History Of AI Advancement

Joseph Capek first wrote about automats in 1917 in his short story Opilec. His brother Karel Capek wrote the play Rossum’s Universal Robots (RUR) in 1921.

Foundational Era：

Turing Test

Artificial Intelligence is a term that has been around for decades. In 1950, the fundamental concepts of AI began to take shape. Alan Turing published Computing Machinery and Intelligence in 1950 and discussed how to build intelligent machines and how to test their intelligence. The human evaluator needs to know which one is the machine to test the human candidates and the machine with a natural language conversation. This idea was groundbreaking because it challenged the old view that machines could only perform mechanical procedures. Even today, we use the Turning test method to determine if a computer can think like a human.

Development of Neural Network Models

McCulloch and Pitts were the first to use Alan Turing’s notion of computation to understand neural activity. In 1943, Warren McCulloch and Walter Pitts proposed one of the earliest neural network models, the McCulloch-Pitts neuron. This model was inspired by the biological neurons found in the brain. It received binary inputs, processed them using a weighted sum mechanism, and produced a binary output based on a predefined threshold function. This early neural network model first posits that neurons are equivalent to logic gates and that neural networks are digital circuits. The McCulloch-Pitts neuron was a simplified mathematical model that laid the groundwork for future advancements in artificial neural networks.

Implementation of Learning Algorithms

At the same time, Canadian psychologist Donald Hebb proposed Hebbian learning. This neuropsychological theory claims neural connections are strengthened when two neurons are activated simultaneously. This idea provided a crucial theoretical framework for learning algorithms in neural network research. It is used for pattern classification and enables machines to adapt and learn from input data. The input layer can have many units, and the output layer only has one unit. The Hebbian learning rule algorithm updates the weights between neurons in the neural network for each training sample. It’s used widely for machines to adapt and learn from input data.

Early Applications Era:

Advancements in Learning and Adaptation

Allen Newell, Herbert A. Simon, and J.C. Shaw developed the computer program General Problem Solver in 1957. GPS computer programs can solve a wide range of problems in various domains. Unlike neural network models, GPS uses symbolic reasoning and logical inference to solve complex problems. It can solve logic puzzles, mathematical problems, and symbolic reasoning tasks. The system compares the current state of the problem to the desired goal state and selects actions that reduce the difference between the two. It was implemented in the third-order programming language, IPL.

Natural language processing program development

Joseph Weizenbaum developed an early natural language processing computer program, ELIZA, from 1964 to 1967. ELIZA was one of the first chatterbots and programs capable of attempting the Turing test. It examined the text for keywords, applied values to them, and transformed the input into an output. However, it wasn’t considered understood the same way humans do.

Terry Winograd also developed SHRDLU from 1968 to 1970 at MIT. It was an interactive program designed to demonstrate natural language understanding and interaction with a computer. In the program, the user can instruct SHRDLU to move various objects around in the “blocks world.” SHRDLU also includes a basic memory to supply context. It can search back through the interactions to find the proper context.

While SHRDLU focused on understanding and executing commands in a specific domain, ELIZA aimed to simulate conversation by responding with predefined patterns. Despite their different approaches to natural language processing, both programs contributed to early AI research.

Autonomous robotics

In the late 1960s, along with symbolic AI and expert systems, the field of robotics progressed. Stanford Research Institute developed the first general-purpose mobile robot, Shakey the Robot. Shakey represented a pioneering achievement in autonomous navigating and planning. It demonstrated the potential of AI technologies to enable robots to perceive and interact with their environment in real-world settings.

Shakey was equipped with various sensors, including cameras, range finders, and bump sensors, which provided detailed information about its surroundings. This sensory data formed the basis for Shakey’s internal model of the environment, including the locations of objects and obstacles. Using this model, Shakey employed simulation to predict future states and evaluate potential actions to plan and execute tasks autonomously.

Shakey’s control system was organized hierarchically. It could analyze commands and break them down into basic chunks by itself. This hierarchical architecture allowed Shakey to integrate sensory information, reason about its environment, and generate high-level plans autonomously. Its control system operates in a closed-loop fashion, continuously sensing its environment, updating its internal state, and adjusting its actions in response to environmental changes. This adaptive control mechanism allowed Shakey to navigate dynamic environments and respond effectively to unforeseen circumstances. Shakey’s programming breakthrough laid the foundation for further research in robotics and AI.

Expert systems

In the 1960s, the concept of expert systems began to emerge. It was formally introduced around 1965 by Edward Feigenbaum. Expert systems would mimic the decision-making process of a human expert. The program would learn how to respond in any situation from an expert in a field. Once the program has learned about every situation, it can advise others. Dendral was one of the earliest expert systems. It was designed to analyze chemical compounds. It utilized a knowledge base of chemical and heuristic rules to analyze data and generate hypotheses about molecular structures.

