The history of artificial intelligence doesn't begin with the first computer, but with a question that has preoccupied humankind since antiquity: Can thinking be replicated by rules and machines? AI emerged as an independent field of research in 1956 with the Dartmouth Summer Research Project on Artificial Intelligence. The path to this point led through philosophy, logic, and mathematics – from the automata of antiquity through Leibniz, Boole, and Frege to Alan Turing's Turing machine of 1936 and his Turing test of 1950. Understanding this development allows us to better assess what an AI system can truly achieve – and what it cannot.
Artificial intelligence ( AI ) refers to methods that enable computer systems to perform tasks such as learning, natural language processing, pattern recognition, and reasoning. As the founder of Berger+Team in Bolzano, I've gained over 20 years of practical experience in placing this development at the intersection of digital communication, branding, and technical implementation. For your business decisions, understanding AI is particularly useful because it helps you distinguish between concrete capabilities and general performance promises.
- The origins of AI and historical milestones
- Understanding the development of AI through its methods
- Consequences for your company
- Frequently asked questions about the history and application of AI
History of Artificial Intelligence: Origins and Milestones
The history of artificial intelligence is not a linear story of progress. Research findings, economic expectations, and available computing power have developed at different rates. Furthermore, a successful application does not prove that a system can reliably solve every other task.
Antiquity and Philosophy: The desire to reconstruct thought
Before computers existed, there was the idea of mechanizing thought. These philosophical roots belong to the prehistory of AI and explain why the topic has fascinated people for centuries.
- Antiquity: The bronze giant Talos From Greek mythology, it is considered an early image of an artificial, thinking creature. Similar ideas can be found in many cultures.
- Aristotle: His syllogistics formulated for the first time rules with which conclusions can be formally derived – a foundation of logical thinking.
- Leibniz (around 1700): The philosopher and mathematician, with his guiding principle "Calculemus," demanded a computational resolution of disputed issues and built calculating machines. He combined logic and mechanics.
- Descartes (17th century): His dualism of mind and body and the idea of animals as machines shaped the debate about whether thinking can be explained mechanically at all.
Logic and mathematics as a foundation (Boole, Frege, Turing)
In the 19th and early 20th centuries, the philosophical question became a mathematical problem.
- 1854 – George Boole: In "An Investigation of the Laws of Thought," Boole described an algebra that links statements with the values true and false. Boolean algebra later became the basis of digital technology.
- 1879 – Gottlob Frege: With his "Concept Writing" system, Frege created a formal logic that made conclusions clearly representable – a milestone for the symbolic representation of knowledge.
- 1936 – Alan Turing: In "On Computable Numbers" Turing described the Turing machine, an abstract model of computation, and revealed the limits of computability. This concept is the theoretical basis for every computer.
1950–1956: Turing Test and Dartmouth
- 1950 – Alan Turing: In “Computing Machinery and Intelligence” Turing described the "Imitation Game". This led to the discussion about the Turing testCan a person reliably distinguish between a machine and a human based on written communication? The test is an indicator of language ability, not proof of consciousness.
- 1955 – Research proposal: John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon formulated the Application for the Dartmouth Summer Research Project on Artificial IntelligenceThe application already used the term "Artificial Intelligence".
- 1956 – Dartmouth: The research project is considered the institutional starting point of the AI field. However, the theoretical foundations and initial work on intelligent machines originated earlier.
1960s to 1990s: Rules, expert systems and AI winter
During this phase of AI development, symbolic approaches shaped many research programs. Systems were expected to solve problems based on explicitly described rules and knowledge bases.
- From the 1960s onwards: Expert systems As DENDRAL demonstrated, specialized expertise can be used for narrowly defined tasks. An expert system combines a knowledge base with a method that derives conclusions from it. In the 1980s, expert systems also gained economic importance.
- AI Winter: Particularly in parts of the 1970s and from the late 1980s onward, disappointed expectations, technological limitations, and economic factors led to a decline in funding. The exact timeframes vary depending on the research area and region – sources are inconsistent, and AI research never completely came to a standstill.
- 1997 – Deep Blue: IBM's chess computer won a match against the reigning world champion Garry Kasparov. This milestone demonstrated specialized playing strength, not general human intelligence.
Since 2012: Deep Learning, Transformers and generative AI
- 2012 – AlexNet: A deep neural network achieved a significant lead in the ImageNet image recognition competition. The result reinforced the importance of Deep Learning for image processing; however, neural networks were not a new invention.
- 2016 – AlphaGo: Google DeepMind's system defeated Lee Sedol in the board game Go. AlphaGo combined, among other things, neural networks, reinforcement learning, and search algorithms.
- 2017 – Transformer: Ashish Vaswani and his co-authors presented in “Attention Is All You Need” credentials for Transform-Architecture. This architecture became the basis for many major language models.
- 2022 – ChatGPT: OpenAI published on November 30th ChatGPT as a publicly accessible research previewThe publication facilitated access to generative AI, but that was not the invention of generative AI.
