Artificial intelligence has spent seven decades climbing a long, uneven ladder as what started as a handful of researchers wondering whether machines could mimic human thought has become a technology that writes reports, generates images, debugs code, and even tries to finish multi-step jobs with only light human supervision.
The progress is real and measurable. However, the claim that machines have already surpassed us across every domain of intelligence is not.
The 1950s: Teaching Computers to Follow Rules
The formal birth of AI is usually dated to the summer of 1956, when a small group of researchers gathered at Dartmouth College. Their bet was that if the processes behind learning and intelligence could be described precisely enough, a machine should be able to reproduce them.
Early systems worked by encoding rules and logical relationships. Programmers told the computer exactly how to solve a problem and the approach produced some impressive demonstrations, but it also exposed hard limits.
Real-world situations are messy and writing enough rules to cover them proved exhausting and brittle.
Machine Learning: Letting the Data Do the Teaching
The next major shift arrived when researchers stopped trying to hand-code every rule and instead let systems find patterns in data.
Train a model on thousands of legitimate and fraudulent transactions, and it could start flagging suspicious ones. Show it enough labelled photographs, and it could learn to recognise objects.
This approach opened doors in finance, healthcare, manufacturing and telecoms. Progress was uneven, computing power was limited, good data was scarce, and over-promising led to the familiar “AI winters” when funding and interest dried up.
Still, the field kept advancing, especially through neural networks loosely inspired by the brain’s interconnected neurons.
Deep Learning: The 2010s Breakthrough
Deep learning are neural networks with many layers and it delivered the improvements that changed the public conversation. In 2012, a deep-learning system scored a win in the ImageNet image-recognition contest and similar leaps followed in speech recognition, translation and medical imaging.
The key change was that machines no longer needed humans to pre-define every relevant feature. They could discover patterns themselves but that did not equal understanding or consciousness. It did mean computers became far better at specific, high-volume pattern-matching tasks.
Generative AI: Machines That Create
The 2020s brought generative systems that learn statistical patterns from vast collections of text, images or audio and then produce new material on demand. The 2017 transformer architecture made large language models practical. Suddenly, ordinary users could chat with AI in everyday language instead of writing code.
Businesses began testing the tools for customer support, research, marketing and software development. Individuals used them to summarise documents, draft emails and explore ideas. The technology also introduced a new problem which are confident-sounding answers that were simply wrong. Plausible is not the same as true and human uman verification stayed essential.
Reasoning Models and Agentic AI: Moving Beyond Instant Answers
Recent models spend extra computing effort on harder problems such as mathematics, multi-step planning and scientific questions before answering.
Some check intermediate steps or explore alternative approaches. Benchmarks show clear gains, yet performance remains uneven. Strength on one test does not guarantee reliability on the next unfamiliar task.
Alongside this sits agentic AI which is systems that try to complete broader goals by planning steps, using tools, retrieving information and interacting with software.
An agent might research a market, organise findings and draft a report. These systems are not fully autonomous. They still misinterpret instructions, hit unexpected obstacles and make mistakes.
Accountability questions become sharper when an AI system can change records or send messages on its own.
AGI, Superintelligence and the Language of Ambition
Artificial general intelligence (AGI) is described as a system that can handle a wide range of intellectual tasks at human level, rather than excelling only in narrow areas.
Artificial superintelligence (ASI) is the more ambitious idea that it is a system that substantially exceeds human capability across virtually all cognitive domains.
The 1950s: Teaching Computers to Follow Rules
The formal birth of AI is usually dated to the summer of 1956, when a small group of researchers gathered at Dartmouth College. Their bet was straightforward: if the processes behind learning and intelligence could be described precisely enough, a machine should be able to reproduce them.
Early systems worked by encoding rules and logical relationships. Programmers told the computer exactly how to solve a problem. The approach produced some impressive demonstrations, but it also exposed hard limits. Real-world situations are messy. Writing enough rules to cover them proved exhausting and brittle.
