4.
- DOGMA is a software research architecture inspired by genomic organization. Its current canonical line deliberately avoids transformer self-attention. It asks whether local interaction, recurrence, selective expression, persistent bounded state, and evolutionary search can produce useful learning behavior.
One Worked Example
Suppose the byte sequence ACGTACGT enters a small DOGMA model. Overlapping four-symbol windows are \[ (A,C,G,T),\ (C,G,T,A),\ (G,T,A,C),\ (T,A,C,G),\ (A,C,G,T). \] Each window produces a local representation. A causal convolution mixes nearby windows; a recurrent state carries information forward; a gate chooses how strongly each state is expressed. The model predicts the next byte. This is a neural computation on digital symbols. No molecule is synthesized.
If the correct next byte is \(C\) and the model assigns it probability \(0.5\), the loss is \[ -\log (0.5) \approx 0.693. \] Across \(N\) held-out bytes, average cross-entropy \(\bar {\mathcal {L}}\) gives perplexity \(\operatorname {PPL}=\exp (\bar {\mathcal {L}})\). Lower perplexity means better teacher-forced prediction on that split. It does not guarantee readable free generation, factual answers, reasoning, or general intelligence.
The Recursive Research Loop
A DOGMA candidate is never allowed to rewrite production weights directly. The controlled loop is: \[ \text {parent} \rightarrow \text {one bounded mutation} \rightarrow \text {train} \rightarrow \text {frozen probes} \rightarrow \text {archive or promote}. \] The candidate must retain real-data anchors, avoid evaluation overlap, run with multiple fixed seeds, and pass separate sequence, explanation, and self-critique probes. A quality-diversity archive retains useful stepping stones; only a reproducible global improvement replaces the live checkpoint.
A Twelve-Week Route
| Week | Reading | Question to answer |
| 1 | Chapters 1 and 2 | How do biological and molecular computation differ? |
| 2 | Chapter 3 | What does regulation compute? |
| 3 | Chapter 4; Appendix A | What do loss, gradient, and perplexity measure? |
| 4 | Chapter 5 | What problem does self-attention solve? |
| 5 | Chapter 6 | How are genomic foundation models evaluated? |
| 6 | Chapter 7 | Which DOGMA mechanisms are canonical, historical, or hypothetical? |
| 7 | Chapters 8 and 10 | How can local structure become a computational prior? |
| 8 | Chapters 9 and 11 | How are regimes, types, and programs represented? |
| 9 | Chapter 12 | Why separate exploration, archives, and promotion? |
| 10 | Chapters 13 and 14 | Why can lower perplexity coexist with worse generation? |
| 11 | Chapters 16 and 17 | What evidence would count as progress toward AGI? |
| 12 | Chapter 18 | Reproduce one baseline and write a failure analysis. |
Evidence Checklist
Before accepting any DOGMA claim, ask:
- 1.
- Is this a biological fact, an external result, an implementation statement, a measured DOGMA result, or a hypothesis?
- 2.
- Are the data source, split, checkpoint, evaluation version, and random seeds recorded?
- 3.
- Is the comparison matched for parameters, training bytes, and compute?
- 4.
- Does the evaluation test free generation and downstream behavior, not only teacher-forced likelihood?
- 5.
- Could contamination, cherry-picking, or synthetic-data feedback explain the result?
Starter Exercises
- S.1.
- Explain in two sentences why the central dogma is a biological theory, not proof that DOGMA will work.
- S.2.
- Compute perplexity for average losses \(0.5\), \(1.0\), and \(2.0\).
- S.3.
- Design one probe for canonical DNA generation and one probe for a readable biological explanation. State exactly what counts as a pass.
- S.4.
- Give an example of a candidate that belongs in the archive but should not replace the live model.
