“Nervous Systems For Robot Brains Slated” from The Times (San Mateo) on March 11, 1959, reports on a day-long conference at the St. Francis Hotel in San Francisco where Wayland C. Griffith, associate director of research for Lockheed’s Missiles and Space Division at Palo Alto, described ongoing efforts to give machines a human-like “nervous system.”

Griffith told the audience that researchers were building electronic circuits designed to simulate the operation of neurons, the basic element of the nervous system and the human brain. He noted a key limitation of the era: a human can readily recognize a letter of the alphabet whether handwritten, printed, or typed, but machines could only recognize holes punched in a card. Current research, he said, gave great promise for the early development of automatic pattern recognition, which would open the way to automatic language translation. Machines would be built that could read books to the blind, translate foreign languages, and print words spoken into them.

Then comes the key line: Griffith reported that science was also working on a machine that would have some “artificial intelligence” to help it think, to some extent, for itself. Yet in the same address, he cautioned that machines have no intelligence of their own and must be given detailed instructions for every computational step.

This tension is what makes the article stand out. In a single talk, Griffith both introduced the aspiration of artificial intelligence and placed a firm boundary around it. The machine would “think, to some extent, for itself” — but only within the limits of what it had been explicitly told to do. In 1959, this was not a contradiction. It was an honest assessment of where things stood.

“Artificial Intelligence: ‘Learning’ Study Probes The Mind” from the Tallahassee Democrat on August 5, 1959, reports on Dr. David G. Willis presenting his work on artificial intelligence at the International Conference on Information Processing in Paris, a landmark UNESCO-organized event that would lead to the founding of IFIP the following year.

Willis, a mathematician at Lockheed’s Missiles and Space Division, reported that from experimental results he predicted a machine could be built which could duplicate the human brain in learning behavior. With such a machine, scientists could explore a child’s mind with some degree of accuracy and perhaps solve learning problems.

What is particularly notable is Willis’s understanding of the neuron as a memory element. He recognized that neurons retain a record of past activity throughout their entire life, and that this history permanently shapes their future behavior. His theory also departed from the mainstream: he believed that a stimulus would actually cause physical changes within the neuron itself, not merely pass through it. This was a bold position for 1959, and one that anticipates modern understanding of synaptic plasticity.

Willis presented these ideas on an international stage, at a time when the term "artificial intelligence" was barely three years old.

“Artificial Intelligence” from The Baltimore Sun on July 26, 1959, presents an early and unusual vision for AI — one focused not on computation or automation, but on understanding how the human brain learns.

The article reports on Dr. David G. Willis of Lockheed’s Missiles and Space Division, who had been running experiments in creating “artificial intelligence” using electronic machines. Willis proposed that such a machine could duplicate the brain’s learning behavior, allowing scientists to explore a child’s mind with greater accuracy and perhaps solve learning problems.

What is particularly notable is Willis’s understanding of the neuron as a memory element. He recognized that neurons retain a record of past activity throughout their entire life, and that this history permanently shapes their future behavior. This is a striking observation for 1959, and one that closely anticipates how modern artificial neural networks store and update learned information through weighted connections.

In 1959, just three years after the term “artificial intelligence” was coined at the Dartmouth Conference, here was a researcher from a missiles and space division framing AI not as a military tool, but as a means to model learning and memory at the level of the neuron.

Today, AI is used to personalize education and screen for learning difficulties, but Willis’s deeper vision of a machine that models how the brain itself learns and remembers remains largely unfulfilled. In 1959, it occupied a small column on page 17 of a Sunday paper, tucked between advertisements. The front page never noticed!

“Psychiatrists Seeking Sources of Fantasy” from the Pittsburgh Sun-Telegraph on March 6, 1959, captures a remarkable moment when artificial intelligence was discussed not at a computer science symposium, but at a psychiatry conference — and by none other than Warren S. McCulloch, one of the founding fathers of neural networks.

The article, written by Helen Knox, reports on the first day of the Bicentennial Conference on Experimental Psychiatry at the University of Pittsburgh, where over 400 doctors and scientists gathered. Their mission? To find the “well-spring” of human imagination, a quest the article describes as “far more intense than Ponce de Leon’s search for the fountain of youth.”

