YEEAP · FUSE-AI-101
Foundations of Artificial Intelligence
Reading Material # 1
Module 1: Introduction, History, and Evolution of AI
A Comprehensive Introduction to Artificial Intelligence —
What it is, Where it Came from, and Why it Matters
Placid Solutions & Consulting · Florida University Southeast
(FUSE-SPS). InnoLearn Systems ·
Module 1 Overview: Understanding Artificial
Intelligence. From Ancient Dreams to Modern
Machines
Introduction: Why This Module Matters
Artificial intelligence has become one of the most consequential technologies in human history.
It powers the search engines that find information for us, the recommendation systems that
suggest what we watch, the fraud detection systems that protect our bank accounts, and the
diagnostic tools that help doctors identify diseases. Within the next decade, it will reshape nearly
every profession, redefine how governments deliver services, and transform the economic
landscape of nations. For young people in Ethiopia and across Africa, understanding AI is not
optional — it is a survival skill for the twenty-first century economy.
This module provides the essential foundation for everything that follows in this course. Before
we can analyze AI applications, before we can build effective prompts, before we can evaluate
ethical risks, we need a precise and honest understanding of what AI actually is. We need to
separate fact from science fiction, capability from hype, and current reality from future possibility.
We also need to understand how AI arrived at its present state — because the history of artificial
intelligence is not just a collection of interesting facts. It is a story of intellectual ambition,
repeated failure, unexpected breakthroughs, and lessons that remain urgently relevant today.
By the end of this module, you will have a coherent mental map of AI — its definitions, its key
concepts, its historical arc, and its current moment. This mental map is the foundation on which
all your subsequent learning will be built.
Part One: What Is Artificial Intelligence?
The term "artificial intelligence" was coined in 1956 by John McCarthy, who defined it as "the
science and engineering of making intelligent machines." That definition, while historically
significant, leaves much to be clarified. What does it mean for a machine to be intelligent? And
what kinds of tasks does AI actually perform?
A more operational definition frames AI as the field of computer science dedicated to building
systems that can perform tasks that would normally require human intelligence. These tasks
include reasoning and problem-solving, learning from experience, understanding natural
language, recognizing patterns in images and sounds, making decisions under uncertainty, and
generating creative content. The range of tasks AI can perform has expanded dramatically in
recent years, but it is important to understand that current AI systems are specialized tools — they
excel within defined domains but lack the general-purpose intelligence that humans possess.
Within artificial intelligence, there are several crucial distinctions that learners must understand
from the outset. The broadest category is AI itself — the overall field. Within AI, machine learning
refers to a family of techniques in which systems learn from data rather than being explicitly
programmed with rules. Deep learning is a subset of machine learning that uses artificial neural
networks with many layers to process complex information. And generative AI refers to systems
that can produce new content — text, images, audio, code — rather than simply classifying or
predicting from existing data.
Narrow AI, General AI, and Superintelligence
Perhaps the most important conceptual distinction in AI is between narrow AI, general AI, and
superintelligence. All AI systems that exist today — without exception — are examples of narrow
AI. Narrow AI systems are designed to perform specific tasks and can only operate within the
domain for which they were trained. A system trained to detect cancer in medical images cannot
write poetry. A system trained to play chess cannot detect fraud in financial transactions. No
matter how impressive their performance in their specific domain, narrow AI systems have no
general understanding of the world.
General AI — sometimes called Artificial General Intelligence, or AGI — refers to a hypothetical
system that could perform any intellectual task a human can perform, with the same flexibility
and adaptability. AGI does not exist today. Whether it will ever exist, and when, is one of the most
actively debated questions in the field. Some researchers believe AGI is decades away; others
believe it is centuries away; and some believe the current deep learning paradigm cannot achieve
it at all.
Superintelligence refers to a hypothetical system that surpasses human intelligence across all
domains. Like AGI, superintelligence does not exist and remains firmly in the realm of
speculation. It is, however, the subject of serious academic study by researchers concerned with
AI safety and the long-term trajectory of the technology.
Critical Clarification: The AI systems you interact with today — including large
language models like ChatGPT and Gemini — are sophisticated narrow AI tools. They
are genuinely impressive, but they are not intelligent in the full human sense, and they
do not 'understand' the world the way humans do.
Part Two: Key Terminology Every AI Learner Must Know
Before tracing the history of AI, it is essential to establish a shared vocabulary. The field is filled
with technical terms that are often used loosely in popular media, leading to widespread
confusion. The following core terms will appear throughout this course and should be understood
precisely.