Contents
Preface
Reader’s Roadmap
DOGMA in One Hour
I The Biological Computer
1 The Code of Life — DNA as an Information System
1.1 The Molecular Basis of Biological Information
1.1.1 Step 1: The Atomic Ingredients
1.1.2 Step 2: The Nucleoside — Sugar Plus Base
1.1.3 Step 3: The Nucleotide — Adding the Phosphate
1.1.4 Step 4: The Single Strand — A Polynucleotide Chain
1.1.5 Step 5: The Double Helix — Watson–Crick Base Pairing
1.1.6 DNA Structure: Summary of the Build-up
1.2 DNA as Information Storage
1.2.1 Information Density: Nature’s Storage Medium
1.2.2 Binary Encoding of DNA
1.2.3 The Four-Letter Alphabet: Why Four?
1.2.4 DNA Data Storage
1.3 The Central Dogma of Molecular Biology
1.3.1 Transcription: DNA to RNA, Step by Step
1.3.2 The Genetic Code: A 64-to-21 Mapping
1.3.3 Translation: The Ribosomal Decoder
1.3.4 Post-Translational Modification: Runtime Polymorphism
1.4 DNA as Information: A Quantitative Analysis
1.4.1 Shannon Entropy of Genomic Sequences
1.4.2 Redundancy in the Genetic Code
1.4.3 Comparing DNA, Natural Language, and Code
1.5 DNA Replication: Copying the Code
1.5.1 Overview: Semi-Conservative Replication
1.5.2 Step-by-Step Mechanism
1.5.3 Error Correction: Proofreading and Repair
1.6 DNA as a Storage and Computing Medium
1.6.1 DNA Strand Displacement as a Computational Primitive
1.6.2 Chemical Reaction Networks and Turing Completeness
1.7 From Molecules to Architecture: The Abstraction Bridge
1.7.1 The Genome as Source Code
1.7.2 Why Previous Biological Metaphors in CS Were Shallow
1.8 Worked Examples
1.9 Exercises
1.10 Key Takeaways
2 A History of DNA Computing — From Adleman to Molecular Programming
2.1 Adleman’s Breakthrough (1994)
2.1.1 The Hamiltonian Path Problem
2.1.2 Encoding Graphs in DNA
2.1.3 Why It Worked—And Why It Didn’t Scale
2.2 DNA Logic Gates and Boolean Computation
2.2.1 Binary Encoding in DNA
2.2.2 The AND Gate
2.2.3 The OR Gate
2.2.4 The NOT Gate (Inverter)
2.2.5 NAND and Universal Computation
2.2.6 XOR Gate and Parity Detection
2.3 DNA Arithmetic: Computing with Molecules
2.3.1 The Half Adder
2.3.2 The Full Adder
2.3.3 DNA Subtraction via Two’s Complement
2.3.4 Comparison and Decision Making
2.4 Strand Displacement Cascades
2.4.1 The Three-Strand Mechanism
2.4.2 Kinetic Control via Toehold Length
2.4.3 Seesaw Gates: Scalable Digital Logic
2.5 Restriction Enzyme Computing and Splicing Systems
2.5.1 Restriction Enzymes as Programmable Cutters
2.5.2 Splicing: DNA Recombination as Computation
2.5.3 How Splicing Achieves Turing Completeness
2.5.4 Rothemund’s DNA Turing Machine (1996)
2.5.5 Surface-Based DNA Computing
2.6 Whiplash PCR: Hairpin-Based State Machines
2.6.1 The Hairpin State Machine
2.6.2 Computation Cycle
2.7 DNA Nanotechnology and Structural Computing
2.7.1 DNA Origami
2.7.2 Tile Self-Assembly and Turing Completeness
2.8 Chemical Reaction Networks: Turing-Complete Chemistry
2.8.1 Formal Definition
2.8.2 Computing with CRNs: Step-by-Step Examples
2.8.3 Turing Completeness of CRNs
2.8.4 Deterministic vs. Stochastic Semantics
2.9 DNA Neural Networks
2.9.1 Architecture
2.9.2 How the Winner-Take-All Works
2.9.3 Significance for DOGMA
2.10 Molecular Programming Languages
2.10.1 Visual DSD
2.10.2 NUPACK
2.10.3 The Compiler Analogy
2.11 From Molecules to Silicon: The DOGMA Bridge
2.12 The State of the Art and Open Challenges
2.13 Exercises
2.14 Key Takeaways
3 Gene Regulation as Computation
3.1 The Regulatory Grammar of DNA
3.1.1 Promoters: The Entry Point
3.1.2 Enhancers: Remote Amplifiers
3.1.3 Silencers and Insulators
3.2 Transcription Factor Binding: Step by Step
3.2.1 How Transcription Factors Bind DNA
3.2.2 Activators vs. Repressors
3.2.3 Cooperative Binding and the Hill Function
3.3 Transcription Factors as Logic Gates
3.3.1 Combinatorial Regulation
3.3.2 The Lac Operon: A Biological AND Gate
3.4 Gene Regulatory Networks
3.4.1 Network Motifs
3.4.2 The Repressilator: A Biological Ring Oscillator
3.4.3 The Toggle Switch: A Biological Flip-Flop
3.4.4 GRNs as Programs
3.5 Boolean Networks in Biology
3.5.1 Kauffman’s Random Boolean Networks
3.5.2 Attractors as Cell Types
3.6 Signal Transduction as Computation
3.6.1 Kinase Cascades as Amplifiers
3.6.2 The MAPK Pathway: A Signal Processing Pipeline
3.6.3 Feedback Loops in Signaling
3.7 Epigenetics: Computation That Modifies the Reader
3.7.1 DNA Methylation
3.7.2 Histone Modification
3.7.3 The Histone Code
3.7.4 Epigenetic Memory: State Without Sequence Change
3.8 Alternative Splicing: One Genome, Many Programs
3.8.1 The Mechanism
3.8.2 Computational Significance
3.9 The ENCODE Project: Quantifying the Regulatory Genome
3.10 Synthetic Biology: Engineering Biological Circuits
3.11 From Biological Regulation to DOGMA
3.12 Exercises
3.13 Key Takeaways
II The Intelligence Stack
4 Machine Learning Foundations for Biological AI
4.1 The Learning Problem
4.1.1 Supervised Learning as Function Approximation
4.1.2 Loss Functions
4.1.3 The Bias-Variance Tradeoff
4.1.4 Unsupervised and Self-Supervised Learning
4.2 Neural Networks from First Principles
4.2.1 The Perceptron: A Single Neuron
4.2.2 Activation Functions
4.2.3 Multi-Layer Perceptrons (MLPs)
4.3 Training Neural Networks
4.3.1 Backpropagation: Computing Gradients Efficiently
4.3.2 Gradient Descent Variants
4.3.3 The Loss Landscape
4.3.4 Regularization: Preventing Overfitting
4.4 Convolutional Neural Networks
4.4.1 The Convolution Operation
4.4.2 Key CNN Concepts
4.5 Representation Learning
4.5.1 Embeddings: Mapping Symbols to Geometry
4.5.2 The Geometry of Meaning
4.5.3 Positional Encoding
4.6 Sequence Models: From RNNs to State Spaces
4.6.1 Recurrent Neural Networks
4.6.2 Long Short-Term Memory (LSTM)
4.6.3 State Space Models
4.7 Evolutionary Algorithms
4.7.1 The Genetic Algorithm: Step by Step
4.7.2 Quality-Diversity Algorithms
4.8 Multi-Objective Optimization
4.8.1 Pareto Optimality
4.8.2 Scalarization Methods
4.9 Bringing It Together: The DOGMA Learning Stack
4.10 Exercises
4.11 Key Takeaways
5 Transformers and Large Language Models
5.1 The Attention Mechanism
5.1.1 From Alignment to Attention
5.1.2 Intuition: Queries, Keys, and Values
5.1.3 Scaled Dot-Product Attention
5.1.4 Why the Scaling Factor Matters
5.1.5 Worked Example: Self-Attention with 3 Tokens
5.1.6 Visualizing Self-Attention
5.2 Multi-Head Attention
5.2.1 Why Multiple Heads?
5.2.2 Head Splitting and Concatenation
5.2.3 What Different Heads Learn
5.2.4 Multi-Head Attention Diagram
5.3 The Transformer Architecture
5.3.1 Encoder and Decoder Stacks
5.3.2 Residual Connections and Layer Normalization
5.3.3 The Complete Transformer Block
5.4 Positional Encoding
5.4.1 The Permutation Equivariance Problem
5.4.2 Sinusoidal Positional Encodings
5.4.3 Learned and Relative Position Encodings
5.5 GPT and Autoregressive Language Models
5.5.1 Decoder-Only Architecture
5.5.2 The Next-Token Prediction Objective
5.5.3 Causal Masking
5.6 Training Large Language Models
5.6.1 Training Data Scale
5.6.2 Compute Requirements
5.7 Scaling Laws for Language Models
5.7.1 Power-Law Relationships
5.7.2 Compute-Optimal Training (Chinchilla)
5.8 Emergent Abilities of Large Language Models
5.8.1 In-Context Learning
5.8.2 Few-Shot, One-Shot, and Zero-Shot Learning
5.8.3 Chain-of-Thought Reasoning
5.8.4 Instruction Following
5.9 Instruction Tuning and RLHF
5.9.1 Supervised Instruction Tuning
5.9.2 Reinforcement Learning from Human Feedback
5.10 Key Models: A Comparative View
5.10.1 GPT Family (OpenAI)
5.10.2 BERT and Encoder-Only Models
5.10.3 LLaMA and Open-Source Models
5.10.4 Comparison Table
5.11 Limitations of the Transformer Paradigm
5.11.1 Quadratic Attention Cost
5.11.2 No Persistent State
5.11.3 No Native Typed Regions
5.11.4 Output-First Bias
5.11.5 No Evolutionary Self-Modification
5.11.6 Summary: What Transformers Cannot Do That Biology Can
5.12 Exercises
5.13 Key Takeaways
6 Biological Foundation Models
6.1 What Is a Foundation Model?
6.1.1 Definition and Core Idea
6.1.2 An Analogy: The Medical Student
6.1.3 Why Biology Is Special
6.1.4 The Biological Modalities
6.2 AlphaFold: Solving Protein Structure Prediction
6.2.1 The Protein Folding Problem
6.2.2 AlphaFold2: The Pipeline Step by Step
6.2.3 The AlphaFold2 Loss Function
6.2.4 AlphaFold2 Architecture Diagram
6.2.5 Why AlphaFold2 Uses “Only” 93M Parameters
6.3 ESM-2: The Language of Proteins
6.3.1 Proteins as Language
6.3.2 Training Objective: Masked Language Modeling
6.3.3 Architecture and Scale
6.3.4 Emergent Properties: Structure from Sequence
6.3.5 ESM-2 Architecture Overview
6.3.6 What ESM-2 Encodes
6.4 DNABERT: Tokenizing and Learning from DNA
6.4.1 The Challenge of DNA Tokenization
6.4.2 Pre-Training DNABERT
6.4.3 What DNABERT Can Predict
6.5 Nucleotide Transformer: Scaling Across Species
6.5.1 Multi-Species Pre-Training
6.5.2 Cross-Species Transfer Learning