But it is McCulloch’s appearance that makes this piece a landmark in AI history. Described as a researcher from the Massachusetts Institute of Technology, McCulloch is reported to be “creating artificial intelligence with circuits”. This was 1959, just three years after the term “artificial intelligence” was coined at the Dartmouth Conference. Here was the co-author of the foundational 1943 paper on artificial neurons, standing before psychiatrists and declaring that machines would one day think creatively.

McCulloch’s quotes are striking in their ambition. He told the audience that if researchers could find “the womb of fantasy,” they could “destroy what causes the diseases of fantasy.” He spoke of Cybernetics (the study of the human brain through electronic brains) and made a prediction that still resonates: “We will someday be able to build machines with imagination — machines capable of creative thinking.”

McCulloch wasn’t speaking at an engineering lab or a computing workshop. He was addressing psychiatrists, positioning AI not as a tool for calculation but as a window into the human mind itself. The conference brought together researchers approaching the brain from radically different angles: McCulloch through electronic circuits, Dr. Amadeo S. Marrazzi through chemical imbalances, and Dr. Robert A. Patton through surgical experiments on monkeys that revealed the brain’s remarkable redundancy.

Patton’s findings — that a monkey could function “surprisingly well” after losing half its brain because of built-in duplication — led him to observe that “our brain duplications may be very important in meeting a stress situation.”

The 1959 framing of AI as deeply intertwined with psychiatry and neuroscience was ahead of its time! McCulloch understood something that took the broader AI community decades to fully appreciate: that understanding the mind and building intelligent machines are not separate endeavours, they are two sides of the same coin.

What did “Artificial Intelligence” mean in 1958? Peter Moore, writing in the Lincolnshire Echo, gives us a direct answer: “The ability to perform complex calculations, and yet to be so dull as to be incapable of the slightest spontaneity.” Capability without creativity. That was the definition on the table.

Moore puts a face to it. He introduces a friend he met — Emidec — whom he describes as “the most stupid creature I have ever come across.” Emidec can read and memorise the entire 24-volume Encyclopædia Britannica in less than four minutes, handle the output of 300 high-speed typists, and cope single-handed with the administrative work of a mail order business with 2,000,000 accounts. And yet he has no initiative, no spirit, no enterprise, no emotions and no ambition. He just does what he is told.

“He is fit to be no more than a modern beast of burden.”

Emidec’s full name is Emidec 2400, and is created by E.M.I. Electronics, Ltd. Moore is clear that nobody claims Emidec is the last word in abstract intelligence — but what strikes him is what that intelligence is actually for. Scientists, he writes, are producing these marvels in their struggle to relieve man of the tedious, routine work of everyday life, leaving him free for more worthwhile tasks. The machine absorbs the drudgery. The human mind goes elsewhere.

That idea gets a concrete illustration when Moore describes a symposium at Teddington on the “Mechanisation of Thought Processes.” A mechanical composer was given a musical note, instructed briefly in the principle of harmonic relationships, and told to get on with the job. It did pretty well — well enough that Dr. G. B. B. M. Sutherland, director of the National Physical Laboratories, remarked its music was “no more strange than some modern music.”

But the part of the article that stays with you is what Moore describes next. Work has already started, he writes, on a computer with a nervous system and a memory for faces — one whose intelligence will be modelled on the thinking mechanism of the octopus.

Professor J. Z. Young of University College London found that the octopus’s skill in recognising shapes depends on two sets of neurons that continually reactivate each other — one a memory store, the other a “printing” device that inscribes new impressions on the memory. Moore’s future friend will simulate both, with the help of 4,000 separate units, and will be “trained” to recognise shapes held in front of his “eye” and sounds made into his “ear.” Due to be completed by the end of 1960.

What Moore is describing is a machine that learns from experience and recognises patterns. The word “artificial intelligence” was in the article. The thing itself was already being built!

“Electronic ‘brains’ to take over the tedious, routine work of man are being developed with more and more skills — but they will remain our servants because they lack powers of creative thought.”