These AI milestones distinguish three levels: the development of a method, the demonstration of a concrete achievement, and the widespread adoption of an application. This distinction is precisely what helps you separate marketing promises from genuine progress.
Understanding the development of AI through its methods
The development of AI is not a simple chain reaction. Rule-based and machine learning methods coexist and can be combined. Therefore, for your company, what matters is the suitability for a task, not just the age of a method.
Symbolic AI and expert systems
Symbolic AI works with explicitly represented knowledge, symbols, and rules. A rule-based expert system combines a knowledge base with a process that derives conclusions from it. The advantage lies in its transparent rules. However, expert knowledge must be captured and maintained; unaccounted-for exceptions can lead to incorrect results. Not every if-then automation is an expert system.
Machine learning, neural networks and deep learning
Machine learning refers to methods in which models are adapted based on data, rather than defining each decision rule individually. An example is the automatic categorization of texts, such as requests for proposals and complaints.
Neural networks are a family of models within machine learning. Deep learning utilizes neural networks with multiple processing layers. Crucially, performance on training data is not the only factor; reliability with new, representative data is essential.
Generative AI, Transformers and large language models
Generative AI generates content such as text, images, audio, or program code based on learned patterns. The term describes a function, not a single model architecture. Transformers are an architecture of neural networks: their attention mechanisms weigh relationships between components of an input. Many large language models use transformers and generate text step by step based on possible next text components, so-called tokens.
Even at this stage in the history of AI, the following applies: linguistic fluency is no proof of factual accuracy. Convincingly formulated answers can contain false information or fabricated sources.
Anyone wishing to delve deeper into the fundamental question of which tasks machines should and should not perform will find the appropriate classification in our article on the philosophy of AI.
What the emergence of AI means for your company
At Berger+Team, we combine brand strategy, web development, marketing, and technical implementation as a collective of freelancers. My benchmark when working with SMEs: AI must improve a specific task, support your team, and respect your responsibility to your customers – creating a true win-win-win situation.
AI can support humans. The responsibility for its use remains with humans.
A practical selection rule can be derived from the history of artificial intelligence: Test performance in the intended application, instead of inferring general reliability from a single demonstration. Five questions will help you with this:
- Which task should work better? "Pre-sorting inquiries by topic" is verifiable; "making our company intelligent" is not.
- Which method is sufficient? For fixed specifications, classic automation may suffice. Variable texts may require a learning model.
- Which data may be used? Check data quality, data protection, access rights and confidentiality before submitting it to a service.
- How do you recognize the benefit? Compare processing time, errors, and rework with the previous process.
- Who reviews and decides? Define which results require human approval.
A hypothetical example : A language model prepares answers to customer inquiries instead of sending them out unchecked. Your team checks facts, commitments, and tone. Only a comparison of total time spent and answer quality reveals whether its use is worthwhile.
For implementation, you'll find guidance on the first AI use case in your company . If you'd like support, we can assist you with our AI strategy consulting and the integration of AI tools into existing systems. The benefits should not only include lower costs, but also reduced workload for your team and reliable quality for your customers.
Frequently asked questions about the history and application of AI
When did the history of artificial intelligence begin?
As an independent field of research, AI is usually associated with Dartmouth in 1956. Its mathematical and philosophical foundations, however, reach back to antiquity. Alan Turing's 1950 publication and his Turing machine from 1936 are important early reference points; there is no single inventor of AI.
Is artificial intelligence the same as machine learning?
No. AI is the overarching term ; machine learning refers to data-driven learning methods within this field. Symbolic methods and expert systems, on the other hand, can work with explicitly predefined knowledge.
Does generative AI make rule-based systems obsolete?
No. You can, for example, generate text drafts and then apply fixed validation rules for mandatory information. Such rules complement generative methods, but do not replace full professional review.
Does success in chess or Go prove general intelligence?
No. High performance in a limited task initially only demonstrates that specialized skill. Therefore, when choosing your tools, what matters is performance in your specific application, not success under other conditions.
Can you trust a large language model as a source of knowledge?
Use language models as a tool, but verify important statements using appropriate original sources. A plausible-sounding answer is not a verified fact. Prices, contracts, technical specifications, and legal matters, in particular, require clear approval guidelines.
How do you maintain responsibility when using AI?
Define permissible tasks, allowed data, auditing obligations, and responsible individuals. Internal AI guidelines for small teams can provide orientation . This ensures transparency regarding which decisions are made by humans and which tasks are handled by a system.
What was the AI Winter and when did it take place?
The term "AI winter" refers to periods of significant funding cuts in AI research, particularly in parts of the 1970s and from the late 1980s onward. The precise boundaries vary depending on the research area and region. Research never completely ceased – many current methods build upon work from these years.
Sources
- Turing, AM — “Computing Machinery and Intelligence,” Mind (1950)
- McCarthy, Minsky, Rochester, Shannon — Dartmouth Summer Research Project proposal, Stanford (1955)
- Vaswani et al. — “Attention Is All You Need,” arXiv (2017)
- OpenAI — Introducing ChatGPT, openai.com (2022)
