AI Use Statement of Principles
The discovery of fire (400,000 to 1,000,000 years ago), bronze (c. 3300 BCE), the wheel (c. 3500 BCE), and iron and steel (c. 1200–1000 BCE) are widely recognised as defining landmarks in the early his-tory of human civilisation.
Although earlier hominin species such as “Aus-tralopithecus” are believed to date back more than four million years, the long evolutionary path through “Homo habilis” (2.4 million to 1.4 million years ago), “Homo erectus” (1.9 million to 110,000 years ago), and “Homo neanderthalensis” (400,000 to 40,000 years ago) is estimated to have led only to the emergence of modern “Homo sapiens sapiens” around 300,000 years ago (early fossils from Jebel Irhoud, Morocco).
There appears, therefore, to exist a direct correlation between techno-logical innovation and the pace of human development that not only ena-bled our species to adapt to its environment but also transformed the ways in which it began to shape, understand, and ultimately master the world around it (There appears, therefore, to exist a direct correlation between technological innovation and the pace of human development that not only enabled our species to adapt to its environment but also transformed the ways in which it began to shape, understand, and ulti-mately master the world around it (Agriculture is perhaps the most illus-trative example, as it is believed to have first emerged around 10,000–12,000 years ago during the “Neolithic Revolution”).
As a technology and as a system, the term “artificial intelligence” refers to computational architectures designed to perform tasks that would typically require human cognitive abilities, such as perception, pattern recognition, reasoning, prediction, decision‑making, or the generation of content.
These systems are, in essence, machine‑learning technologies capable of making their own inferences and decisions by learning statistical relationships from data, representing information through mathematical models, and producing outputs autonomously or semi‑autonomously. They have advanced far beyond the capabilities of traditional online search engines.
In engineering terms, an AI system is:
A computational system that uses machine‑learning, logic‑based, or statistical modelling techniques to infer patterns from data and generate outputs — such as predictions, classifications, recommendations, or actions — without being explicitly programmed for every possible scenario.
AI systems operate through a sequence of computational steps:
A. Data Acquisition
The system collects or is trained on datasets (text, images, audio, sensor data, numerical records, etc.). This data forms the basis for learning.
B. Model Training
The system uses algorithms — most commonly machine learning or deep learning — to detect statistical patterns in the data. Training involves adjusting internal parameters (weights) to minimise error in predictions.
C. Model Representation
The trained model becomes a mathematical structure (e.g., a neural network) that encodes relationships between inputs and outputs.
D. Inference
When the system receives new input, it applies the trained model to generate an output, such as:
• a classification (“spam / not spam”)
• a prediction (“credit risk: high / low”)
• a recommendation (“show this ad”)
• a generated output (text, image, audio)
• an action (robotic movement, automated decision)
E. Feedback and Updating
Some systems continue learning after deployment (online learning), while others remain fixed until retrained.
3. Core Components of AI Systems
1. Algorithms
Mathematical procedures that learn patterns from data. Examples include neural networks, decision trees, support vector machines, and transformers.
2. Models
The trained mathematical structures that perform inference.
3. Training Data
The information used to teach the model. Its quality, representativeness, and biases directly influence system behaviour.
4. Compute Infrastructure
Hardware such as GPUs, TPUs, and cloud clusters used to train and run models.
5. Outputs
Predictions, classifications, decisions, or generated content.
6. Human Oversight
Required in many systems to ensure safety, accuracy, and compliance.
4. Legal‑Style Definition (suitable for your AI Act text)
Artificial Intelligence (AI) refers to computational systems designed to operate with varying levels of autonomy and that, through machine‑learning, statistical, or logic‑based approaches, infer patterns from data and generate outputs such as predictions, classifications, recommendations, decisions, or content. These systems function by training mathematical models on data, applying those models to new inputs, and producing results that influence or determine physical or digital processes.