An algorithm is a set of step-by-step instructions that a computer follows to complete a task.
Algorithms are the fundamental building blocks of all software, including AI systems. A model is
the output of a machine learning training process — it is the mathematical structure, encoded with
learned parameters, that an AI system uses to make predictions or generate outputs. Training is
the process of exposing a machine learning model to large quantities of data so that it can adjust
its internal parameters and improve its performance on a given task. Inference is the process of
using a trained model to make predictions or generate outputs from new, previously unseen
inputs.
A dataset is the collection of examples used to train, validate, and test an AI model. The quality,
size, and representativeness of the training dataset are perhaps the most important determinants
of a model's performance. A parameter is a numerical value within a model that is adjusted during
training. Large language models like GPT-4 contain hundreds of billions of parameters. A neural
network is a computational structure loosely inspired by the structure of biological neurons in the
brain, consisting of layers of interconnected nodes. Neural networks are the dominant
architecture behind modern deep learning systems.
Labels, in the context of supervised learning, are the correct answers associated with training
examples. For instance, in a dataset used to train a medical image classifier, each image would be
labeled with the correct diagnosis. Labeling large datasets is time-consuming, expensive, and a
major constraint on AI development in resource-limited environments.
Part Three: The Birth of AI — 1940s to 1960s
The intellectual roots of artificial intelligence reach back millennia — to ancient myths of
mechanical servants, to philosophical debates about the nature of mind, and to the mathematical
work of Enlightenment thinkers who imagined that human reasoning could be formalized and
mechanized. But the modern scientific field of AI has a clear and traceable origin in the mid-
twentieth century.
The pivotal figure in AI's founding is Alan Turing, the British mathematician whose contributions
to computing theory laid the groundwork for the entire field. In his landmark 1950 paper
"Computing Machinery and Intelligence," Turing asked a question that would define the field for
decades: "Can machines think?" Rather than attempting to answer this philosophically loaded
question directly, Turing proposed an operational test — now known as the Turing Test — in which
a human judge engages in a text conversation with both a human and a machine, without knowing
which is which. If the judge cannot reliably distinguish between the two, Turing argued, the
machine could be said to exhibit intelligent behavior.
The formal founding of AI as a scientific discipline is typically dated to the summer of 1956, when
John McCarthy organized a workshop at Dartmouth College in New Hampshire, USA. The
proposal for this workshop — signed by McCarthy, Marvin Minsky, Nathaniel Rochester, and
Claude Shannon — contained the famous assertion that "every aspect of learning or any other
feature of intelligence can in principle be so precisely described that a machine can be made to
simulate it." This optimism, which in retrospect vastly underestimated the difficulty of the
problem, set the tone for the early field.
The early years of AI produced genuinely impressive results in narrow domains. Programs like
the Logic Theorist and the General Problem Solver could prove mathematical theorems and solve
structured puzzles. ELIZA, created at MIT in the 1960s, could simulate a psychotherapist by
pattern-matching user inputs to scripted responses — and famously convinced some users that
they were speaking with a real person. These demonstrations generated enormous excitement and
attracted substantial government funding, particularly in the United States.
Part Four: The AI Winters — Cycles of Promise and Disappointment
The early optimism about AI proved premature. The systems of the 1950s and 1960s excelled at
well-defined problems with clear rules and limited scope, but they could not handle the ambiguity,
context-dependence, and open-endedness of real-world tasks. Translating natural language,
recognizing objects in photographs, understanding speech — tasks that humans perform
effortlessly and automatically — proved extraordinarily difficult for rule-based AI systems.
By the early 1970s, the limitations of early AI were becoming undeniable. In 1973, the British
government commissioned a report — the Lighthill Report — that offered a devastating critique
of AI research progress. The report argued that the field had failed to deliver on its promises, that
combinatorial explosion made many AI problems computationally intractable, and that much of
the work being done had no practical applications. The British government subsequently cut most
of its AI funding, triggering the first AI Winter.
The second AI Winter occurred in the late 1980s, following the rise and collapse of expert systems.
Expert systems were AI programs that encoded the knowledge of human specialists — doctors,
engineers, financial analysts — in the form of explicit "if-then" rules. Companies invested heavily
in building expert systems for tasks ranging from medical diagnosis to oil exploration. Initially,
expert systems appeared transformative — they could reproduce the reasoning of specialists in
specific domains with impressive accuracy. But they were brittle: they could not handle situations
outside their encoded knowledge, could not learn from new cases, and were extraordinarily
expensive to build and maintain.