The Modern Era：

Neural networks

In the 1980s, several advancements in neural networks and connectionism laid the foundation for developing modern deep-learning techniques. David Rumelhart and James L. McClelland introduced the concept of Parallel Distributed Processing models (PDP) in their book “Parallel Distributed Processing.” The theory assumes the mind comprises many units connected in a neural network. Mental processes are interactions between those units that excite and inhibit each other in parallel rather than sequential operations. This explained how a network’s interconnected nodes, or neurons, could simulate human cognition. McClelland’s research has focused on understanding human cognition through computational modeling, particularly memory, learning, and language processing.

“Godfather of AI” Geoffrey Hinton has also co-authored several influential papers that helped revive interest in neural networks. His research significantly contributed to advancements in diverse areas, such as image recognition and natural language processing. This progress paved the way for a neural network-based AI system tackling complex real-world challenges, including tasks like image recognition, natural language understanding, speech recognition, and even autonomous driving.

Machine learning

In the 1990s, several Machine Learning algorithms improved artificial intelligence capabilities in various domains. One significant advancement was refining decision tree algorithms, a supervised learning method. Decision trees enable AI systems to make complex decisions based on input data. They were widely applied in healthcare diagnostics, financial analysis, and customer relationship management.

Support Vector Machines (SVMs) emerged as another important machine learning during this period. They can handle high-dimensional data and nonlinear relationships and are widely used in pattern recognition and image classification. They can efficiently classify data points based on their features and revolutionized fields. It acts as computer vision to identify and categorize images accurately.

Probabilistic graphical molders also gained prominence in the 1990s. They were powerful tools for representing and reasoning under uncertainty. These models provided a framework for AI systems to make decisions in situations where the outcome is uncertain or probabilistic. By incorporating probabilistic reasoning, AI systems could better assess risk, make more informed decisions, and navigate uncertain environments more effectively.

Advancements in reinforcement learning algorithms also marked a significant breakthrough in the 1990s. Reinforcement learning techniques improved game playing, robotics control systems, and autonomous vehicle control. By learning from trial-and-error interactions with the environment, AI agents could optimize their behavior to achieve specific goals, leading to more efficient and adaptive systems.

Additionally, knowledge representation and reasoning continued developing in formalisms. These formalisms provided ways to represent and reason about structured knowledge. They enabled intelligent systems to understand and reason about complex information, facilitating the development of knowledge-based applications in different domains, such as expert systems, natural language understanding, and intelligent tutoring systems.

Famous AI achievement

After decades of research and development, in 1997, IBM’s Deep Blue defeated world chess champion Garry Kasparov. Deep Blue was a chess-playing expert system. The development began in 1985. Chess was believed to be exclusive to human intelligence. Computer scientists believed that playing chess was a good measurement of the effectiveness of artificial intelligence. This event demonstrated that AI’s sophisticated evaluation functions, brute-force search algorithms, parallel processing, endgame databases, and opening book strategies can compete at the game’s highest levels.

The Current Era:

In the 21st century, deep learning algorithms, especially convolutional neural networks, have surpassed human-level accuracy in tasks such as image recognition. These advancements have improved the accuracy of medical image analysis and revolutionized fields like healthcare. AI-powered systems assist doctors in diagnosing diseases and interpreting medical scans with unprecedented precision.

Natural language processing (NLP) has also seen significant progress. AIs can perform complex tasks such as answering questions, translating languages, and classifying text according to sentiment analysis. This has led to developing virtual assistants and language translation tools that bridge diverse language and cultural communication.

Moreover, AI technologies, including machine learning and deep learning, are increasingly used in drug discovery and development. They can predict molecular properties and identify potential drug candidates. This process has optimized drug designs. AI techniques are also applied to address global challenges such as climate change, environmental monitoring, and sustainable development. AI-powered systems analyze vast amounts of data to model climate patterns, monitor biodiversity, and optimize resource management. If we use those technologies wisely, they can help mitigate environmental impact and promote sustainability.

While self-driving cars like Waymo have been successfully tested in San Francisco and Phoenix, scientists continue optimizing various areas, such as sensor fusion and perception with redundant safety sensor systems, real-time decision-making and planning, highly accurate maps, precise localization, etc.

Final Thoughts

AI advancements can contribute to positive social outcomes if we prioritize ethical considerations. We all should engage in this process, advocating for inclusive and representative data collection for AI training. With transparent and unbiased models, AI systems can process large-scale data analysis to identify and address healthcare access, education, and resource allocation disparities. The technology leap can be used to advance our social and physical environment.

However, we also need to consider potential risks and unintended consequences carefully. With proper oversight and collaboration between AI experts and the broader community, we can continue to harness AI’s power for good. Let’s work together towards a future where AI creates a better world.