The Difference Between Symbolic AI and Connectionist AI

Industries ranging from banking to health care use AI to meet needs.  As people learn about AI, they often come across two methods of research: symbolic AI and connectionist AI.

Symbolic AI

Symbolic AI goes by several other names, including rule-based AI, classic AI and good old-fashioned AI (GOFA). Much of the early days of artificial intelligence research centered on this method, which relies on inserting human knowledge and behavioural rules into computer codes.

Humans regularly use symbols to assign meaning to the things and events in their environment. For example, if someone told a friend they just purchased a bouquet of roses, the person hearing that news could quickly conjure an image of the flowers. The idea behind symbolic AI is that these symbols become the building blocks of cognition.

An application made with this kind of AI research processes strings of characters representing real-world entities or concepts through symbols. The symbols can be arranged hierarchically or through lists and networks. Such arrangements tell the AI algorithm how the symbols relate to each other.

Processing of the information happens through something called an expert system. It contains if/then pairings that instruct the algorithm how to behave. You can think of an expert system as a human-created knowledge base. A component called an inference engine refers to the knowledge base and selects rules to apply to given symbols.

When Does Symbolic AI Work Well or Falter?

Symbolic AI works well with applications that have clear-cut rules and goals. If an AI algorithm needs to beat a human at chess, a programmer could teach it the specifics of the game. That framework gives the AI the boundaries within which to operate.

However, it falls short in applications likely to encounter variations. For example, a machine vision program might look at a product from several possible angles. It's time-consuming to create rules for every possibility. The real world has a tremendous amount of data and variations, and no one could anticipate all fluctuations in a given environment.

Connectionist AI

Although people focused on the symbolic type for the first several decades of artificial intelligence's history, a newer model called connectionist AI is more popular now. It models AI processes based on how the human brain works and its interconnected neurons. This model uses something called a perceptron to represent a single neuron.

A system built with connectionist AI gets more intelligent through increased exposure to data and learning the patterns and relationships associated with it. In contrast, symbolic AI gets hand-coded by humans. One example of connectionist AI is an artificial neural network. Each one contains hundreds of single units, artificial neurons or processing elements. They have a layered format with weights forming connections within the structure. The weights are adjustable parameters.

Every processing element contains weighted units, a transfer function and an output. Something to keep in mind about the transfer function is that it assesses multiple inputs and combines them into one output value. Each weight evaluates importance and directionality, and the weighted sum activates the neuron. Then, the activated signal passes through the transfer function and produces a single output.

When to Consider Connectionist AI

Connectionist AI is a good choice when people have a lot of high-quality training data to feed into the algorithm. Although this model gets more intelligent with increased exposure, it needs a foundation of accurate information to start the learning process. The health care industry commonly uses this kind of AI, especially when there is a wealth of medical images to use that humans checked for correctness or provided annotations for context.

However, it often cannot explain how it arrived at a solution. Thus, people should not select it as the sole or primary choice if they need to disclose to an outside party why the AI made the conclusion it did. Consider the example of using connectionist AI to decide the fate of a person accused of murder. In that case, people would likely consider it cruel and unjust to rely on AI that way without knowing why the algorithm reached its outcome.

Combining the Two Approaches

Statistics indicate that AI's impact on the global economy will be three times higher in 2030 than today. The parties that experience the most success will likely be those that use a combination of these two methods.

Since connectionist AI learns through increased information exposure, it could help a company assess supply chain needs or changing market conditions. However, if a business needs to automate repetitive and relatively simple tasks, symbolic AI could get them done. For example, if an office worker wants to move all invoices from certain clients into a dedicated folder, symbolic AI's rule-based structure suits that need.

It's easy to see that both these kinds of AI have their merits. Some scientists want to go further by blending the two into something called neuro-symbolic AI. This model learns about the world by observing it and getting question-answer pairs for inputs.

One neural network is trained on images containing scenes with small sets of objects. Another learns based on question-and-answer pairs about things in those scenes. For example, a question could ask, "What color is the bicycle?" and the answer could be "red." Another part of the system lets it recognize symbolic concepts within the text. Then, they can find visual representations of the questions or their answers within a training set's images.

Scientists working with neuro-symbolic AI believe that this approach will let AI learn and reason while performing a broad assortment of tasks without extensive training. It'll be fascinating to watch the progress made in this area.

The Evolution of a Promising Technology

AI is now something known by the mainstream and widely used. However, the distinctions here show why it's crucial to understand how certain types operate before choosing one.

Guest Author: Shannon Flynn, Managing Editor of Rehack

Interested in hearing more from our experts? you can see our previous experts blog series below:

Top AI Resources - Directory for Remote Learning
10 Must-Read AI Books in 2020
13 ‘Must-Read’ Papers from AI Experts
Top AI & Data Science Podcasts
30 Influential Women Advancing AI in 2019
‘Must-Read’ AI Papers Suggested by Experts - Pt 2
30 Influential AI Presentations from 2019
AI Across the World: Top 10 Cities in AI 2020
Female Pioneers in Computer Science You May Not Know
10 Must-Read AI Books in 2020 - Part 2
Top Women in AI 2020 - Texas Edition
2020 University/College Rankings - Computer Science, Engineering & Technology
How Netflix uses AI to Predict Your Next Series Binge - 2020
Top 5 Technical AI Presentation Videos from January 2020
20 Free AI Courses & eBooks
5 Applications of GANs - Video Presentations You Need To See
250+ Directory of Influential Women Advancing AI in 2020
The Isolation Insight - Top 50 AI Articles, Papers & Videos from Q1
Reinforcement Learning 101 - Experts Explain
The 5 Most in Demand Programming Languages in 2020