6.5.3 Downstream Task Performance
6.6 Evo: Genome-Scale DNA Language Modeling
6.6.1 The Context Length Challenge
6.6.2 State Space Models for DNA
6.6.3 Single-Nucleotide Resolution
6.6.4 Genome-Scale Generation
6.7 Enformer: From DNA Sequence to Gene Expression
6.7.1 The Gene Expression Prediction Task
6.7.2 Architecture: Convolution Then Attention
6.7.3 Long-Range Regulatory Interactions
6.7.4 Limitations
6.8 Generative Protein Design: RFdiffusion
6.8.1 From Prediction to Design
6.8.2 RFdiffusion
6.8.3 The Inverse Folding Pipeline
6.9 Comparison of Biological Foundation Models
6.10 What Biology Teaches AI
6.10.1 Symmetry as Architectural Prior
6.10.2 Multi-Scale Structure
6.10.3 Evolutionary Conservation as a Learning Signal
6.10.4 The Generative Turn
6.11 How DOGMA Differs: An Explicit Comparison
6.11.1 DOGMA vs. AlphaFold2
6.11.2 DOGMA vs. ESM-2
6.11.3 DOGMA vs. DNABERT / Nucleotide Transformer
6.11.4 DOGMA vs. Evo
6.11.5 DOGMA vs. Enformer
6.12 The Missing Piece: Why Current Models Are Domain-Specific
6.12.1 The Fragmentation Problem
6.12.2 The Central Dogma as Unifying Principle
6.12.3 Why Unification Is Hard
6.13 The Convergence with DOGMA
6.13.1 The Genome as Structured Program
6.13.2 Regulation as Conditional Computation
6.13.3 Evolution as Optimization
6.14 Exercises
6.15 Key Takeaways
III The DOGMA Architecture
7 DOGMA — The Six-Stage Pipeline
7.1 The DOGMA Thesis
7.1.1 Why Current LLMs Are Structurally Limited
7.2 The Genomic Base: DOGMA’s Atomic Unit
7.2.1 Anatomy of a Genomic Base
7.2.2 Genomic Sequences and Regions
7.2.3 Genomic Embedding: From Raw Bytes to the 4-Tuple Base Representation
7.3 The Six-Stage Pipeline: Overview
7.3.1 Stage 1: TetraMemory
7.3.2 Stage 2: DNA Computing Block
7.3.3 Stage 3: Regulation Gate
7.3.4 Stage 4: Codon Translation
7.3.5 Stage 5: Strand Attention
7.3.6 Stage 6: Epigenetic + Allosteric Mechanisms
7.4 The Expression Folder
7.5 Full Architecture Diagram
7.6 The DogmaBlock: Complete Forward Pass
7.7 Architecture Comparison: DOGMA vs. Alternatives
7.7.1 Complexity Summary
7.8 Worked Example: Tracing Data Through All Six Stages
7.9 The Evolutor State
7.10 Why DOGMA Is Not Just a Modified Transformer
7.11 Exercises
7.12 Key Takeaways
8 HelixHash — Structural Representation of Computational Genomes
8.1 The Double-Helix as a Data Structure
8.1.1 Step-by-Step Construction of the Double Helix
8.1.2 Watson–Crick Complementarity as a Hash Function
8.1.3 Strand Alignment Score
8.2 The Bond Matrix
8.2.1 The 4×4 Hydrogen Bond Matrix
8.2.2 Bond Energies as Attention Weights
8.2.3 SantaLucia Nearest-Neighbor Parameters
8.2.4 Worked Example: Computing \(\Delta G^{\circ }\) for ACGTACGT
8.2.5 The Full Bond Matrix
8.2.6 Bond Strength
8.3 Motif Detection and Analysis
8.3.1 Defining Genomic Motifs
8.3.2 How HelixHash Detects Motifs
8.3.3 Motif Frequency Spectrum
8.3.4 Hash Function Design for DNA-Like Sequences
8.4 HelixHash as Locality-Preserving Encoding
8.4.1 Why Helical Representation Preserves Locality
8.4.2 Comparison with Standard Positional Encodings
8.4.3 Locality-Sensitive Hashing on Helical Structure
8.5 Strand Sketches
8.5.1 MinHash Lineage and Locality-Sensitive Hashing
8.6 Novelty Detection
8.6.1 Archive-Based Novelty
8.6.2 Detecting Novel vs. Seen Patterns
8.6.3 Cluster-Aware Selection
8.6.4 Temporal Novelty Decay
8.6.5 The Selection Objective
8.7 HelixHash as Embedding
8.7.1 Motif Embedding Layer
8.7.2 Dual-Stream Architecture
8.8 The HelixHash Algorithm
8.9 Theoretical Properties
8.10 Connection to the DOGMA Pipeline
8.11 Exercises
8.12 Key Takeaways
9 Tera — The Hidden Regime State
9.1 What Is a Regime State?
9.1.1 Everyday Analogies
9.1.2 Why Does DOGMA Need a Regime State?
9.2 The Problem of Multi-Objective Evolution
9.2.1 Why Single-Objective Optimization Fails
9.2.2 The Three Core Pressures: Quality, Diversity, and Coherence
9.2.3 The Pareto Landscape
9.3 The Tera Regime State
9.3.1 The Six Pressure Dimensions
9.3.2 Visualizing the Pressure Space
9.3.3 How Pressures Interact and Balance
9.3.4 Exponential Memory Decay
9.3.5 Regime Detection from Evaluation History
9.4 Mutation Families
9.4.1 What Are Mutation Families?
9.4.2 The Six Mutation Families
9.4.3 Adaptive Selection of Mutation Strategies
9.4.4 Within-Family Operator Selection
9.4.5 The Pressure Thermostat
9.5 The Tera State Machine
9.5.1 Formal Definition
9.5.2 Transition Rules
9.5.3 State Diagram
9.5.4 Output Function: Mutation Distribution by State
9.6 Meta-Policy: Learning Which Mutations Work Best
9.6.1 The Problem of Starting from Scratch
9.6.2 Cross-Run Persistent Learning
9.6.3 Shared Regime Priors
9.6.4 Comparison to Meta-Learning
9.7 Connection to Biological Evolution
9.7.1 Organisms Adapt to Environmental Pressures
9.7.2 Multi-Scale Adaptation
9.8 Pressure Landscape Visualization
9.9 Worked Example: Five Evolutionary Steps
9.9.1 Step 1: Discovery of a Grounding Problem
9.9.2 Step 2: Grounding Problem Persists
9.9.3 Step 3: Grounding Improves, Schema Problem Emerges
9.9.4 Step 4: Regime Transition to Schema
9.9.5 Step 5: Schema Takes Over
9.10 Extended Mutation Family Catalog
9.11 The Tera Double-Helix Complex
9.11.1 Forward and Reverse Strand Dynamics
9.12 Toward Post-Transformer Architectures
9.13 Exercises
9.14 Key Takeaways
10 TetraData Structure and TetraMemory
10.1 Building a Tetrahedron from Scratch
10.1.1 Step 0: A Single Vertex (0D)
10.1.2 Step 1: Add a Second Vertex (1D — an Edge)
10.1.3 Step 2: Add a Third Vertex (2D — a Triangle)
10.1.4 Step 3: Add a Fourth Vertex (3D — the Tetrahedron!)
10.1.5 The Construction at a Glance
10.2 Why Tetrahedra? — A Deeper Look
10.2.1 Tetrahedra vs. Triangles: A Detailed Comparison
10.2.2 The Biological Justification
10.3 Linking Tetrahedra — The LinkedTetraChain
10.3.1 Adding Vertex 4: The Second Tetrahedron
10.3.2 Adding Vertex 5: The Third Tetrahedron
10.3.3 The General Pattern: LinkedTetraChain
10.3.4 Growth Table
10.4 DNA Mapping — The Key Insight
10.4.1 Four Vertices, Four Bases
10.4.2 Worked Example: Mapping \(5'\)-ACGTACGT-\(3'\)
10.4.3 Why Not Triangles? An Information Density Argument
10.4.4 The Six Edges: Pairwise Base Interactions
10.5 The TetraData Structure (TDS) — Formal Definition
10.5.1 Chirality
10.5.2 Face-Sharing Adjacency
10.6 36 Degrees of Freedom
10.6.1 Detailed DOF Breakdown
10.6.2 Comparison: Triangle vs. Tetrahedron DOF
10.6.3 TetraTokenization
10.7 TetraMemory: Structured Memory for DOGMA
10.7.1 The Quadratic Attention Problem
10.7.2 Face Propagation: Information Flow Through Shared Faces
10.7.3 Volume as Attention Weight
10.7.4 Edge Weights as Pairwise Compatibility
10.7.5 Long-Range Information Flow
10.7.6 Complexity Comparison
10.8 The DoubleHelixMesh
10.8.1 Sense and Antisense Strands
10.8.2 Watson–Crick Cross-Links
10.8.3 Advantages of the Double Helix Structure
10.9 Delaunay Tetrahedralization
10.10 Multimodal Applications of TDS
10.11 Worked Examples
10.12 Exercises
10.13 Key Takeaways
11 CodonForge — The DNA Compiler and SDANS
11.1 Biological Foundations: Codon Translation in Nature
11.1.1 The Central Dogma of Molecular Biology
11.1.2 Step-by-Step: From mRNA to Amino Acid
11.1.3 The Complete Biological Codon Table
11.2 DNA as a Programming Language
11.2.1 Opcode Families
11.3 Gene Specifications
11.3.1 Gene Structure in Detail
11.3.2 Forward and Reverse Strands
11.4 The Compilation Pipeline
11.4.1 Stage 1: Parse
11.4.2 Stage 2: Compile
11.4.3 Stage 3: Optimize
11.5 Worked Example: Compiling “What is DNA?”
11.5.1 Stage 1: Parse — Building the Genomic IR
11.5.2 Stage 2: Compile — Generating the Codon Program
11.5.3 Stage 3: Optimize — Refining the Program
11.6 Execution Semantics
11.6.1 Example Execution Trace
11.7 The Word2FASTA Bridge
11.7.1 Why Encode Text as DNA?
11.7.2 The Encoding Algorithm
11.7.3 Complete Encoding Example: “Hi”
11.7.4 Decoding: FASTA Back to Text
11.8 SDANS: Strand Displacement Attention Neural System
11.8.1 HelixSim and SDANS: Clarifying the Relationship
11.8.2 Overview
11.8.3 Motivation: Why Do We Need SDANS?
11.8.4 How SDANS Maps Opcodes to Neural Operations
11.8.5 Strand Displacement as Attention
11.9 SDANS Compilation: Step-by-Step
11.9.1 Step 1: Gene-to-Layer Mapping
11.9.2 Step 2: Opcode-to-Operation Translation
11.9.3 Step 3: Parameter Initialization from Codon Hints
11.10 CodonForge vs. LLVM: A Compiler Comparison
11.11 Connection to the DOGMA Architecture
11.11.1 CodonForge in the DOGMA Pipeline
11.11.2 Proposed CodonForge-to-Neural Compilation
11.12 Advanced Topics
11.12.1 Codon Optimization Strategies
11.12.2 Error Detection and Correction
11.12.3 Multi-Gene Regulation Patterns
11.13 Exercises
11.14 Key Takeaways
IV Building the Evolutor
12 Recursive, Evidence-Gated Training
12.1 Three Loops, Three Responsibilities
12.2 The Candidate Lifecycle