The birth of Artificial Intelligence was not a military endeavour, nor a strategic project of statecraft. It emerged from the minds of scientists, mathematicians, and philosophers who sought to understand the nature of intelligence itself. Their reflections continue to shape the field:
• “Can machines think?” — Alan Turing (1950)
• “Artificial intelligence is the science of making machines do things that would require intelligence if done by men.” — Marvin Minsky (1966)
• “Within ten years a digital computer will be the world’s chess champion.” — Herbert A. Simon (1958)
• “A year spent in artificial intelligence is enough to make one believe in God.” — Alan Perlis (1982)
• “Success in creating AI would be the biggest event in human history.” — Stephen Hawking
These statements reflect the scientific curiosity, philosophical ambition, and human imagination that gave rise to the discipline — not the pursuit of weaponry or conflict.
However, as with all great human achievements, the rise of Artificial Intelligence also brings significant responsibilities. Its power and reach have prompted the introduction of strong ethical, transparency, and accountability requirements in the development and use of AI systems. Therefore, it is in this context — and in light of the recent enactment of the European Union’s Artificial Intelligence Act — that we set forth the following policy, affirming our commitment to responsible, lawful, and trustworthy AI practices, in recognition of the human curiosity, creativity, and scientific endeavour that gave rise to this field.
When Reason First Stirred Within the Machine
Artificial Intelligence is a relatively modern scientific discipline, rooted in mathematics, logic, and computer science. Its formal birth is universally credited to the Dartmouth Summer Research Project on Artificial Intelligence of 1956, organised by American mathematicians and computer scientists John McCarthy (1927–2011) and Marvin Minsky (1927–2016), who, in that modest academic gathering, would first coin the term “artificial intelligence”, thereby marking the beginning of a new and remarkable intellectual frontier.
Much like the Manhattan Project, led by J. Robert Oppenheimer, which assembled the greatest scientific minds of its era to explore the deepest structures of the physical world, the Dartmouth Project represented a decisive moment in the United States’ investment in frontier research. Through agencies such as DARPA, the U.S. government recognised the transformative potential of advanced computation and committed itself to supporting the scientists capable of shaping this emerging field. Where the Manhattan Project reshaped the physical world through nuclear physics, Dartmouth reshaped the intellectual world by giving birth to artificial intelligence — the first faint stirring of reason within the machine.
Both the birth of artificial intelligence and the emergence of global digital connectivity share a common origin: the U.S. government’s sustained investment in visionary research through DARPA. If Dartmouth marked the moment when machines first began to exhibit the earliest forms of reasoning, ARPANET — funded by the same agency — marked the moment when computers first learned to speak to one another. Together, these parallel revolutions laid the foundations of the modern digital world.
The Pioneers of ARPANET
ARPANET, the direct ancestor of today’s Internet, was shaped by a small group of extraordinary scientists whose ideas would redefine communication, information, and society.
1. J.C.R. Licklider
Often regarded as the intellectual father of the Internet, Licklider was a psychologist‑engineer from MIT who led ARPA’s computer research division in the early 1960s. He envisioned a future of interactive, networked computing and persuaded ARPA to pursue it.
2. Larry Roberts
The ARPA program manager who built ARPANET. Roberts coordinated the project, selected its contractors, and oversaw the implementation of packet switching — the core innovation that made global networking possible.
3. Bob Kahn
ARPANET’s principal architect and later co‑inventor of TCP/IP, the protocol suite that still powers the Internet today. Kahn worked closely with researchers and engineers to design the system’s communication foundations.
4. Leonard Kleinrock
A UCLA computer scientist whose pioneering work in queueing theory provided the mathematical basis for packet switching. His models guided the design and performance analysis of the early network.
5. The BBN Engineering Team (Bolt Beranek and Newman)
Contracted by ARPA to build the first routers — the Interface Message Processors (IMPs) — BBN’s engineers constructed the hardware that made ARPANET operational in 1969. Their work transformed theoretical ideas into functioning infrastructure.