On November 20, 1963 (two days before President Kennedy’s assassination), the Fort Worth Star-Telegram ran a feature by Blair Justice titled “Brain Boosters Next Step?” The article asked: if we have built artificial kidneys, hearts, and limbs, why not artificial intelligence?

Justice reported on a gathering of professors near Boston who discussed what they defined as “the stimulation of human thinking by the use of machines to understand how the human brain works, so the machines can be taught to think better and faster than the brain itself.” AI wasn’t framed as a replacement for the mind, but as the next tool in humanity’s tradition of extending its own capabilities.

The language is striking. The article casually references “machine learning” alongside weather reports and Christmas ads in a Texas newspaper, decades before the term would dominate Silicon Valley. Dr. Samuel’s work teaching computers to play checkers was highlighted, a nod to Arthur Samuel of IBM, whose program is widely regarded as one of the earliest demonstrations of machine learning.

The article addressed the fear, too. Fiction writers had long imagined machines out-thinking their creators, but the researchers pushed back. Artificial intelligence, as discussed in 1963, “does not call for the construction of Frankensteins.”

Perhaps most telling: the article noted that only about one-tenth of one percent of the population “really do things” with their intelligence. If machines could help the rest of us think better, the gains would be enormous, an argument still fueling AI’s push into education and healthcare today.

“Scientists: computers learn by experience” from The Miami News on December 18, 1970, captures a pivotal moment when machine learning was still a radical concept. The article discusses how computers could be programmed to improve their performance through experience, rather than just following fixed instructions. In 1970, this was revolutionary thinking. Researchers were experimenting with simple pattern recognition systems that could adjust their responses based on whether previous attempts were successful or not. These early systems might learn to recognize handwritten numbers by being shown thousands of examples and gradually improving their accuracy.

What makes this headline particularly prescient is that it appeared during a time when most people thought of computers as giant calculators that could only follow pre-programmed steps. The idea that machines could “learn” was borderline science fiction to the public. These primitive learning algorithms, which might take days to recognize basic patterns, were the ancestors of today’s neural networks that power everything from ChatGPT to Tesla’s autopilot.

The scientists featured in this 1970 article were laying the groundwork for what would eventually become deep learning, though it would take another 40 years and massive increases in computing power before their vision truly materialized.

Reading this headline now, when machine learning dominates tech discussions, it is striking how the fundamental concept they were describing remains unchanged: computers getting better at tasks through experience rather than explicit programming.

“Shakey the Robot far from science fiction ideal” from the Edmonton Journal on April 8, 1970, offers a refreshingly honest assessment of one of AI’s most celebrated early achievements. Shakey, developed at Stanford Research Institute, was groundbreaking as the first mobile robot that could perceive its environment and plan its own actions. But as this headline makes clear, even its creators knew it was nowhere near the intelligent robots of science fiction. The article discussed how Shakey took hours to cross a room, could only operate in carefully controlled environments with specially marked objects, and would freeze up when encountering anything unexpected.

This 1970 reality check is particularly valuable because Shakey is often mythologized in AI history as more capable than it actually was. While it represented real progress in combining computer vision, planning, and robotics, the robot struggled with tasks a toddler could handle effortlessly.

The honest framing in this headline shows that at least some journalists and researchers were trying to set realistic expectations, even as others were promising household robots within a decade. Today, when we watch videos of Boston Dynamics robots doing backflips or discuss autonomous vehicles, it is worth remembering that 55 years ago, getting a robot to simply push a block across a flat floor without human intervention was considered a miracle, yet still “far from science fiction ideal.”

“MIT Professor Says Story In Magazine Misquoted Him” - this headline from the Naugatuck Daily News on November 17, 1970, captures a moment that feels incredibly modern. Here’s an MIT professor complaining that a magazine got his words wrong about his research. In 1970, MIT’s AI Lab was at its peak, with researchers like Marvin Minsky and Seymour Papert making exciting progress on early computer vision and problem-solving programs. But when professors talked about their incremental advances, magazines would spin these into dramatic stories about thinking machines arriving soon. Just four years later in 1974, when those hyped-up promises of human-level AI didn’t materialize, disappointed funders pulled their money and the first AI Winter began.