12.3 Data Provenance and Model Collapse
12.4 A Multi-Probe Evaluator
12.5 Why Keep a Quality-Diversity Archive?
12.6 Controlled Mutation
12.7 Promotion Contract
12.8 Connection to Self-Improving Systems
12.9 Exercises
12.10 Key Takeaways
13 Foundation Models on Genomic Data
13.1 What Is Genomic Data?
13.1.1 From Molecules to Sequences
13.1.2 The FASTA Format: Step by Step
13.1.3 Beyond DNA: RNA and Protein Sequences
13.1.4 Ambiguity Codes and Real-World Messiness
13.2 Tokenizing Genomic Sequences
13.2.1 Why Subword Tokenization Fails for Genomic Data
13.2.2 Three Approaches to Genomic Tokenization
13.2.3 Comparison Table: Tokenization Methods
13.3 DOGMA’s Unified Tokenization: CodonForge
13.3.1 The CodonForge Vocabulary
13.3.2 Adaptive Tokenization Logic
13.3.3 Why Unified Tokenization Matters
13.4 Byte-Level Tokenization as Universal Fallback
13.5 The Bigram Baseline
13.5.1 Character-Level Statistics of Genomic Sequences
13.5.2 What Bigrams Reveal About Biology
13.6 Pre-Training Objectives for Genomic Foundation Models
13.6.1 Masked Language Modeling on DNA
13.6.2 Next-Token Prediction on Codons
13.6.3 Combining MLM and NTP
13.7 Curriculum Learning for Genomic Data
13.7.1 What Is Curriculum Learning?
13.7.2 Why Curriculum Learning Helps
13.7.3 DOGMA’s Four-Stage Curriculum
13.7.4 Stage Transitions and Mixing
13.8 Training Data Pipeline: GenBank to Batching
13.8.1 Data Sources
13.8.2 Preprocessing Steps
13.8.3 Structured Rendering
13.8.4 Curriculum Assignment and Batching
13.9 TinyGPT: Minimal Transformer for Genomic Sequences
13.9.1 Decoder-Only Architecture Adapted for DNA
13.9.2 Byte-Level Embedding with Positional Encoding
13.9.3 Training on Mixed Text/Genome Corpora
13.10 Helix-Aware Training
13.10.1 Why the Double Strand Matters for Training
13.10.2 Dual-Stream Architecture
13.10.3 Reverse Complement Consistency Loss
13.10.4 Motif-Level Feature Channels
13.10.5 Performance Comparison: Standard vs. Helix-Aware
13.11 Multi-Scale TetraMemory During Pretraining
13.11.1 What Is TetraMemory?
13.11.2 How TetraMemory Integrates with Training
13.11.3 Multi-Scale Attention Patterns
13.12 Cross-Modal Training
13.12.1 The Four Modalities
13.12.2 Cross-Modal Training Objectives
13.13 Evaluation of Genomic Foundation Models
13.13.1 Intrinsic Metrics
13.13.2 Downstream Benchmarks
13.13.3 Required Comparison with Existing Models
13.14 Proposed Large-Scale DOGMA Configuration
13.14.1 From Foundation Model to DOGMA Pipeline
13.14.2 Training Configuration Alignment
13.14.3 Transfer Learning Strategy
13.15 DOGMA for Multimodal Intelligence
13.15.1 TetraMesh: Converting 3D Structures to Byte-Level Tokens
13.15.2 Image to DNA Encoding: A Detailed Pipeline
13.15.3 Video to DNA Encoding
13.15.4 Audio to DNA Encoding
13.15.5 Cross-Modal Generation
13.15.6 The Multimodal Corpus Builder
13.15.7 Multi-Express: One Genome, Many Realizations
13.15.8 Visual Overviews
13.15.9 Use Cases for DOGMA Multimodal Intelligence
13.16 DOGMA for Conversational AI
13.16.1 The ChatSession Architecture
13.16.2 Central Dogma Context Enrichment
13.16.3 Realization Focus for Conversational Output
13.16.4 Measured ATCG-to-Language Bridge
13.16.5 Streaming Generation
13.17 Exercises
13.18 Key Takeaways
14 The Measured DOGMA Baseline
14.1 Implemented Architecture
14.2 Data Snapshot
14.3 What the Cycles Measured
14.4 A Separate Hybrid Result
14.5 Why Perplexity and Generation Disagreed
14.6 The v2 Correction
14.7 Evaluation Program
14.7.1 Tier 1: Software Correctness
14.7.2 Tier 2: Controlled Sequence Tasks
14.7.3 Tier 3: Genomic Benchmarks
14.7.4 Tier 4: Recursive-Learning Evidence
14.8 Near-Term Experiments
14.9 Exercises
14.10 Key Takeaways
V Toward AGI
15 Scaling Laws and Post-Transformer Architectures
15.1 Neural Scaling Laws
15.1.1 The Kaplan Scaling Laws
15.1.2 The Chinchilla Correction
15.1.3 Visualizing Scaling Laws
15.1.4 Emergent Abilities
15.2 The Limits of Dense Scaling
15.2.1 Quadratic Attention Cost
15.2.2 Energy and Environmental Cost
15.2.3 Data Exhaustion
15.2.4 Capability Plateaus
15.3 State Space Models: From S4 to Mamba
15.3.1 The Continuous-Time State Space
15.3.2 S4: Structured State Spaces for Sequences
15.3.3 Mamba: Selective State Spaces
15.3.4 The Hardware-Aware Scan Algorithm
15.4 RWKV and Linear Attention
15.4.1 The Quadratic Bottleneck in Attention
15.4.2 Linear Attention via Kernel Decomposition
15.4.3 RWKV: Receptance Weighted Key Value
15.5 Mixture of Experts (MoE)
15.5.1 The MoE Architecture
15.5.2 Expert Parallelism and Load Balancing
15.5.3 Expert Specialization
15.6 How DOGMA Scales
15.6.1 The Four Scaling Axes of DOGMA
15.6.2 Sample Efficiency Through Biological Priors
15.6.3 Why Four-Path Diversity Helps Scaling
15.6.4 Parameter Allocation: DOGMA vs. Transformer
15.6.5 Per-Block FLOP Analysis
15.6.6 Complexity Analysis
15.7 Hybrid Architectures
15.7.1 Attention + SSM Hybrids
15.7.2 Attention + MoE Hybrids
15.7.3 Where DOGMA Fits in the Hybrid Landscape
15.8 Comprehensive Architecture Comparison
15.9 Worked Example: Comparing Architectures on a Long Sequence
15.10 Exercises
15.11 Key Takeaways
16 Paths Toward AGI and Post-Transformer Intelligence
16.1 What Is Artificial General Intelligence?
16.1.1 Definitions and the Capability Spectrum
16.1.2 Narrow AI vs. General AI: A Comparison
16.1.3 Why Current LLMs Are Broad but Not General
16.2 The Scaling Hypothesis
16.2.1 What the Scaling Hypothesis Claims
16.2.2 Chinchilla and Compute-Optimal Training
16.2.3 Arguments For the Scaling Hypothesis
16.2.4 Arguments Against the Scaling Hypothesis
16.3 Emergent Abilities in Large Models
16.3.1 What Is Emergence?
16.3.2 Step-by-Step Examples of Emergence
16.3.3 Are Emergent Abilities Real?
16.4 Why Transformers May Not Be Enough
16.4.1 The Quadratic Attention Bottleneck
16.4.2 Reasoning Limitations
16.4.3 The World Model Problem
16.5 Post-Transformer Architectures
16.5.1 State Space Models: Mamba and Beyond
16.5.2 Mixture of Experts
16.5.3 Liquid Neural Networks
16.5.4 Neuromorphic Computing
16.6 DOGMA as a Post-Transformer Candidate
16.6.1 Addressing the Compute Wall: Regulated Propagation
16.6.2 Addressing the Reasoning Wall: Dynamic Regulation Depth
16.6.3 Addressing the Grounding Wall: Structured State as World Model
16.6.4 Compositional Verification Through Tetrahedral Structure
16.7 The Biological Precedent for AGI
16.7.1 Evolution Produced General Intelligence
16.7.2 Three Timescales of Biological Intelligence
16.7.3 Genomic Organization vs. Flat Sequence Prediction
16.7.4 Multi-Scale Organization Mirrors Hierarchical Reasoning
16.8 The Alignment Problem
16.8.1 Instrumental Convergence
16.8.2 Goodhart’s Law and Specification Gaming
16.8.3 Scalable Oversight
16.9 Architecture Comparison
16.10 The DOGMA Roadmap Toward AGI
16.10.1 Phase 1: Epigenetic Memory Layer
16.10.2 Phase 2: Ribosomal Translation
16.10.3 Phase 3: Allosteric Regulation
16.10.4 Phase 4: Chromosome-Scale Hierarchy and Horizontal Gene Transfer
16.11 Open Problems and Challenges
16.12 A Vision for the Future
16.13 Exercises
16.14 Key Takeaways
17 Ethics, Safety, and the Responsibility of Intelligence
17.1 Why AI Safety Matters Now
17.1.1 The Capability–Alignment Gap
17.1.2 Concrete Examples of AI Failures
17.2 The Alignment Problem in Depth
17.2.1 Outer Alignment
17.2.2 Inner Alignment and Mesa-Optimizers
17.2.3 Alignment in Evolutionary Systems
17.2.4 Multi-Objective Optimization as Implicit Alignment
17.3 RLHF and Constitutional AI
17.3.1 RLHF: Step by Step
17.3.2 Constitutional AI
17.3.3 Limitations of Current Alignment Techniques
17.4 Interpretability and Mechanistic Transparency
17.4.1 Feature Visualization
17.4.2 Circuit Analysis
17.4.3 DOGMA’s Transparency Advantages
17.5 DOGMA’s Safety Advantages: Biological Inspiration
17.5.1 The Immune System: Detecting and Neutralizing Threats
17.5.2 Homeostasis: Maintaining Stability Under Perturbation
17.5.3 Regulatory Networks: Constitutive and Inducible Safety
17.6 Dual-Use Concerns: DNA Computing and Biosecurity
17.6.1 Where Computational Biology Meets Synthetic Biology
17.6.2 Risk Taxonomy for Genomic AI
17.6.3 Responsible Disclosure
17.7 Fairness, Bias, and Representation
17.7.1 Sources of Bias
17.7.2 Mitigation Strategies
17.8 Environmental Impact
17.8.1 Compute Costs and Carbon Footprint
17.8.2 Efficiency as an Ethical Imperative
17.9 Governance and Regulation
17.9.1 AI Governance Frameworks
17.9.2 Biosafety Governance
17.9.3 Toward Integrated AI-Biosafety Governance
17.10 The Bio-Digital Ethics Boundary
17.10.1 Intellectual Property and Biological Inspiration
17.10.2 The Status of Computational Organisms
17.10.3 Informed Consent in Bio-Digital Research
17.11 Responsible Development Principles
17.11.1 Incremental Capability with Safety Checkpoints
17.11.2 Red-Teaming Genomic AI Systems