Historical Lesson: The AI Winters teach us that hype is the enemy of progress. When
expectations dramatically outpace actual capabilities, the backlash — in funding cuts,
credibility loss, and institutional retreat — can set back an entire field for years. This
lesson remains urgently relevant as AI experiences another period of extraordinary
enthusiasm today.
The AI Winters, however painful, were not wasted time. Researchers continued working, refining
mathematical techniques, developing new approaches, and building theoretical foundations that
would later enable dramatic breakthroughs. The history of AI is one of the clearest examples of
how persistent long-term research, even through periods of apparent failure, ultimately generates
transformative results.
Part Five: The Neural Network Revolution and the Rise of Machine
Learning
The transition from the second AI Winter to the modern era of AI was driven by three converging
developments: the theoretical rehabilitation of neural networks through a technique called
backpropagation, the exponential growth of available data, and the dramatic increase in
computational power enabled by modern processors and, later, graphics processing units (GPUs).
Neural networks had been proposed as early as the 1940s, inspired by the structure of biological
brains. The basic idea was to create networks of simple computational units — artificial neurons
— that could, by adjusting the strength of connections between them, learn to recognize patterns
in data. But early neural networks were limited to a single layer and could only solve linearly
separable problems. The crucial theoretical breakthrough was the development and
popularization of the backpropagation algorithm in the 1980s, which provided an efficient
method for training multi-layer neural networks.
The key figures in the neural network renaissance are Geoffrey Hinton, Yann LeCun, and Yoshua
Bengio — often called the "Godfathers of Deep Learning." Working through the 1980s, 1990s, and
2000s, often against the prevailing skepticism of the mainstream AI community, they developed
the theoretical and practical foundations of modern deep learning. LeCun's convolutional neural
networks achieved impressive results in recognizing handwritten digits in the 1990s, but the full
power of deep learning would not be demonstrated until the 2010s, when the scale of available
data and computing power finally matched the potential of the approach.
The decisive breakthrough came in 2012, when a team led by Geoffrey Hinton — including his
students Alex Krizhevsky and Ilya Sutskever — entered the ImageNet competition, a large-scale
computer vision challenge involving the classification of over one million images into one
thousand categories. Their deep convolutional neural network, AlexNet, reduced the error rate by
nearly half compared to the previous best result. This achievement sent shockwaves through the
research community and marked the beginning of the deep learning era. Within a few years, deep
learning had transformed computer vision, speech recognition, and natural language processing,
achieving performance levels that had seemed unattainable only a decade before.
Part Six: The Modern AI Era — Transformers, Large Language Models,
and Generative AI
The most recent chapter in AI's history — and the one most directly relevant to the world learners
are entering today — is defined by the Transformer architecture and the large language models it
enabled. In 2017, researchers at Google published a paper titled "Attention Is All You Need,"
introducing the Transformer — a novel neural network architecture that processes sequences of
data, such as words in a sentence, using a mechanism called "attention." Instead of processing
words one by one in sequence, Transformers can process all words simultaneously, allowing them
to capture long-range relationships in language with unprecedented effectiveness.
The Transformer architecture unleashed a cascade of breakthroughs. BERT (Bidirectional
Encoder Representations from Transformers), developed at Google in 2018, transformed natural
language understanding tasks. GPT (Generative Pre-trained Transformer), developed at OpenAI
beginning in 2018, demonstrated that language models trained on massive text corpora could
generate coherent, contextually appropriate text — and that this capability scaled dramatically
with model size and training data. GPT-2, GPT-3, and GPT-4 represented successive generations
of this scaling, each dramatically more capable than the last.
The release of ChatGPT to the public in November 2022 marked an inflection point in AI's
relationship with society. For the first time, a highly capable AI system was made freely accessible
to the general public through a simple conversational interface. ChatGPT reached one hundred
million users in two months — the fastest adoption of any consumer application in history. This
sparked a wave of competitive development from Google (Gemini), Meta (LLaMA), Anthropic
(Claude), and a rapidly expanding ecosystem of specialized AI applications.
Generative AI — systems that can create new content rather than simply analyzing existing
content — represents the current frontier. Modern generative AI systems can write essays and
reports, generate photorealistic images from text descriptions, compose music, write and debug
code, synthesize human voices, and engage in sophisticated multi-turn conversations. They are
being deployed across virtually every sector of the economy, and their capabilities continue to
advance at a pace that challenges even experts to track.
Part Seven: The Accelerating Pace of Change and Its Implications for
Africa
One of the most striking features of the current AI era is the pace of change. The gap between a
research breakthrough and a deployed product that affects millions of people has compressed
from years and decades to months. Large language models that would have been considered
science fiction five years ago are now accessible on smartphones to anyone with an internet
connection.