Machine Learning: Letting the Data Do the Teaching
The next major shift arrived when researchers stopped trying to hand-code every rule and instead let systems find patterns in data. Train a model on thousands of legitimate and fraudulent transactions, and it could start flagging suspicious ones. Show it enough labelled photographs, and it could learn to recognise objects.
This approach opened doors in finance, healthcare, manufacturing and telecoms. Progress was uneven. Computing power was limited, good data was scarce, and over-promising led to the familiar “AI winters” when funding and interest dried up. Still, the field kept advancing, particularly through neural networks loosely inspired by the brain’s interconnected neurons.
Deep Learning: The 2010s Breakthrough
Deep learning as neural networks with many layers delivered the improvement that changed the public conversation. In 2012 a deep-learning system scored a dramatic win in the ImageNet image-recognition contest. Similar leaps followed in speech recognition, translation and medical imaging.
The key change was that machines no longer needed humans to pre-define every relevant feature. They could discover patterns themselves. That did not equal understanding or consciousness. It did mean computers became far better at specific, high-volume pattern-matching tasks.
Generative AI: Machines That Create
The 2020s brought generative systems that learn statistical patterns from vast collections of text, images or audio and then produce new material on demand. The 2017 transformer architecture made large language models practical. Suddenly ordinary users could chat with AI in everyday language instead of writing code.
Businesses began testing the tools for customer support, research, marketing and software development. Individuals used them to summarise documents, draft emails and explore ideas. The technology also introduced a new problem: confident-sounding answers that were simply wrong. Plausible is not the same as true. Human verification stayed essential.
Reasoning Models and Agentic AI: Moving Beyond Instant Answers
Recent models spend extra computing effort on harder problems such as mathematics, multi-step planning, scientific questions before answering.
Some check intermediate steps or explore alternative approaches. Benchmarks show clear gains, yet performance remains uneven. Strength on one test does not guarantee reliability on the next unfamiliar task.
Alongside this are agentic AI which are systems that try to complete broader goals by planning steps, using tools, retrieving information and interacting with software. An agent might research a market, organise findings and draft a report.
These systems are not fully autonomous. They still misinterpret instructions, hit unexpected obstacles and make mistakes. Accountability questions become sharper when an AI system can change records or send messages on its own.
AGI, Superintelligence and the Language of Ambition
Artificial general intelligence (AGI) is usually described as a system that can handle a wide range of intellectual tasks at human level, rather than excelling only in narrow areas. Artificial superintelligence (ASI) is the more ambitious idea: a system that substantially exceeds human capability across virtually all cognitive domains.
Human intelligence includes learning from sparse experience, adapting to novel situations, exercising judgement and understanding social context.
However, current systems show varying competence across these areas.
In late September 2026 an executive order directed U.S. federal agencies to prefer the terms ‘Super Intelligence’ and ‘SI’.
What does this mean in practice?
Businesses do not need to wait for superintelligence to gain value. AI already shortens research time, supports customer service, analyses information and automates routine work.
In Nigeria and other African economies the potential includes higher productivity, local-language services and more efficient operations, provided electricity, connectivity, relevant data, skills and safeguards are in place.
The concentration of advanced computing and expertise among a small number of companies and countries will shape who captures the economic upside.
The history of AI is not a straight line from rule-based programs to human-level minds and then beyond but it is a series of practical advances in how machines represent information, learn from data, generate content, solve harder problems and interact with their environments.
Artificial Intelligence has left the laboratory and its next chapter will be decided by reliability, access, governance and the distribution of benefits as by raw technical power.
Related News
Folake Balogun is a technology journalist covering Africa’s digital economy, with a focus on startups, fintechs, venture capital, artificial intelligence, and emerging technologies. Her work explores the intersection of technology, business, and society, highlighting how innovation is reshaping industries and everyday life across Africa and global markets. She translates complex trends into insightful and impactful stories for a wider audience.
