forward reasoning
**Forward reasoning** (also called **forward chaining** or **data-driven reasoning**) is the problem-solving strategy of **starting from known facts, premises, or given information and systematically applying rules** to derive new facts — building toward a conclusion step by step from the ground up.
**How Forward Reasoning Works**
1. **Start with Known Facts**: Gather all given information, premises, and initial conditions.
2. **Apply Rules**: Look for rules or inference steps that can be applied to the known facts.
3. **Derive New Facts**: Each rule application produces new information that gets added to the knowledge base.
4. **Repeat**: Continue applying rules to the growing knowledge base.
5. **Conclude**: Eventually derive the answer, or exhaust all applicable rules.
**Forward Reasoning Example**
```
Given:
- All birds have feathers.
- All animals with feathers can fly
(simplified rule).
- A robin is a bird.
Forward reasoning:
Step 1: Robin is a bird. (given)
Step 2: Robin has feathers.
(from rule 1 + step 1)
Step 3: Robin can fly.
(from rule 2 + step 2)
Conclusion: A robin can fly.
```
**Forward vs. Backward Reasoning**
- **Forward**: Start with data → apply rules → see what you can conclude. Explores broadly.
- **Backward**: Start with a specific goal → find what's needed → check availability. More focused.
- **Trade-Off**: Forward reasoning may derive many irrelevant intermediate facts. Backward reasoning may miss useful derivations that aren't obviously goal-related.
**When to Use Forward Reasoning**
- **Exploratory Analysis**: "Given these facts, what can we conclude?" — when you don't have a specific goal.
- **Data Processing Pipelines**: Process input data through a series of transformations → each step produces intermediate results → final output.
- **Sequential Computation**: Mathematical calculations where each step depends on the previous — compound interest, iterative algorithms, simulations.
- **Causal Reasoning**: "If X happens, then Y follows, then Z follows..." — tracing forward through causal chains.
- **Story/Scenario Generation**: Build a narrative forward from initial conditions — each event triggers subsequent events.
**Forward Reasoning in LLM Prompting**
- **Standard CoT** is essentially forward reasoning — the model starts from the problem statement and builds toward the answer step by step.
- Explicit instruction: "Given these facts, derive new conclusions step by step."
- **Stepwise prompting**: "What follows from fact 1? Now given that and fact 2, what follows?"
**Forward Reasoning Strengths**
- **Natural and Intuitive**: Mirrors how humans often think about problems — "if this, then that."
- **Complete**: Will eventually derive all possible conclusions from the given facts (if rules are exhaustive).
- **Easy to Follow**: Each step clearly follows from the previous — reasoning traces are easy to verify.
**Forward Reasoning Weaknesses**
- **Combinatorial Explosion**: With many facts and rules, the number of possible derivations grows rapidly — many may be irrelevant to the actual question.
- **No Goal Direction**: Without backward guidance, forward reasoning may spend effort deriving facts that don't contribute to the answer.
- **Efficiency**: For problems with a specific target, backward reasoning is often more efficient.
**Combining Forward and Backward**
- The most effective reasoning often combines both — backward reasoning identifies what's needed, forward reasoning builds from available facts toward those needs. This **bidirectional** approach is used in both AI systems and human expert reasoning.
Forward reasoning is the **most natural and commonly used reasoning strategy** — it builds knowledge incrementally from what is known, making it the default reasoning mode for both humans and language models.