Code & Consciousness

The Definitive History of AI: Part 2 - The Golden Years (1956-1974)

Welcome to the second installment of our seven-part series on the history of Artificial Intelligence. Today, we'll delve into the period from 1956 to 1974, often referred to as the "Golden Years" of AI.

The Dartmouth Conference: A Closer Look

The Dartmouth Conference of 1956, which we briefly mentioned in Part 1, deserves a more detailed examination. This eight-week gathering, organized by John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon, marked the official birth of AI as a field of study.

The conference brought together researchers from various disciplines, including computer science, psychology, and mathematics. While the attendees had diverse views on how to approach the creation of artificial intelligence, they shared a common belief in the possibility of creating machines that could think.

Key outcomes of the Dartmouth Conference included:

• The coining of the term "Artificial Intelligence" by John McCarthy
• The establishment of AI as a distinct field of research
• The formation of a community of AI researchers
• The setting of an ambitious research agenda for AI

Early AI Programs and Achievements

The period following the Dartmouth Conference saw rapid development in AI, with several groundbreaking programs demonstrating the potential of the field:

1. General Problem Solver (GPS) (1957): Developed by Allen Newell and Herbert Simon, GPS was designed to work as a universal problem-solving machine. While it was successful in solving simple problems, it struggled with more complex tasks, highlighting the challenges of creating general intelligence.
2. ELIZA (1966): Created by Joseph Weizenbaum at MIT, ELIZA was one of the first chatbots. It could engage in dialogue by using pattern matching and substitution methodology. While ELIZA could convincingly simulate a psychotherapist in conversation, it had no true understanding of the dialogue.
3. STUDENT (1967): Developed by Daniel Bobrow for his PhD dissertation, STUDENT could solve high school algebra word problems. It was one of the first demonstrations of natural language processing in AI.
4. SHRDLU (1970): Created by Terry Winograd, SHRDLU demonstrated natural language understanding in a limited block world. Users could interact with the program using natural language to move blocks around, and SHRDLU could answer questions about the state of the world.
5. MYCIN (1972): Developed at Stanford by Edward Shortliffe, MYCIN was an early expert system that could diagnose blood infections and recommend antibiotics, with a performance level comparable to human experts.

Key Areas of AI Research

During this period, several key areas of AI research emerged:

• Natural Language Processing: Researchers worked on enabling computers to understand and generate human language, as demonstrated by programs like ELIZA and SHRDLU.
• Knowledge Representation: This involved finding ways to represent real-world information in a form that a computer could use to solve complex problems.
• Problem Solving: Programs like GPS focused on developing general methods for solving a wide range of problems.
• Machine Learning: Early work began on enabling computers to learn from experience, although significant progress in this area would come later.
• Robotics: The first digital and programmable robot, Unimate, was installed in 1961 to lift hot pieces of metal from a die casting machine and stack them.

The Optimism of the Golden Years

The rapid progress made during this period led to a wave of optimism about the potential of AI. Researchers made bold predictions about the future of AI:

• Herbert Simon predicted in 1965 that "machines will be capable, within twenty years, of doing any work a man can do."
• Marvin Minsky stated in 1967 that "within a generation... the problem of creating 'artificial intelligence' will substantially be solved."

This optimism helped secure significant funding for AI research, particularly from the Defense Advanced Research Projects Agency (DARPA) in the United States. The field attracted many talented researchers and students, further accelerating progress.

Challenges and Limitations

Despite the significant progress and optimism, researchers began to encounter challenges that hinted at the complexity of creating true artificial intelligence:

• Intractability: Many of the problems AI was trying to solve proved to be computationally intractable when scaled up to real-world complexity.
• The Frame Problem: First formulated by John McCarthy and Patrick J. Hayes in 1969, this problem highlighted the difficulty of representing the effects of actions in logic without having to represent explicitly a large number of intuitively obvious non-effects.
• Brittleness: AI systems of this era often performed well on specific problems they were designed for but failed when presented with slightly different scenarios.
• Common Sense Reasoning: Giving machines the kind of common sense knowledge that humans use to navigate the world proved much more difficult than initially anticipated.

Conclusion

The period from 1956 to 1974 was truly a golden age for AI. It saw the birth of the field, rapid progress in various areas of AI research, and a wave of optimism about the potential of artificial intelligence. The achievements of this era laid the groundwork for many aspects of modern AI.

However, the challenges encountered during this period also foreshadowed the difficulties that would lead to the first "AI winter" in the following years. The realization of the true complexity of creating human-like intelligence was just beginning to dawn on the AI community.

Coming Up Next

Join us tomorrow for Part 3 of "The Definitive History of AI" series, where we'll explore "The First AI Winter (1974-1980)". We'll delve into the challenges, limitations, and reduced funding that characterized this difficult period in AI development.