17.11.3 Community Oversight and Peer Review
17.11.4 Ethical Review for Evolutionary Experiments
17.12 A Worked Example: Safety-Augmented Evolutionary Training
17.13 Exercises
17.14 Key Takeaways
VI Practicum
18 Hands-On Labs — Building DOGMA from Scratch
18.1 Environment Setup
18.2 Lab 1: Implementing Watson-Crick Base Pairing in Code
18.2.1 Learning Objectives
18.2.2 Background
18.2.3 Step-by-Step Instructions
18.2.4 Expected Output
18.2.5 Discussion Questions
18.3 Lab 2: Building a Toehold-Mediated Strand Displacement Simulator
18.3.1 Learning Objectives
18.3.2 Background
18.3.3 Step-by-Step Instructions
18.3.4 Expected Output
18.3.5 Discussion Questions
18.4 Lab 3: Implementing a DNA Logic Gate (AND) as a CRN
18.4.1 Learning Objectives
18.4.2 Background
18.4.3 Step-by-Step Instructions
18.4.4 Expected Output
18.4.5 Discussion Questions
18.5 Lab 4: Building TetraMemory from Scratch
18.5.1 Learning Objectives
18.5.2 Background
18.5.3 Step-by-Step Instructions
18.5.4 Expected Output
18.5.5 Discussion Questions
18.6 Lab 5: Implementing the Thermodynamic Attention Mechanism
18.6.1 Learning Objectives
18.6.2 Background
18.6.3 Step-by-Step Instructions
18.6.4 Expected Output
18.6.5 Discussion Questions
18.7 Lab 6: Training a Minimal DOGMA Model
18.7.1 Learning Objectives
18.7.2 Background
18.7.3 Step-by-Step Instructions
18.7.4 Expected Output
18.7.5 Discussion Questions
18.8 Lab 7: Comparing DOGMA vs. Transformer on a Genomic Sequence Task
18.8.1 Learning Objectives
18.8.2 Background
18.8.3 Step-by-Step Instructions
18.8.4 Expected Output
18.8.5 Discussion Questions
18.9 Lab 8: Visualizing DOGMA’s Internal Representations
18.9.1 Learning Objectives
18.9.2 Background
18.9.3 Step-by-Step Instructions
18.9.4 Expected Output
18.9.5 Discussion Questions
18.10 Putting It All Together: The Lab Progression
18.11 Key Takeaways
19 Epilogue: The Future of Intelligence
19.1 The Journey from DNA to DOGMA
19.1.1 A Molecule That Changed Everything
19.1.2 From Biological Principles to Computational Architecture
19.2 What DOGMA Teaches Us About Intelligence
19.2.1 Intelligence Is Substrate-Independent
19.2.2 Evolution Found Solutions We Are Only Now Discovering
19.2.3 Regulation Is as Important as Computation
19.3 The Central Dogma Revisited: From Biology to DOGMA
19.3.1 Crick’s Original Formulation
19.3.2 The DOGMA Pipeline: A Computational Analog
19.3.3 What the Mapping Reveals
19.4 The Convergence of Biology and Computation
19.4.1 Wet Computing: DNA as a Computational Substrate
19.4.2 DNA Data Storage
19.4.3 Synthetic Biology and Programmable Life
19.4.4 In Vivo Computing: The Cell as a Computer
19.4.5 The Horizon of Hybrid Systems
19.5 Open Research Frontiers
19.5.1 Open Problem 1: Scaling Evolutionary Synthesis
19.5.2 Open Problem 2: Formal Theory of Regulatory Attention
19.5.3 Open Problem 3: Cross-Modal Genomic Representations
19.5.4 Open Problem 4: Interpretability Through Biological Analogy
19.5.5 Open Problem 5: Energy-Efficient Inference via Biological Sparsity
19.5.6 Open Problem 6: Developmental Programs for AI
19.5.7 Open Problem 7: Bridging Genomic Foundation Models and DOGMA
19.6 The Next Decade of Post-Transformer AI
19.6.1 Beyond Bigger Transformers
19.6.2 Predictions and Possibilities
19.6.3 What We Are Not Predicting
19.6.4 The Role of Hardware Co-Evolution
19.7 What Makes Intelligence “General”?
19.7.1 Beyond Benchmark Scores
19.7.2 Generality as Architectural Property
19.7.3 Intelligence as Understanding, Not Just Performance
19.8 Honest Assessment: Where We Stand
19.8.1 What Has Been Demonstrated
19.8.2 What Remains
19.8.3 The Community We Need
19.9 A Letter to the Next Generation
19.9.1 The Importance of Interdisciplinary Thinking
19.9.2 What You Can Do Now
19.9.3 The Moment You Are Entering
19.10 Key Takeaways: The Entire Book in Review
19.11 Final Reflection
A Mathematical Foundations
A.1 Functions, Vectors, and Tensors
A.2 Probability and Information
A.3 Gradients and Optimization
A.4 Convolution and Recurrence
A.5 Gates and Bounded Memory
A.6 Multi-Objective Evaluation
A.7 Uncertainty and Reproduction
A.8 Complexity
A.9 Review Problems
B Canonical DOGMA Formal Specification
B.1 Model Contract
B.2 Prohibited Canonical Components
B.3 Checkpoint Contract
B.4 Corpus Contract
B.5 Recursive-Cycle Contract
B.6 Evaluation Contract
B.7 Archive and Promotion
B.8 Claim Contract
C Glossary
Reader’s Roadmap
DOGMA in One Hour
I The Biological Computer
1 The Code of Life — DNA as an Information System
1.1 The Molecular Basis of Biological Information
1.1.1 Step 1: The Atomic Ingredients
1.1.2 Step 2: The Nucleoside — Sugar Plus Base
1.1.3 Step 3: The Nucleotide — Adding the Phosphate
1.1.4 Step 4: The Single Strand — A Polynucleotide Chain
1.1.5 Step 5: The Double Helix — Watson–Crick Base Pairing
1.1.6 DNA Structure: Summary of the Build-up
1.2 DNA as Information Storage
1.2.1 Information Density: Nature’s Storage Medium
1.2.2 Binary Encoding of DNA
1.2.3 The Four-Letter Alphabet: Why Four?
1.2.4 DNA Data Storage
1.3 The Central Dogma of Molecular Biology
1.3.1 Transcription: DNA to RNA, Step by Step
1.3.2 The Genetic Code: A 64-to-21 Mapping
1.3.3 Translation: The Ribosomal Decoder
1.3.4 Post-Translational Modification: Runtime Polymorphism
1.4 DNA as Information: A Quantitative Analysis
1.4.1 Shannon Entropy of Genomic Sequences
1.4.2 Redundancy in the Genetic Code
1.4.3 Comparing DNA, Natural Language, and Code
1.5 DNA Replication: Copying the Code
1.5.1 Overview: Semi-Conservative Replication
1.5.2 Step-by-Step Mechanism
1.5.3 Error Correction: Proofreading and Repair
1.6 DNA as a Storage and Computing Medium
1.6.1 DNA Strand Displacement as a Computational Primitive
1.6.2 Chemical Reaction Networks and Turing Completeness
1.7 From Molecules to Architecture: The Abstraction Bridge
1.7.1 The Genome as Source Code
1.7.2 Why Previous Biological Metaphors in CS Were Shallow
1.8 Worked Examples
1.9 Exercises
1.10 Key Takeaways
2 A History of DNA Computing — From Adleman to Molecular Programming
2.1 Adleman’s Breakthrough (1994)
2.1.1 The Hamiltonian Path Problem
2.1.2 Encoding Graphs in DNA
2.1.3 Why It Worked—And Why It Didn’t Scale
2.2 DNA Logic Gates and Boolean Computation
2.2.1 Binary Encoding in DNA
2.2.2 The AND Gate
2.2.3 The OR Gate
2.2.4 The NOT Gate (Inverter)
2.2.5 NAND and Universal Computation
2.2.6 XOR Gate and Parity Detection
2.3 DNA Arithmetic: Computing with Molecules
2.3.1 The Half Adder
2.3.2 The Full Adder
2.3.3 DNA Subtraction via Two’s Complement
2.3.4 Comparison and Decision Making
2.4 Strand Displacement Cascades
2.4.1 The Three-Strand Mechanism
2.4.2 Kinetic Control via Toehold Length
2.4.3 Seesaw Gates: Scalable Digital Logic
2.5 Restriction Enzyme Computing and Splicing Systems
2.5.1 Restriction Enzymes as Programmable Cutters
2.5.2 Splicing: DNA Recombination as Computation
2.5.3 How Splicing Achieves Turing Completeness
2.5.4 Rothemund’s DNA Turing Machine (1996)
2.5.5 Surface-Based DNA Computing
2.6 Whiplash PCR: Hairpin-Based State Machines
2.6.1 The Hairpin State Machine
2.6.2 Computation Cycle
2.7 DNA Nanotechnology and Structural Computing
2.7.1 DNA Origami
2.7.2 Tile Self-Assembly and Turing Completeness
2.8 Chemical Reaction Networks: Turing-Complete Chemistry
2.8.1 Formal Definition
2.8.2 Computing with CRNs: Step-by-Step Examples
2.8.3 Turing Completeness of CRNs
2.8.4 Deterministic vs. Stochastic Semantics
2.9 DNA Neural Networks
2.9.1 Architecture
2.9.2 How the Winner-Take-All Works
2.9.3 Significance for DOGMA
2.10 Molecular Programming Languages
2.10.1 Visual DSD
2.10.2 NUPACK
2.10.3 The Compiler Analogy
2.11 From Molecules to Silicon: The DOGMA Bridge
2.12 The State of the Art and Open Challenges
2.13 Exercises
2.14 Key Takeaways
3 Gene Regulation as Computation
3.1 The Regulatory Grammar of DNA
3.1.1 Promoters: The Entry Point
3.1.2 Enhancers: Remote Amplifiers
3.1.3 Silencers and Insulators
3.2 Transcription Factor Binding: Step by Step
3.2.1 How Transcription Factors Bind DNA
3.2.2 Activators vs. Repressors
3.2.3 Cooperative Binding and the Hill Function
3.3 Transcription Factors as Logic Gates
3.3.1 Combinatorial Regulation
3.3.2 The Lac Operon: A Biological AND Gate
3.4 Gene Regulatory Networks
3.4.1 Network Motifs
3.4.2 The Repressilator: A Biological Ring Oscillator
3.4.3 The Toggle Switch: A Biological Flip-Flop
3.4.4 GRNs as Programs
3.5 Boolean Networks in Biology
3.5.1 Kauffman’s Random Boolean Networks
3.5.2 Attractors as Cell Types
3.6 Signal Transduction as Computation
3.6.1 Kinase Cascades as Amplifiers
3.6.2 The MAPK Pathway: A Signal Processing Pipeline
3.6.3 Feedback Loops in Signaling
3.7 Epigenetics: Computation That Modifies the Reader
3.7.1 DNA Methylation
3.7.2 Histone Modification
3.7.3 The Histone Code
