In this series of articles titled ‘Demystifying AI Algorithms’, I would like to explore the basic nature of AI/ML algorithm categories and develop simple to understand perspective of the algorithms and their applications.  Consciously I will not get deeply into technical aspects and will deal only with those applications that explicitly touch our personal and work lives. There could be some technical examples for the sake of details, which can be ignored. In this first part, I will bring out perspectives on so called ‘Good Old Fashioned Artificial intelligence (GOFAI)’.

The thumb of Old-fashioned

We will start from where it all started and tell you what is now referred to as ‘Good Old Fashioned Artificial intelligence (GOFAI) which was prominent during mid-1950s to mid-1990s and is still relevant. Not that all young people ignore elderly old-fashioned people. Trying to mimic experts was the earliest approach to install intelligence in machines. Rules and thumb rules in expert’s knowledgebase get processed in their brain to come out with answers for questions which often surprise the ordinary souls. I refer these categories of algorithms as ‘Rules Rule’.

How Rules rule?

In this approach knowledge is represented symbolically, and logical rules framed to simulate human intelligence. Knowledge can also be coded into rules and represented as trees which can be searched. Inverse deduction is one of the methods used to arrive at a result based on available rules.

To get a feel of how inverse deduction works let’s take up a simple rule set: ‘Cow eats grass’, ‘Sheep eats grass’, ‘Horse eats horse grams’, ‘Grass is plant material’, ‘Horse grams is plant material’, ‘Herbivores eat plant material’.

From the above rule we can deduce that ‘Horse is herbivores’ and the process is known as inverse deduction. The below implementation in prolog program can produce result as to whether cow is a herbivore.

% Facts

eats(cow, grass).

eats(sheep, grass).

eats(horse, horse_grams).

eats(deer, grass).

eats(lion, deer).

is_plant_material(grass).

is_plant_material(horse_grams).

is_animal_material(deer).

% Rule: Herbivores eat plant material

herbivore(X) :- eats(X, Y), is_plant_material(Y).

% Rule: Carnivores eat animal material

carnivore(X) :- eats(X, Y), is_animal_material(Y).

% Query: Is Horse a herbivore?

?- herbivore(horse).

% Query: Is Lion a carnivore?

?- carnivore(lion).

% Query: Is Horse a herbivore?

?- herbivore(horse).

We define the facts about what each animal eats, and that grass and horse grams are plant material. When we run this Prolog program, it will deduce that “Horse is Herbivores” based on the given rules.  Unlike the imperative programming languages such as python or Java where we explicitly specify control flow using if-then-else statements,  Prolog is a declarative language where we specify what we want to achieve rather than how to achieve it.

What is the nature of GOFAI?

If your problem can be solved using a set of rules or a tree structure, and you can find the solution by following these rules or searching the tree, then it’s a good fit for the GOFAI approach.

GOFAI is well-suited for building expert systems that emulate human expertise in specific domains. MYCIN was an early expert system developed in the 1970s by Edward Shortliffe at Stanford University. Its primary purpose was to assist doctors in diagnosing and recommending treatments for patients with blood diseases, particularly bacterial infections.

Pathfinding, game playing, and puzzle solving are some of the use cases where search techniques like search, breadth-first search, and depth-first search can be used to navigate through tree representations. Other applications include medical diagnosis, legal reasoning, financial advice, and troubleshooting complex machinery.

Problems like automatic planning, scheduling, resource allocation involve searches through possible states starting from initial state and executing actions to achieve a specific goal can be handled through algorithms that are of GOFAI category.

Limitations of GOFAI

There are several notable limitations that have influenced the shift towards other AI approaches like machine learning and neural networks:

1. They lack the flexibility to adapt to new or unexpected scenarios.
2. GOFAI systems do not learn from experience.
3. As the complexity of the problem domain increases, the number of rules required can grow exponentially leading to scalability issues.

Why should we get along with this old-fashioned guy?

GOFAI provides a foundational understanding of AI principles and techniques. You may wonder the relevance of this old-fashioned fellow in the current AI world where we hear a lot about LLMs like ChatGPT.

1. GOFAI systems are often more transparent and easier to understand compared to complex machine learning models. This makes them valuable in applications where explainability is crucial, such as legal and medical decision-making.
2. Modern AI often integrates GOFAI principles with machine learning techniques to create hybrid systems. For example, combining symbolic reasoning with neural networks can enhance the interpretability and robustness of AI models.

I have pointed out how the AI world is taking a leap back to GOFAI  in my earlier blog – ‘Back to basics: Machine Learning for Human Understanding ( https://ai-positive.com/2024/05/30/back-to-basics-machine-learning-for-human-understanding/ ). In the world of ever growing AI models amidst web of AI/ML algorithms, there is an opportunity for GOFAI, which can be investigated further.