Why These Scientists Matter
Together, these researchers invented packet switching, the foundation of all modern networks; built the first wide‑area computer network; created the first internet communication protocols; and transmitted the first computer‑to‑computer message (UCLA → SRI, 1969). Their work transformed ARPANET from a Cold War research initiative into the global Internet — a system that now connects billions of people and devices across the world with the effort of a single fingertip.
The First Computers: The Foundations Beneath It All
The first computers were built two decades before ARPANET and long before the birth of AI as a discipline. They emerged during the pressures of the Second World War and the early post‑war period.
1. Konrad Zuse — Z3 (Germany, 1941)
Often considered the first fully functional, programmable, automatic digital computer. Built by German engineer Konrad Zuse.
2. Colossus (United Kingdom, 1943–1944)
The world’s first electronic, digital, programmable computer, designed by Tommy Flowers at Bletchley Park to break encrypted German communications.
3. ENIAC (United States, 1945)
The first general‑purpose electronic digital computer, developed by John Presper Eckert and John Mauchly at the University of Pennsylvania.
4. EDVAC and the Stored‑Program Concept (1945)
The idea of storing programs in memory — the foundation of modern computing — was formalised by John von Neumann.
These early machines provided the technological substrate upon which both ARPANET and AI research would later build.
DARPA and the Birth of AI Research
The Dartmouth Project was funded primarily by the United States Department of Defense (DoD), which played a decisive role in establishing and supporting the first major AI laboratories. Later, through DARPA, the U.S. government financed several foundational programs that significantly shaped the trajectory of early AI research, including the Speech Understanding Research Program, the Heuristic Programming Project, and early automated reasoning and theorem‑proving systems.
DARPA’s investments directly supported the work of McCarthy, Minsky, and their collaborators. Through substantial research grants, millions of dollars were allocated in the early decades of AI, driven by the belief that machines capable of human‑level intelligence could be developed within a generation. This environment of ambition and investment enabled the founders of AI to establish the discipline as a formal scientific field.
John McCarthy and Marvin Minsky
John McCarthy, often celebrated as the “Father of Artificial Intelligence”, was a mathematician and computer scientist whose vision and intellectual leadership defined the early field. He not only co‑founded AI as a discipline but also coined the term “Artificial Intelligence” in 1955. As one of the principal organisers of the Dartmouth Conference, McCarthy helped create the academic and conceptual foundation upon which modern AI is built.
Marvin Minsky, a cognitive scientist and pioneering theorist, co‑organised the Dartmouth Conference and later co‑founded the MIT Artificial Intelligence Laboratory. His work on symbolic reasoning, neural networks, and theories of the mind profoundly influenced the development of AI research for decades. The Dartmouth Conference is universally recognised as the moment when AI became a formal scientific discipline.
The First True AI Program: Logic Theorist (1955–1956)
The first genuine AI program was Logic Theorist, created by Allen Newell, Herbert A. Simon, and J.C. Shaw. Presented at the Dartmouth workshop, it is widely regarded as the first system capable of human‑like problem solving and automated reasoning. Logic Theorist could prove mathematical theorems from Principia Mathematica — a feat no machine had previously achieved.
Earlier Foundations of Machine Intelligence
Although AI formally began in 1956, several earlier milestones laid the intellectual groundwork.
Alan Turing’s Early Work (1930s–1950s)
In 1935, Turing described the concept of a universal computing machine, the theoretical basis of all modern computers. In 1950, he published Computing Machinery and Intelligence, introducing what later became known as the Turing Test. These contributions were not AI systems themselves, but they provided the essential theoretical foundation for machine intelligence.
Claude Shannon’s Theseus (1950)
In 1950, Claude Shannon built Theseus, a mechanical mouse capable of navigating a maze and remembering its path. This device is often cited as one of the earliest demonstrations of machine‑learning‑like behaviour, predating the formal birth of AI.