3.7.4 Epigenetic Memory: State Without Sequence Change
3.8 Alternative Splicing: One Genome, Many Programs
3.8.1 The Mechanism
3.8.2 Computational Significance
3.9 The ENCODE Project: Quantifying the Regulatory Genome
3.10 Synthetic Biology: Engineering Biological Circuits
3.11 From Biological Regulation to DOGMA
3.12 Exercises
3.13 Key Takeaways
II The Intelligence Stack
4 Machine Learning Foundations for Biological AI
4.1 The Learning Problem
4.1.1 Supervised Learning as Function Approximation
4.1.2 Loss Functions
4.1.3 The Bias-Variance Tradeoff
4.1.4 Unsupervised and Self-Supervised Learning
4.2 Neural Networks from First Principles
4.2.1 The Perceptron: A Single Neuron
4.2.2 Activation Functions
4.2.3 Multi-Layer Perceptrons (MLPs)
4.3 Training Neural Networks
4.3.1 Backpropagation: Computing Gradients Efficiently
4.3.2 Gradient Descent Variants
4.3.3 The Loss Landscape
4.3.4 Regularization: Preventing Overfitting
4.4 Convolutional Neural Networks
4.4.1 The Convolution Operation
4.4.2 Key CNN Concepts
4.5 Representation Learning
4.5.1 Embeddings: Mapping Symbols to Geometry
4.5.2 The Geometry of Meaning
4.5.3 Positional Encoding
4.6 Sequence Models: From RNNs to State Spaces
4.6.1 Recurrent Neural Networks
4.6.2 Long Short-Term Memory (LSTM)
4.6.3 State Space Models
4.7 Evolutionary Algorithms
4.7.1 The Genetic Algorithm: Step by Step
4.7.2 Quality-Diversity Algorithms
4.8 Multi-Objective Optimization
4.8.1 Pareto Optimality
4.8.2 Scalarization Methods
4.9 Bringing It Together: The DOGMA Learning Stack
4.10 Exercises
4.11 Key Takeaways
5 Transformers and Large Language Models
5.1 The Attention Mechanism
5.1.1 From Alignment to Attention
5.1.2 Intuition: Queries, Keys, and Values
5.1.3 Scaled Dot-Product Attention
5.1.4 Why the Scaling Factor Matters
5.1.5 Worked Example: Self-Attention with 3 Tokens
5.1.6 Visualizing Self-Attention
5.2 Multi-Head Attention
5.2.1 Why Multiple Heads?
5.2.2 Head Splitting and Concatenation
5.2.3 What Different Heads Learn
5.2.4 Multi-Head Attention Diagram
5.3 The Transformer Architecture
5.3.1 Encoder and Decoder Stacks
5.3.2 Residual Connections and Layer Normalization
5.3.3 The Complete Transformer Block
5.4 Positional Encoding
5.4.1 The Permutation Equivariance Problem
5.4.2 Sinusoidal Positional Encodings
5.4.3 Learned and Relative Position Encodings
5.5 GPT and Autoregressive Language Models
5.5.1 Decoder-Only Architecture
5.5.2 The Next-Token Prediction Objective
5.5.3 Causal Masking
5.6 Training Large Language Models
5.6.1 Training Data Scale
5.6.2 Compute Requirements
5.7 Scaling Laws for Language Models
5.7.1 Power-Law Relationships
5.7.2 Compute-Optimal Training (Chinchilla)
5.8 Emergent Abilities of Large Language Models
5.8.1 In-Context Learning
5.8.2 Few-Shot, One-Shot, and Zero-Shot Learning
5.8.3 Chain-of-Thought Reasoning
5.8.4 Instruction Following
5.9 Instruction Tuning and RLHF
5.9.1 Supervised Instruction Tuning
5.9.2 Reinforcement Learning from Human Feedback
5.10 Key Models: A Comparative View
5.10.1 GPT Family (OpenAI)
5.10.2 BERT and Encoder-Only Models
5.10.3 LLaMA and Open-Source Models
5.10.4 Comparison Table
5.11 Limitations of the Transformer Paradigm
5.11.1 Quadratic Attention Cost
5.11.2 No Persistent State
5.11.3 No Native Typed Regions
5.11.4 Output-First Bias
5.11.5 No Evolutionary Self-Modification
5.11.6 Summary: What Transformers Cannot Do That Biology Can
5.12 Exercises
5.13 Key Takeaways
6 Biological Foundation Models
6.1 What Is a Foundation Model?
6.1.1 Definition and Core Idea
6.1.2 An Analogy: The Medical Student
6.1.3 Why Biology Is Special
6.1.4 The Biological Modalities
6.2 AlphaFold: Solving Protein Structure Prediction
6.2.1 The Protein Folding Problem
6.2.2 AlphaFold2: The Pipeline Step by Step
6.2.3 The AlphaFold2 Loss Function
6.2.4 AlphaFold2 Architecture Diagram
6.2.5 Why AlphaFold2 Uses “Only” 93M Parameters
6.3 ESM-2: The Language of Proteins
6.3.1 Proteins as Language
6.3.2 Training Objective: Masked Language Modeling
6.3.3 Architecture and Scale
6.3.4 Emergent Properties: Structure from Sequence
6.3.5 ESM-2 Architecture Overview
6.3.6 What ESM-2 Encodes
6.4 DNABERT: Tokenizing and Learning from DNA
6.4.1 The Challenge of DNA Tokenization
6.4.2 Pre-Training DNABERT
6.4.3 What DNABERT Can Predict
6.5 Nucleotide Transformer: Scaling Across Species
6.5.1 Multi-Species Pre-Training
6.5.2 Cross-Species Transfer Learning
6.5.3 Downstream Task Performance
6.6 Evo: Genome-Scale DNA Language Modeling
6.6.1 The Context Length Challenge
6.6.2 State Space Models for DNA
6.6.3 Single-Nucleotide Resolution
6.6.4 Genome-Scale Generation
6.7 Enformer: From DNA Sequence to Gene Expression
6.7.1 The Gene Expression Prediction Task
6.7.2 Architecture: Convolution Then Attention
6.7.3 Long-Range Regulatory Interactions
6.7.4 Limitations
6.8 Generative Protein Design: RFdiffusion
6.8.1 From Prediction to Design
6.8.2 RFdiffusion
6.8.3 The Inverse Folding Pipeline
6.9 Comparison of Biological Foundation Models
6.10 What Biology Teaches AI
6.10.1 Symmetry as Architectural Prior
6.10.2 Multi-Scale Structure
6.10.3 Evolutionary Conservation as a Learning Signal
6.10.4 The Generative Turn
6.11 How DOGMA Differs: An Explicit Comparison
6.11.1 DOGMA vs. AlphaFold2
6.11.2 DOGMA vs. ESM-2
6.11.3 DOGMA vs. DNABERT / Nucleotide Transformer
6.11.4 DOGMA vs. Evo
6.11.5 DOGMA vs. Enformer
6.12 The Missing Piece: Why Current Models Are Domain-Specific
6.12.1 The Fragmentation Problem
6.12.2 The Central Dogma as Unifying Principle
6.12.3 Why Unification Is Hard
6.13 The Convergence with DOGMA
6.13.1 The Genome as Structured Program
6.13.2 Regulation as Conditional Computation
6.13.3 Evolution as Optimization
6.14 Exercises
6.15 Key Takeaways
III The DOGMA Architecture
7 DOGMA — The Six-Stage Pipeline
7.1 The DOGMA Thesis
7.1.1 Why Current LLMs Are Structurally Limited
7.2 The Genomic Base: DOGMA’s Atomic Unit
7.2.1 Anatomy of a Genomic Base
7.2.2 Genomic Sequences and Regions
7.2.3 Genomic Embedding: From Raw Bytes to the 4-Tuple Base Representation
7.3 The Six-Stage Pipeline: Overview
7.3.1 Stage 1: TetraMemory
7.3.2 Stage 2: DNA Computing Block
7.3.3 Stage 3: Regulation Gate
7.3.4 Stage 4: Codon Translation
7.3.5 Stage 5: Strand Attention
7.3.6 Stage 6: Epigenetic + Allosteric Mechanisms
7.4 The Expression Folder
7.5 Full Architecture Diagram
7.6 The DogmaBlock: Complete Forward Pass
7.7 Architecture Comparison: DOGMA vs. Alternatives
7.7.1 Complexity Summary
7.8 Worked Example: Tracing Data Through All Six Stages
7.9 The Evolutor State
7.10 Why DOGMA Is Not Just a Modified Transformer
7.11 Exercises
7.12 Key Takeaways
8 HelixHash — Structural Representation of Computational Genomes
8.1 The Double-Helix as a Data Structure
8.1.1 Step-by-Step Construction of the Double Helix
8.1.2 Watson–Crick Complementarity as a Hash Function
8.1.3 Strand Alignment Score
8.2 The Bond Matrix
8.2.1 The 4×4 Hydrogen Bond Matrix
8.2.2 Bond Energies as Attention Weights
8.2.3 SantaLucia Nearest-Neighbor Parameters
8.2.4 Worked Example: Computing \(\Delta G^{\circ }\) for ACGTACGT
8.2.5 The Full Bond Matrix
8.2.6 Bond Strength
8.3 Motif Detection and Analysis
8.3.1 Defining Genomic Motifs
8.3.2 How HelixHash Detects Motifs
8.3.3 Motif Frequency Spectrum
8.3.4 Hash Function Design for DNA-Like Sequences
8.4 HelixHash as Locality-Preserving Encoding
8.4.1 Why Helical Representation Preserves Locality
8.4.2 Comparison with Standard Positional Encodings
8.4.3 Locality-Sensitive Hashing on Helical Structure
8.5 Strand Sketches
8.5.1 MinHash Lineage and Locality-Sensitive Hashing
8.6 Novelty Detection
8.6.1 Archive-Based Novelty
8.6.2 Detecting Novel vs. Seen Patterns
8.6.3 Cluster-Aware Selection
8.6.4 Temporal Novelty Decay
8.6.5 The Selection Objective
8.7 HelixHash as Embedding
8.7.1 Motif Embedding Layer
8.7.2 Dual-Stream Architecture
8.8 The HelixHash Algorithm
8.9 Theoretical Properties
8.10 Connection to the DOGMA Pipeline
8.11 Exercises
8.12 Key Takeaways
9 Tera — The Hidden Regime State
9.1 What Is a Regime State?
9.1.1 Everyday Analogies
9.1.2 Why Does DOGMA Need a Regime State?
9.2 The Problem of Multi-Objective Evolution
9.2.1 Why Single-Objective Optimization Fails
9.2.2 The Three Core Pressures: Quality, Diversity, and Coherence
9.2.3 The Pareto Landscape
9.3 The Tera Regime State
9.3.1 The Six Pressure Dimensions
9.3.2 Visualizing the Pressure Space
9.3.3 How Pressures Interact and Balance
9.3.4 Exponential Memory Decay
9.3.5 Regime Detection from Evaluation History
9.4 Mutation Families
9.4.1 What Are Mutation Families?
9.4.2 The Six Mutation Families
9.4.3 Adaptive Selection of Mutation Strategies
9.4.4 Within-Family Operator Selection
9.4.5 The Pressure Thermostat
9.5 The Tera State Machine
9.5.1 Formal Definition
9.5.2 Transition Rules
9.5.3 State Diagram