For Africa, and for Ethiopia specifically, this pace of change creates both opportunities and risks.
On the opportunity side, AI technologies can help address some of Africa's most pressing
development challenges — improving healthcare access in underserved communities, making
quality education more accessible, increasing agricultural productivity for smallholder farmers,
and extending financial services to populations that have historically been excluded from formal
banking. The mobile-first digital infrastructure that has developed across Africa creates a
platform on which AI-powered services can be rapidly deployed at scale.
On the risk side, the pace of AI development creates genuine dangers of exclusion. If AI systems
are primarily trained on data from North America and Europe, they will perform poorly on African
languages, contexts, and use cases. If access to AI tools requires high-speed internet connections
or expensive devices, the benefits of AI will accrue primarily to urban, educated populations and
widen existing inequalities. And if African workers are not equipped with the skills to work
alongside AI, automation could displace employment without creating replacement
opportunities.
This is precisely why programs like YEEAP and courses like FUSE-AI-101 are not merely
educational initiatives — they are investments in economic inclusion and national
competitiveness. The young people who complete this course are not passive recipients of a
technology being developed elsewhere. They are future participants in shaping how AI develops
in Africa, what it is used for, and who benefits from it.
Conclusion: Building Your Mental Map
This module has covered foundational territory: what AI is and what it is not, the key terminology
of the field, the historical arc from the founding of the discipline through two cycles of winter and
revival to the current era of large language models and generative AI. This is not merely historical
background — it is the context that will make everything else in this course comprehensible and
meaningful.
As you proceed through the sub-modules, keep returning to the historical perspective. The
patterns of the past — the cycles of overestimation and disappointment, the importance of data
and computation, the irreplaceable role of long-term basic research — are not relics. They are live
dynamics that continue to shape how AI develops today. The learner who understands this history
is better equipped to evaluate AI claims critically, to identify genuine breakthroughs from
marketing hype, and to think strategically about where AI is likely to go next.
A Word of Encouragement: Every expert in AI was once a beginner who built their
understanding one concept at a time. The mental map you are building in this module
is the foundation on which deep expertise is constructed. Take your time with it,
question everything, and connect what you learn to the world you live in and the future
you want to help build.
Placid Solutions & Consulting · FUSE-SPS · YEEAP Program ·
References
Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A.,
Shyam, P., Sastry, G., Askell, A., Agarwal, S., Herbert-Voss, A., Krueger, G., Henighan, T., Child,
R., Ramesh, A., Ziegler, D. M., Wu, J., Winter, C., … Amodei, D. (2020). Language models are
few-shot learners. Advances in Neural Information Processing Systems, 33, 1877–1901.
[Link]
[Link]
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of deep
bidirectional transformers for language understanding. In J. Burstein, C. Doran, & T. Solorio
(Eds.), Proceedings of the 2019 Conference of the North American Chapter of the Association
for Computational Linguistics: Human Language Technologies, Volume 1 (pp. 4171–4186).
Association for Computational Linguistics. [Link]
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). ImageNet classification with deep
convolutional neural networks. Advances in Neural Information Processing Systems, 25, 1097–
1105. [Link]
convolutional-neural-networks
LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning applied to
document recognition. Proceedings of the IEEE, 86(11), 2278–2324.
[Link]
McCarthy, J., Minsky, M. L., Rochester, N., & Shannon, C. E. (2006). A proposal for the
Dartmouth Summer Research Project on Artificial Intelligence, August 31, 1955. AI Magazine,
27(4), 12–14. [Link]
Minsky, M., & Papert, S. (1969). Perceptrons: An introduction to computational geometry. MIT
Press.
Rosenblatt, F. (1958). The perceptron: A probabilistic model for information storage and
organization in the brain. Psychological Review, 65(6), 386–408.
[Link]
Rumelhart, D. E., Hinton, G. E., & Williams, R. J. (1986). Learning representations by back-
propagating errors. Nature, 323(6088), 533–536. [Link]
Turing, A. M. (1937). On computable numbers, with an application to the
Entscheidungsproblem. Proceedings of the London Mathematical Society, s2-42(1), 230–265.
[Link]
Turing, A. M. (1950). Computing machinery and intelligence. Mind, 59(236), 433–460.
[Link]
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., &
Polosukhin, I. (2017). Attention is all you need. In Advances in Neural Information Processing
Systems 30 (pp. 5998–6008). Curran Associates. [Link]