The Genesis of Artificial Intelligence: Dartmouth and the Symbolic Approach
The history of modern artificial intelligence (AI) formally begins in 1956, during the Dartmouth Conference. There, figures like John McCarthy, Claude Shannon, and Allen Newell proposed that human reasoning could be replicated by machines. This event not only coined the term "Artificial Intelligence" but also established the symbolic approach as the dominant paradigm: the idea that intelligence could be modeled by manipulating symbols and logical rules. This period, known as the Golden Years (1956–1974), was marked by overflowing enthusiasm, though limited by the hardware capabilities of the time.
LISP: The Language That Shaped Symbolic AI
In that context, LISP (List Processing) emerged, developed by John McCarthy between 1956 and 1958. Unlike languages like FORTRAN, aimed at numerical computation, LISP was designed to manipulate symbolic expressions, essential for representing knowledge and reasoning. Its structure based on linked lists offered unprecedented flexibility for building syntactic trees, semantic networks, and expert systems. Projects like the Advice Taker, a pioneer in common-sense logic, found in LISP the ideal vehicle for their implementation.
Technical Elegance and Homoiconicity
LISP was based on S-expressions as its fundamental unit, allowing data and functions to be represented with a uniform syntax. The car and cdr operations, optimized for the IBM 704 hardware, facilitated navigation through list structures. But its most radical innovation was homoiconicity: the ability to treat code as data. This property enabled metaprogramming and program self-modification, anticipating concepts that are key today in adaptive AI and reflective systems.
Laboratories, Emblematic Systems, and Functional Notation
During the 60s and 70s, LISP established itself as the central language in laboratories like the MIT AI Lab and Stanford AI Lab. McCarthy introduced conditional expressions and adopted Church's λ-notation to define recursive functions. Its prefix syntax, known as Cambridge Polish, simplified complex symbolic operations. Thanks to these features, LISP was the basis for systems like MACSYMA (symbolic algebra), SHRDLU (natural language), and MYCIN (medical diagnosis), which marked milestones in AI history.
Technical Limitations and Language Evolution
Despite its virtues, LISP faced technical challenges. The high cost of memory in the 70s and 80s limited its industrial adoption. Garbage collection, though conceptually elegant, took time to be implemented efficiently. Furthermore, its performance in numerical calculations was inferior to FORTRAN, restricting its use in scientific applications. These limitations spurred the diversification of the language into multiple dialects, each with different approaches.
Fragmentation and Consolidation: Towards Common Lisp
After 1965, dialects such as MacLisp, focused on performance, and Interlisp, which offered an interactive development environment with tools like DWIM, emerged. This fragmentation led, in the 80s, to the creation of Common Lisp (CL), which sought to integrate the best of its predecessors. CL introduced innovations like SETF and complex argument lists, though its association with Lisp Machines marginalized other dialects. Still, CL became the standard for industrial and academic applications.
The Logicist Tradition and the Challenges of Reasoning
In parallel, McCarthy developed the logicist tradition of AI, based on formal logic as a means to represent knowledge. He proposed the situation calculus to model actions and changes, and faced problems like the Frame Problem and the Qualification Problem. To address the uncertainty of human knowledge, he introduced circumscription, a form of non-monotonic reasoning that allows for default inferences and retraction in the face of new evidence. These ideas remain relevant in explainable AI and hybrid systems.
LISP as a Precursor to Functional Programming
LISP also laid the foundations for functional programming (FP), influenced by Church's λ-calculus. Its emphasis on first-class functions and recursion established principles that later expanded into dialects like Scheme, ML, and Haskell. The latter advanced functional purity with lazy evaluation and monads, key tools for managing side effects without compromising code integrity. FP became an essential paradigm for designing robust and modular systems.
AI Winters and the Transition to Machine Learning
AI faced two periods of stagnation known as the AI Winters (1974–1980 and 1987–1993), caused by unfulfilled expectations and technical limitations. These moments spurred a paradigm shift: from the symbolic approach to machine learning. The increase in computing power allowed for the training of advanced statistical models, including recurrent neural networks like LSTM. This transition laid the groundwork for the current data-driven AI boom.
Generative Models and Intelligent Chatbots
Today, AI is in a new stage starring generative models and chatbots like ChatGPT, DeepSeek, and Claude. These tools combine natural language processing with contextual reasoning, acting as assistants in complex tasks. In computer engineering, they are used to accelerate software development, from coding to documentation. Their effectiveness depends on the quality of the prompts and the user's technical knowledge, posing new challenges in training and professional ethics.
Technical Evaluation of AI Assistants in API Development
When evaluating these tools in creating a REST API, relevant technical differences are observed. DeepSeek stands out for its precise documentation in HTTP routes. ChatGPT implements functional code in Python with data persistence, although it requires minor adjustments. Claude offers clear explanations and fault tolerance but faces difficulties in tests requiring external configurations. These observations show that, while powerful, AI assistants still need expert human supervision.
Towards Neuro-Symbolic AI and the Role of the Ethical Engineer
The legacy of LISP and McCarthy's logicist vision remains relevant. Concepts like functional programming and symbolic reasoning have permeated modern languages like Python and JavaScript. The evolution points towards neuro-symbolic AI, which seeks to combine the interpretability of logic with the adaptability of neural networks. As consensus builds that AI will transform computer engineering in the next five years, it is essential for professionals to take on an ethical and technical role, ensuring progress translates into responsible and sustainable solutions.