OldSchool AI doesn't belong to the trash

OldSchool AI doesn't belong to the trash

Despite the emergence of more advanced machine learning and deep learning techniques, traditional AI still has a place in the world. This is because traditional AI techniques, such as rule-based systems and expert systems, are often better suited for handling problems that require more explicit reasoning and decision-making, as well as those with limited amounts of data. Additionally, traditional AI techniques can be easier to interpret and explain, which is important in domains such as finance, law, and medicine where transparency and accountability are crucial.


Large language models (LLMs) have become the current obsession in the AI research community, with many tech companies investing heavily in them. However, experts are getting concerned that people have blindly started giving control to LLMs and other deep learning models.

Who gave my AI protein powder?

Large language models (LLMs) have been known to generate hallucinations or misleading outputs, especially when dealing with serious business use cases such as finance or healthcare. These hallucinations occur when the model generates plausible yet false information that can mislead businesses and individuals. This can have severe consequences, such as wrong investment decisions or incorrect medical diagnoses.

Highly recommended to watch Wired's "AI Support" video

What is an expert system?

A rule-based system is an AI technique that uses a set of if-then rules to make decisions or draw conclusions. The system has a knowledge base that stores the rules and a reasoning engine that applies them to input data to produce output. The rules are evaluated to determine which apply, and the reasoning engine uses them to make a decision or conclusion. Rule-based systems are relatively simple, transparent, and useful in applications where the problem domain is well understood, but they may struggle with complex or uncertain situations.

So you are saying that I should scrap my LLM project?

LLMs are designed to be best at one thing: human-like communication and fooling you that you are talking to a human. This is exactly what was lacking from your old banking chatbot and why no one used it and we should definitely utilize that capability to its full potential, but where output is deterministic, business processes are written and well-defined we should rely on traditional AI systems or even traditional business logic, rather than giving LLMs control. Do not choose A or B, pick both!

If you are interested in a more in-depth explanation, I highly recommend reading this article from Gary Marcus: Deep Learning Is Hitting A Wall