9.5.4 Output Function: Mutation Distribution by State
9.6 Meta-Policy: Learning Which Mutations Work Best
9.6.1 The Problem of Starting from Scratch
9.6.2 Cross-Run Persistent Learning
9.6.3 Shared Regime Priors
9.6.4 Comparison to Meta-Learning
9.7 Connection to Biological Evolution
9.7.1 Organisms Adapt to Environmental Pressures
9.7.2 Multi-Scale Adaptation
9.8 Pressure Landscape Visualization
9.9 Worked Example: Five Evolutionary Steps
9.9.1 Step 1: Discovery of a Grounding Problem
9.9.2 Step 2: Grounding Problem Persists
9.9.3 Step 3: Grounding Improves, Schema Problem Emerges
9.9.4 Step 4: Regime Transition to Schema
9.9.5 Step 5: Schema Takes Over
9.10 Extended Mutation Family Catalog
9.11 The Tera Double-Helix Complex
9.11.1 Forward and Reverse Strand Dynamics
9.12 Toward Post-Transformer Architectures
9.13 Exercises
9.14 Key Takeaways
10 TetraData Structure and TetraMemory
10.1 Building a Tetrahedron from Scratch
10.1.1 Step 0: A Single Vertex (0D)
10.1.2 Step 1: Add a Second Vertex (1D — an Edge)
10.1.3 Step 2: Add a Third Vertex (2D — a Triangle)
10.1.4 Step 3: Add a Fourth Vertex (3D — the Tetrahedron!)
10.1.5 The Construction at a Glance
10.2 Why Tetrahedra? — A Deeper Look
10.2.1 Tetrahedra vs. Triangles: A Detailed Comparison
10.2.2 The Biological Justification
10.3 Linking Tetrahedra — The LinkedTetraChain
10.3.1 Adding Vertex 4: The Second Tetrahedron
10.3.2 Adding Vertex 5: The Third Tetrahedron
10.3.3 The General Pattern: LinkedTetraChain
10.3.4 Growth Table
10.4 DNA Mapping — The Key Insight
10.4.1 Four Vertices, Four Bases
10.4.2 Worked Example: Mapping \(5'\)-ACGTACGT-\(3'\)
10.4.3 Why Not Triangles? An Information Density Argument
10.4.4 The Six Edges: Pairwise Base Interactions
10.5 The TetraData Structure (TDS) — Formal Definition
10.5.1 Chirality
10.5.2 Face-Sharing Adjacency
10.6 36 Degrees of Freedom
10.6.1 Detailed DOF Breakdown
10.6.2 Comparison: Triangle vs. Tetrahedron DOF
10.6.3 TetraTokenization
10.7 TetraMemory: Structured Memory for DOGMA
10.7.1 The Quadratic Attention Problem
10.7.2 Face Propagation: Information Flow Through Shared Faces
10.7.3 Volume as Attention Weight
10.7.4 Edge Weights as Pairwise Compatibility
10.7.5 Long-Range Information Flow
10.7.6 Complexity Comparison
10.8 The DoubleHelixMesh
10.8.1 Sense and Antisense Strands
10.8.2 Watson–Crick Cross-Links
10.8.3 Advantages of the Double Helix Structure
10.9 Delaunay Tetrahedralization
10.10 Multimodal Applications of TDS
10.11 Worked Examples
10.12 Exercises
10.13 Key Takeaways
11 CodonForge — The DNA Compiler and SDANS
11.1 Biological Foundations: Codon Translation in Nature
11.1.1 The Central Dogma of Molecular Biology
11.1.2 Step-by-Step: From mRNA to Amino Acid
11.1.3 The Complete Biological Codon Table
11.2 DNA as a Programming Language
11.2.1 Opcode Families
11.3 Gene Specifications
11.3.1 Gene Structure in Detail
11.3.2 Forward and Reverse Strands
11.4 The Compilation Pipeline
11.4.1 Stage 1: Parse
11.4.2 Stage 2: Compile
11.4.3 Stage 3: Optimize
11.5 Worked Example: Compiling “What is DNA?”
11.5.1 Stage 1: Parse — Building the Genomic IR
11.5.2 Stage 2: Compile — Generating the Codon Program
11.5.3 Stage 3: Optimize — Refining the Program
11.6 Execution Semantics
11.6.1 Example Execution Trace
11.7 The Word2FASTA Bridge
11.7.1 Why Encode Text as DNA?
11.7.2 The Encoding Algorithm
11.7.3 Complete Encoding Example: “Hi”
11.7.4 Decoding: FASTA Back to Text
11.8 SDANS: Strand Displacement Attention Neural System
11.8.1 HelixSim and SDANS: Clarifying the Relationship
11.8.2 Overview
11.8.3 Motivation: Why Do We Need SDANS?
11.8.4 How SDANS Maps Opcodes to Neural Operations
11.8.5 Strand Displacement as Attention
11.9 SDANS Compilation: Step-by-Step
11.9.1 Step 1: Gene-to-Layer Mapping
11.9.2 Step 2: Opcode-to-Operation Translation
11.9.3 Step 3: Parameter Initialization from Codon Hints
11.10 CodonForge vs. LLVM: A Compiler Comparison
11.11 Connection to the DOGMA Architecture
11.11.1 CodonForge in the DOGMA Pipeline
11.11.2 Proposed CodonForge-to-Neural Compilation
11.12 Advanced Topics
11.12.1 Codon Optimization Strategies
11.12.2 Error Detection and Correction
11.12.3 Multi-Gene Regulation Patterns
11.13 Exercises
11.14 Key Takeaways
IV Building the Evolutor
12 Recursive, Evidence-Gated Training
12.1 Three Loops, Three Responsibilities
12.2 The Candidate Lifecycle
12.3 Data Provenance and Model Collapse
12.4 A Multi-Probe Evaluator
12.5 Why Keep a Quality-Diversity Archive?
12.6 Controlled Mutation
12.7 Promotion Contract
12.8 Connection to Self-Improving Systems
12.9 Exercises
12.10 Key Takeaways
13 Foundation Models on Genomic Data
13.1 What Is Genomic Data?
13.1.1 From Molecules to Sequences
13.1.2 The FASTA Format: Step by Step
13.1.3 Beyond DNA: RNA and Protein Sequences
13.1.4 Ambiguity Codes and Real-World Messiness
13.2 Tokenizing Genomic Sequences
13.2.1 Why Subword Tokenization Fails for Genomic Data
13.2.2 Three Approaches to Genomic Tokenization
13.2.3 Comparison Table: Tokenization Methods
13.3 DOGMA’s Unified Tokenization: CodonForge
13.3.1 The CodonForge Vocabulary
13.3.2 Adaptive Tokenization Logic
13.3.3 Why Unified Tokenization Matters
13.4 Byte-Level Tokenization as Universal Fallback
13.5 The Bigram Baseline
13.5.1 Character-Level Statistics of Genomic Sequences
13.5.2 What Bigrams Reveal About Biology
13.6 Pre-Training Objectives for Genomic Foundation Models
13.6.1 Masked Language Modeling on DNA
13.6.2 Next-Token Prediction on Codons
13.6.3 Combining MLM and NTP
13.7 Curriculum Learning for Genomic Data
13.7.1 What Is Curriculum Learning?
13.7.2 Why Curriculum Learning Helps
13.7.3 DOGMA’s Four-Stage Curriculum
13.7.4 Stage Transitions and Mixing
13.8 Training Data Pipeline: GenBank to Batching
13.8.1 Data Sources
13.8.2 Preprocessing Steps
13.8.3 Structured Rendering
13.8.4 Curriculum Assignment and Batching
13.9 TinyGPT: Minimal Transformer for Genomic Sequences
13.9.1 Decoder-Only Architecture Adapted for DNA
13.9.2 Byte-Level Embedding with Positional Encoding
13.9.3 Training on Mixed Text/Genome Corpora
13.10 Helix-Aware Training
13.10.1 Why the Double Strand Matters for Training
13.10.2 Dual-Stream Architecture
13.10.3 Reverse Complement Consistency Loss
13.10.4 Motif-Level Feature Channels
13.10.5 Performance Comparison: Standard vs. Helix-Aware
13.11 Multi-Scale TetraMemory During Pretraining
13.11.1 What Is TetraMemory?
13.11.2 How TetraMemory Integrates with Training
13.11.3 Multi-Scale Attention Patterns
13.12 Cross-Modal Training
13.12.1 The Four Modalities
13.12.2 Cross-Modal Training Objectives
13.13 Evaluation of Genomic Foundation Models
13.13.1 Intrinsic Metrics
13.13.2 Downstream Benchmarks
13.13.3 Required Comparison with Existing Models
13.14 Proposed Large-Scale DOGMA Configuration
13.14.1 From Foundation Model to DOGMA Pipeline
13.14.2 Training Configuration Alignment
13.14.3 Transfer Learning Strategy
13.15 DOGMA for Multimodal Intelligence
13.15.1 TetraMesh: Converting 3D Structures to Byte-Level Tokens
13.15.2 Image to DNA Encoding: A Detailed Pipeline
13.15.3 Video to DNA Encoding
13.15.4 Audio to DNA Encoding
13.15.5 Cross-Modal Generation
13.15.6 The Multimodal Corpus Builder
13.15.7 Multi-Express: One Genome, Many Realizations
13.15.8 Visual Overviews
13.15.9 Use Cases for DOGMA Multimodal Intelligence
13.16 DOGMA for Conversational AI
13.16.1 The ChatSession Architecture
13.16.2 Central Dogma Context Enrichment
13.16.3 Realization Focus for Conversational Output
13.16.4 Measured ATCG-to-Language Bridge
13.16.5 Streaming Generation
13.17 Exercises
13.18 Key Takeaways
14 The Measured DOGMA Baseline
14.1 Implemented Architecture
14.2 Data Snapshot
14.3 What the Cycles Measured
14.4 A Separate Hybrid Result
14.5 Why Perplexity and Generation Disagreed
14.6 The v2 Correction
14.7 Evaluation Program
14.7.1 Tier 1: Software Correctness
14.7.2 Tier 2: Controlled Sequence Tasks
14.7.3 Tier 3: Genomic Benchmarks
14.7.4 Tier 4: Recursive-Learning Evidence
14.8 Near-Term Experiments
14.9 Exercises
14.10 Key Takeaways
V Toward AGI
15 Scaling Laws and Post-Transformer Architectures
15.1 Neural Scaling Laws
15.1.1 The Kaplan Scaling Laws
15.1.2 The Chinchilla Correction
15.1.3 Visualizing Scaling Laws
15.1.4 Emergent Abilities
15.2 The Limits of Dense Scaling
15.2.1 Quadratic Attention Cost
15.2.2 Energy and Environmental Cost
15.2.3 Data Exhaustion
15.2.4 Capability Plateaus
15.3 State Space Models: From S4 to Mamba
15.3.1 The Continuous-Time State Space
15.3.2 S4: Structured State Spaces for Sequences
15.3.3 Mamba: Selective State Spaces
15.3.4 The Hardware-Aware Scan Algorithm
15.4 RWKV and Linear Attention
15.4.1 The Quadratic Bottleneck in Attention
15.4.2 Linear Attention via Kernel Decomposition
15.4.3 RWKV: Receptance Weighted Key Value