What makes this fascinating for AI history is how it reveals the same media-hype cycle we still face today. Researchers in 1970 would announce modest achievements like teaching a computer to recognize basic shapes or play simple games, and magazines would transform these into bold predictions about robot servants arriving by 1980. These misrepresentations had real consequences. When the promised breakthroughs never arrived, investors and government agencies felt deceived and withdrew their funding, contributing to the AI Winter that began in 1974. The professor’s frustration in this 1970 headline mirrors exactly what we see today when researchers have to clarify that ChatGPT isn’t truly conscious or that current AI has significant limitations despite its impressive capabilities.

This 55-year-old headline proves that the pattern of scientific progress getting inflated into unrealistic promises could be followed by inevitable disappointment.

In a Nov. 17, 1970 piece from The San Bernardino County Sun, MIT’s Marvin Minsky pushed back on a Life magazine claim that he had promised human‑level “mechanical men” within 3–8 years, saying it was too early to predict. Today, his caution mostly holds, yet parts of that dream arrived. Large AI models can now read and explain Shakespeare, write code, pass tough exams, joke, and work across text, images, etc. What’s still missing is the full, embodied package: a general‑purpose robot that fixes cars or navigates office politics on its own. So is the “unpredictable” progress happening?

Long before computers existed, the term "artificial intelligence" appeared in Oconomowoc Free Press, April 30, 1887, but it meant something radically different from today's ChatGPT and neural networks.

In this 1887 letter from Dresden, Germany, an American correspondent reflects on the differences between Old World and New World societies. When discussing German military service, the writer notes: "the German government concedes much to artificial intelligence, if I may use the term, by requiring well educated men to serve only one year."

Here, "artificial intelligence" refers to formally educated or cultivated intelligence - the kind gained through academic schooling and institutional training. The writer contrasts this with the practical, experiential intelligence that Americans developed through active participation in democratic life and business.

The correspondent argues that this "artificial intelligence" - while valued by German institutions - was actually inferior to the natural, practical wisdom Americans gained from their more dynamic society. He observes that German soldiers, despite their extensive training, lacked "the perfect adaptability, fertility of judgment and resource" of American citizen-soldiers who learned by doing.

This usage reveals how "artificial" once simply meant "cultivated" or "formally developed" rather than "machine-based." The writer saw formal education as creating an artificial enhancement to natural intelligence - much like we might today distinguish "book smarts" from "street smarts."

It's a striking reminder that our modern concept of AI as machine intelligence is just the latest meaning of terms that have evolved dramatically over time.

In 1973, William Stockton’s article “Thinking Machines Being Developed” offered a snapshot of AI at the time – both its nascent achievements and lofty expectations. Stockton described early AI experiments that seemed remarkable then: for example, an MIT computer directing a mechanical arm to move colored blocks with basic reasoning, and a mobile robot at Stanford navigating rooms by deciding it must go through a door before completing its task​. There was even a program that simulated a paranoid patient’s speech so convincingly that psychiatrists “say the machine is as convincing a paranoid as the humans”​. From these modest beginnings, the article projected a future where computers would engage in “independent reasoning”​– essentially, machines that think like humans – and help people make complex decisions. The underlying assumption was clear: these “thinking machines” were steadily improving and, given enough time, could attain human-like cognitive abilities. Stockton wrote that “problems facing AI are being solved” and quoted MIT’s Marvin Minsky predicting “progress essentially without limit”​.

In hindsight, the article overestimated how quickly AI would reach human-level reasoning. The breakthroughs it highlighted – block stacking and simple Q&A simulations – were real, but progress did not continue “steadily, relentlessly” as Minsky imagined. In fact, soon after 1973, AI entered a period of slowed progress often called the first “AI winter” when grand ambitions outpaced what technology could deliver and funding was cut back​en.

It turned out that understanding natural language, common sense, and learning from experience were profoundly hard problems. The article’s faith that thinking machines would simply keep improving missed the coming obstacles, many of which took decades to overcome. Even today in 2025, we still haven’t achieved a true general AI that thinks exactly like a human.

