Symbolic AI
AI is therefore a broad field that includes machine learning, deep learning, and approaches that do not involve learning.
From Human Work to AI
When people ask whether a computer is intelligent, a useful starting point is to ask what kind of human intellectual work the computer is designed to perform. Artificial intelligence is the effort to automate intellectual tasks normally performed by humans. This definition describes a goal, not one single technique. AI is therefore a broad field that includes machine learning, deep learning, and approaches that do not involve learning.
Following Explicit Rules
Symbolic AI describes knowledge and decisions directly through explicit rules handcrafted by programmers. A rule can connect stated information to a conclusion: if the required conditions are present, produce the specified result. The important feature is that the system's decision logic is written out explicitly rather than obtained through a learning process.
A Rule-Based Classification
Use explicit facts and an if-then rule to determine whether an object belongs to a named category.
- 1Start with facts
The system is given the facts that an object is a robin and that every robin is a bird.
- 2Apply a rule
The programmer has written the rule: if something is a bird, classify it as an animal.
- 3Produce a conclusion
Because the facts identify the object as a bird, the explicit rule produces the conclusion that it is an animal.
The system reaches the conclusion by applying programmer-written rules to explicit facts.
Two Routes to an Intelligent Result
| Aspect | Symbolic AI | Machine learning |
|---|---|---|
| How knowledge is supplied | Programmers describe knowledge and decisions through explicit rules | The system is built for problems where writing all the rules is extremely difficult |
| Best fit | Problems whose logic can be described clearly | Complex, fuzzy problems |
| Relationship to AI | One approach within the broader field of AI | One approach within the broader field of AI |
The central distinction is how the system obtains the knowledge used to produce an intelligent-looking result.
Symbolic AI and machine learning are two routes to an intelligent-looking result. In symbolic AI, programmers describe the relevant knowledge and decisions directly. In machine learning, the system is intended for problems where writing all the necessary rules would be extremely difficult. Machine learning is not separate from AI; it is one approach within the broader field.
Clear Logic and Fuzzy Inputs
Explicit rules work well when the logic of a problem can be described clearly. The programmer can state the relevant conditions and the decisions that follow from them. The situation changes when the problem is complex and fuzzy. For image classification, speech recognition, and language translation, working out all the explicit rules needed to describe the solution proved intractable. Machine learning arose as an approach for these kinds of problems.
Image classification, speech recognition, and language translation illustrate the complex, fuzzy side of the distinction. The source describes these as tasks for which writing all the explicit rules needed to describe the solution proved intractable. This does not mean that symbolic AI has no value; it means that the suitability of an approach depends on the kind of problem being addressed.
Check Your Classification
A team wants to build a system for a problem whose logic can be described clearly using programmer-written if-then rules. Which approach is the better fit according to the distinction in this article, and why?
Hints
- Look at how the problem's logic is described.
- Compare explicit handcrafted rules with the problem types associated with machine learning.
What do you think happens?
Which approach is the better fit for a problem whose logic can be described clearly using programmer-written if-then rules?
Reveal answer
Symbolic AI
Symbolic AI depends on explicit rules handcrafted by programmers and is suitable for problems whose logic can be described clearly.
Common mistake Treating artificial intelligence and machine learning as synonyms.
What it looks likeSaying that every AI system must learn from data.
Why it failsAI is a broad field that includes machine learning and approaches that do not involve learning.
Do this insteadDescribe machine learning as one approach within the broader field of artificial intelligence.
Common mistake Assuming symbolic AI means that a system learns its rules automatically.
What it looks likeCalling a set of programmer-written if-then rules machine learning simply because the result looks intelligent.
Why it failsSymbolic AI depends on explicit rules handcrafted by programmers.
Do this insteadAsk who supplied the decision logic: programmers state it directly in symbolic AI.
Common mistake Assuming explicit rules are equally practical for every task.
What it looks likeExpecting a complete list of rules to describe every case in image classification or speech recognition.
Why it failsThe source identifies these as complex, fuzzy problems for which working out all the explicit rules proved intractable.
Do this insteadFirst classify the problem as clearly logical or complex and fuzzy before choosing the approach.
Key Takeaways
- 1Artificial intelligence is the effort to automate intellectual tasks normally performed by humans.
- 2AI is a broad field that includes machine learning, deep learning, and approaches that do not involve learning.
- 3Symbolic AI uses explicit rules handcrafted by programmers.
- 4Explicit rules suit problems whose logic can be described clearly.
- 5Machine learning arose as an approach for complex, fuzzy problems such as image classification, speech recognition, and language translation, where writing all the rules is extremely difficult.
At a glance
- Artificial intelligence is defined by the human intellectual tasks it aims to automate, not by one particular technique.
- Symbolic AI represents knowledge and decisions through explicit rules handcrafted by programmers.
- Machine learning is one approach within AI, especially associated in this distinction with complex, fuzzy problems.
- The usefulness of explicit rules depends on whether the problem's logic can be described clearly.
- Image classification, speech recognition, and language translation are examples of problems for which writing all explicit rules proved intractable.