15.5 Mixture of Experts (MoE)
15.5.1 The MoE Architecture
15.5.2 Expert Parallelism and Load Balancing
15.5.3 Expert Specialization
15.6 How DOGMA Scales
15.6.1 The Four Scaling Axes of DOGMA
15.6.2 Sample Efficiency Through Biological Priors
15.6.3 Why Four-Path Diversity Helps Scaling
15.6.4 Parameter Allocation: DOGMA vs. Transformer
15.6.5 Per-Block FLOP Analysis
15.6.6 Complexity Analysis
15.7 Hybrid Architectures
15.7.1 Attention + SSM Hybrids
15.7.2 Attention + MoE Hybrids
15.7.3 Where DOGMA Fits in the Hybrid Landscape
15.8 Comprehensive Architecture Comparison
15.9 Worked Example: Comparing Architectures on a Long Sequence
15.10 Exercises
15.11 Key Takeaways
16 Paths Toward AGI and Post-Transformer Intelligence
16.1 What Is Artificial General Intelligence?
16.1.1 Definitions and the Capability Spectrum
16.1.2 Narrow AI vs. General AI: A Comparison
16.1.3 Why Current LLMs Are Broad but Not General
16.2 The Scaling Hypothesis
16.2.1 What the Scaling Hypothesis Claims
16.2.2 Chinchilla and Compute-Optimal Training
16.2.3 Arguments For the Scaling Hypothesis
16.2.4 Arguments Against the Scaling Hypothesis
16.3 Emergent Abilities in Large Models
16.3.1 What Is Emergence?
16.3.2 Step-by-Step Examples of Emergence
16.3.3 Are Emergent Abilities Real?
16.4 Why Transformers May Not Be Enough
16.4.1 The Quadratic Attention Bottleneck
16.4.2 Reasoning Limitations
16.4.3 The World Model Problem
16.5 Post-Transformer Architectures
16.5.1 State Space Models: Mamba and Beyond
16.5.2 Mixture of Experts
16.5.3 Liquid Neural Networks
16.5.4 Neuromorphic Computing
16.6 DOGMA as a Post-Transformer Candidate
16.6.1 Addressing the Compute Wall: Regulated Propagation
16.6.2 Addressing the Reasoning Wall: Dynamic Regulation Depth
16.6.3 Addressing the Grounding Wall: Structured State as World Model
16.6.4 Compositional Verification Through Tetrahedral Structure
16.7 The Biological Precedent for AGI
16.7.1 Evolution Produced General Intelligence
16.7.2 Three Timescales of Biological Intelligence
16.7.3 Genomic Organization vs. Flat Sequence Prediction
16.7.4 Multi-Scale Organization Mirrors Hierarchical Reasoning
16.8 The Alignment Problem
16.8.1 Instrumental Convergence
16.8.2 Goodhart’s Law and Specification Gaming
16.8.3 Scalable Oversight
16.9 Architecture Comparison
16.10 The DOGMA Roadmap Toward AGI
16.10.1 Phase 1: Epigenetic Memory Layer
16.10.2 Phase 2: Ribosomal Translation
16.10.3 Phase 3: Allosteric Regulation
16.10.4 Phase 4: Chromosome-Scale Hierarchy and Horizontal Gene Transfer
16.11 Open Problems and Challenges
16.12 A Vision for the Future
16.13 Exercises
16.14 Key Takeaways
17 Ethics, Safety, and the Responsibility of Intelligence
17.1 Why AI Safety Matters Now
17.1.1 The Capability–Alignment Gap
17.1.2 Concrete Examples of AI Failures
17.2 The Alignment Problem in Depth
17.2.1 Outer Alignment
17.2.2 Inner Alignment and Mesa-Optimizers
17.2.3 Alignment in Evolutionary Systems
17.2.4 Multi-Objective Optimization as Implicit Alignment
17.3 RLHF and Constitutional AI
17.3.1 RLHF: Step by Step
17.3.2 Constitutional AI
17.3.3 Limitations of Current Alignment Techniques
17.4 Interpretability and Mechanistic Transparency
17.4.1 Feature Visualization
17.4.2 Circuit Analysis
17.4.3 DOGMA’s Transparency Advantages
17.5 DOGMA’s Safety Advantages: Biological Inspiration
17.5.1 The Immune System: Detecting and Neutralizing Threats
17.5.2 Homeostasis: Maintaining Stability Under Perturbation
17.5.3 Regulatory Networks: Constitutive and Inducible Safety
17.6 Dual-Use Concerns: DNA Computing and Biosecurity
17.6.1 Where Computational Biology Meets Synthetic Biology
17.6.2 Risk Taxonomy for Genomic AI
17.6.3 Responsible Disclosure
17.7 Fairness, Bias, and Representation
17.7.1 Sources of Bias
17.7.2 Mitigation Strategies
17.8 Environmental Impact
17.8.1 Compute Costs and Carbon Footprint
17.8.2 Efficiency as an Ethical Imperative
17.9 Governance and Regulation
17.9.1 AI Governance Frameworks
17.9.2 Biosafety Governance
17.9.3 Toward Integrated AI-Biosafety Governance
17.10 The Bio-Digital Ethics Boundary
17.10.1 Intellectual Property and Biological Inspiration
17.10.2 The Status of Computational Organisms
17.10.3 Informed Consent in Bio-Digital Research
17.11 Responsible Development Principles
17.11.1 Incremental Capability with Safety Checkpoints
17.11.2 Red-Teaming Genomic AI Systems
17.11.3 Community Oversight and Peer Review
17.11.4 Ethical Review for Evolutionary Experiments
17.12 A Worked Example: Safety-Augmented Evolutionary Training
17.13 Exercises
17.14 Key Takeaways
VI Practicum
18 Hands-On Labs — Building DOGMA from Scratch
18.1 Environment Setup
18.2 Lab 1: Implementing Watson-Crick Base Pairing in Code
18.2.1 Learning Objectives
18.2.2 Background
18.2.3 Step-by-Step Instructions
18.2.4 Expected Output
18.2.5 Discussion Questions
18.3 Lab 2: Building a Toehold-Mediated Strand Displacement Simulator
18.3.1 Learning Objectives
18.3.2 Background
18.3.3 Step-by-Step Instructions
18.3.4 Expected Output
18.3.5 Discussion Questions
18.4 Lab 3: Implementing a DNA Logic Gate (AND) as a CRN
18.4.1 Learning Objectives
18.4.2 Background
18.4.3 Step-by-Step Instructions
18.4.4 Expected Output
18.4.5 Discussion Questions
18.5 Lab 4: Building TetraMemory from Scratch
18.5.1 Learning Objectives
18.5.2 Background
18.5.3 Step-by-Step Instructions
18.5.4 Expected Output
18.5.5 Discussion Questions
18.6 Lab 5: Implementing the Thermodynamic Attention Mechanism
18.6.1 Learning Objectives
18.6.2 Background
18.6.3 Step-by-Step Instructions
18.6.4 Expected Output
18.6.5 Discussion Questions
18.7 Lab 6: Training a Minimal DOGMA Model
18.7.1 Learning Objectives
18.7.2 Background
18.7.3 Step-by-Step Instructions
18.7.4 Expected Output
18.7.5 Discussion Questions
18.8 Lab 7: Comparing DOGMA vs. Transformer on a Genomic Sequence Task
18.8.1 Learning Objectives
18.8.2 Background
18.8.3 Step-by-Step Instructions
18.8.4 Expected Output
18.8.5 Discussion Questions
18.9 Lab 8: Visualizing DOGMA’s Internal Representations
18.9.1 Learning Objectives
18.9.2 Background
18.9.3 Step-by-Step Instructions
18.9.4 Expected Output
18.9.5 Discussion Questions
18.10 Putting It All Together: The Lab Progression
18.11 Key Takeaways
19 Epilogue: The Future of Intelligence
19.1 The Journey from DNA to DOGMA
19.1.1 A Molecule That Changed Everything
19.1.2 From Biological Principles to Computational Architecture
19.2 What DOGMA Teaches Us About Intelligence
19.2.1 Intelligence Is Substrate-Independent
19.2.2 Evolution Found Solutions We Are Only Now Discovering
19.2.3 Regulation Is as Important as Computation
19.3 The Central Dogma Revisited: From Biology to DOGMA
19.3.1 Crick’s Original Formulation
19.3.2 The DOGMA Pipeline: A Computational Analog
19.3.3 What the Mapping Reveals
19.4 The Convergence of Biology and Computation
19.4.1 Wet Computing: DNA as a Computational Substrate
19.4.2 DNA Data Storage
19.4.3 Synthetic Biology and Programmable Life
19.4.4 In Vivo Computing: The Cell as a Computer
19.4.5 The Horizon of Hybrid Systems
19.5 Open Research Frontiers
19.5.1 Open Problem 1: Scaling Evolutionary Synthesis
19.5.2 Open Problem 2: Formal Theory of Regulatory Attention
19.5.3 Open Problem 3: Cross-Modal Genomic Representations
19.5.4 Open Problem 4: Interpretability Through Biological Analogy
19.5.5 Open Problem 5: Energy-Efficient Inference via Biological Sparsity
19.5.6 Open Problem 6: Developmental Programs for AI
19.5.7 Open Problem 7: Bridging Genomic Foundation Models and DOGMA
19.6 The Next Decade of Post-Transformer AI
19.6.1 Beyond Bigger Transformers
19.6.2 Predictions and Possibilities
19.6.3 What We Are Not Predicting
19.6.4 The Role of Hardware Co-Evolution
19.7 What Makes Intelligence “General”?
19.7.1 Beyond Benchmark Scores
19.7.2 Generality as Architectural Property
19.7.3 Intelligence as Understanding, Not Just Performance
19.8 Honest Assessment: Where We Stand
19.8.1 What Has Been Demonstrated
19.8.2 What Remains
19.8.3 The Community We Need
19.9 A Letter to the Next Generation
19.9.1 The Importance of Interdisciplinary Thinking
19.9.2 What You Can Do Now
19.9.3 The Moment You Are Entering
19.10 Key Takeaways: The Entire Book in Review
19.11 Final Reflection
A Mathematical Foundations
A.1 Functions, Vectors, and Tensors
A.2 Probability and Information
A.3 Gradients and Optimization
A.4 Convolution and Recurrence
A.5 Gates and Bounded Memory
A.6 Multi-Objective Evaluation
A.7 Uncertainty and Reproduction
A.8 Complexity
A.9 Review Problems
B Canonical DOGMA Formal Specification
B.1 Model Contract
B.2 Prohibited Canonical Components
B.3 Checkpoint Contract
B.4 Corpus Contract
B.5 Recursive-Cycle Contract
B.6 Evaluation Contract
B.7 Archive and Promotion
B.8 Claim Contract
C Glossary
Part I
The Biological Computer