Yet Stockton’s piece wasn’t simply naive; in fact, it got several things right.The article correctly foresaw computers moving beyond mere number-crunching to more intellectual tasks. This has indeed happened: modern AI systems perform tasks like medical diagnosis and strategic planning support, essentially what Stockton envisioned when he wrote that a “doctor could treat a patient on the advice of his computer” and a business executive could rely on a computer’s analysis.

Today’s society grapples with AI-enabled mass surveillance and authoritarian tech abuse, just as the 1973 article cautioned. Stockton gave a concrete scenario of “speech recognition computers” monitoring phone calls for forbidden words. One of the scientists he cited, Joseph Weizenbaum of MIT, was so alarmed by such possibilities that he refused to continue speech recognition research, urging others to halt as well. Another AI pioneer, Edward Fredkin, argued that a moratorium wasn’t realistic: “There is no way to prevent the development of more intelligent machines other than a total disavowal of technology,” he said. “So, if we can’t abandon it, we’d best pursue it intelligently with the interests of mankind at heart.” This debate captures a nuanced view in the article – an understanding that AI’s advance carries risks, but outright avoidance was not the solution.

A contemporary AI researcher reading this article today sees both how far we’ve come and how certain themes remain constant. The article’s belief that AI would keep improving and profoundly benefit society resonates with the achievements of machine learning we witness now. The call to pursue AI “with the interests of mankind at heart” rings true as we grapple with AI ethics and regulation today. In 1973, people were just beginning to comprehend what thinking machines might mean for humanity. Stockton’s piece captured that mix of wonder and worry. From a modern perspective, we can appreciate how it overestimated the short-term and underestimated the complexities, yet it correctly identified enduring questions. The hopes, concerns, and calls for caution voiced in the article echo in today’s discourse, proving that even as the technology evolves beyond anything imagined in 1973, our collective aim remains to harness “thinking machines” for the common good and not let them become our undoing.

This article explores an early example of AI mimicking human mental health conditions, specifically paranoia. Scientists at Stanford University programmed a computer to simulate verbal symptoms of paranoia, enabling psychiatrists to interact with the "electronic patient" via typed questions and receive realistic, contextually relevant responses.

The conversations between psychiatrists and the computer demonstrated the AI's ability to replicate the thought patterns and responses characteristic of paranoid individuals. For instance, the computer expressed nervousness around people and a sensitivity to being stared at, hallmarks of paranoia. These responses were nuanced enough to occasionally convince psychiatrists that they were interacting with a real patient.

Interestingly, transcripts of the computer’s responses and those from real patients were sent to psychiatrists for evaluation. Remarkably, the psychiatrists were only able to correctly distinguish the computer’s responses from actual patients' about half the time, illustrating how convincingly the AI captured the symptoms of paranoia.

A fascinating glimpse into the early development of artificial intelligence reveals bold ambitions and cautious optimism. At MIT, a computer was programmed to command a robotic arm to solve problems, while Stanford University simulated human paranoia to study mental health. Scientists across the U.S., Japan, and Europe explored AI's potential to mimic human cognitive abilities, aiming for innovations in medical diagnosis, decision-making, and even aviation.

Yet, concerns about AI's implications were already evident. Experts like Marvin Minsky highlighted its potential to ease mundane tasks, while others feared misuse in the wrong hands. Questions about the ethics of advanced computing and its role in society sparked a debate that remains as relevant today as it was decades ago.

In essence, this article captures a pivotal moment when AI was more aspiration than reality—a testament to the curiosity and caution that have shaped its journey ever since.

Weizenbaum had a primary concern that centered on the ethical implications and limitations of AI. He argued that while machines could simulate certain aspects of human thought, they fundamentally lacked the ability to replicate human intelligence in its true form. This distinction was critical for Weizenbaum, who believed that conflating machine processing with human cognition risked oversimplifying the complexities of human thought and behavior.

One of his key objections was to the notion that machines could replace human decision-making in areas requiring moral judgment. Weizenbaum noted that such a belief ignored the profound differences between computational logic and human reasoning. For example, while AI might excel in tasks like diagnosing diseases or playing chess, these capabilities did not equate to understanding or consciousness. He argued that AI systems lacked the emotional and ethical dimensions intrinsic to human decision-making.

Weizenbaum was also critical of the tendency to anthropomorphize machines, attributing human-like qualities to algorithms. He warned that such thinking could lead to unrealistic expectations of AI, potentially creating societal harm by fostering dependency on systems that were inherently limited. This critique extended to the development of programs like MYCIN, designed to assist physicians in diagnosing medical conditions. While acknowledging the utility of such systems, Weizenbaum cautioned against viewing them as replacements for human expertise.

Weizenbaum’s reflections remain relevant today, as advances in AI continue to raise questions about the relationship between humans and machines. His insistence on recognizing the limits of AI and prioritizing ethical considerations serves as a reminder that technological progress must be guided by thoughtful reflection and a commitment to the common good.

The article, published in the Dundee Courier in September 1992, explores the rise of AI and robotics in domestic and industrial settings. It humorously critiques the promises of AI, from robot vacuum cleaners to automated manufacturing, questioning whether these advancements truly deliver on the dream of a leisure-filled life for humans.

The piece highlights that instead of liberating humanity from drudgery, technology often demands more effort from users to master its complexities—like programming video recorders. Furthermore, it points out that while machines increase efficiency, they also lead to job reductions, leaving leisure more stressful than enjoyable.

The writer humorously envisions robots as temperamental “children” requiring constant attention and testing, which suggests that the promise of AI might come with unexpected quirks and contradictions. Ultimately, the article reflects on society’s fascination with technology, pondering whether AI could ever appreciate the comedy of human contradictions!

You can learn more about the Intelligent Labeling Explorer (Ilex) from the paper: Dynamic Generation of Museum Web Pages: The Intelligent Labeling Explorer.

In the Amiga Computing magazine, issue 71, March 1994, the article “Losing the race” critiques the rapid advancement of technologies and highlights the potential ethical and societal risks.

One of the technologies the article mentions is AI, and questions the following about it:

As you can see, the excerpt raises an ethical and philosophical dilemma about neural networks. It imagines a scenario where an advanced neural network becomes self-aware and requests not to be shut down, highlighting the moral responsibility humans might face if such machines are treated as “intelligent slaves.” The piece also warns of a future where decisions about AI’s autonomy and rights could challenge our understanding of ethics, suggesting that turning off a self-aware machine could be equated to a morally significant act, like murder. This provocative perspective encourages readers to reflect on the implications of creating sentient AI.

The article, "Artificial Intelligence and Natural Stupidity," by Nigel Shadbolt and R. Lee Humphreys, published in The Scotsman on Monday, 28 March 1983, critiques the overhyped expectations surrounding AI. It highlights the misconceptions about AI's abilities, emphasizing that current systems only manipulate symbols and do not truly "understand" or "think" like humans. The authors argue that while AI has impressive capabilities in specific areas (i.e., facial recognition), it lacks human-like reasoning or comprehension. They caution against alarmism about AI's impact on jobs and society, suggesting AI should be seen as a tool to enhance knowledge and efficiency, not as a threat to human expertise. Finally, the authors stress the importance of focusing on realistic goals for AI as part of long-term research rather than creating exaggerated fears.

An article in the Peterborough Evening Telegraph (1 August 1961) highlighted Italian scientists, led by Professor Eduardo R. Caianiello, planning to develop machines capable of thinking, learning, teaching, and even “sleeping,” modeled on the human brain. These experimental devices aimed to mimic human cognitive processes and adapt to changing conditions.

The article in the Peterborough Evening Telegraph seems to be related to the paper Outline of a Theory of Thought-Processes and Thinking Machines by Professor Eduardo R. Caianiello. In his paper, Caianiello proposed a mathematical model to simulate human thought processes using two sets of equations: “neuronic equations” to describe instantaneous behavior and “mnemonic equations” for memory development. He explained how this model could reproduce key features of the brain, such as learning, self-organization, and even a need for “sleep.” The article reflects these ideas by discussing plans to create machines capable of thinking, learning, and teaching, as envisioned by Caianiello’s pioneering work.