Evolución de la IA - (4/7)

Evolution of Artificial Intelligence Concepts

Introduction to Knowledge Engineering

• María Florencia introduces the video, discussing the evolution of artificial intelligence (AI) concepts and the significance of knowledge engineering.

• The previous video covered systems based on knowledge and expert systems, emphasizing their importance in AI development.

Emergence of Backpropagation Algorithm

• Following an initial period of stagnation in AI, new architectural proposals emerged, including backpropagation.

• Rummelhart, Hinton, and Williams introduced the backpropagation algorithm for training multilayer artificial neural networks with hidden layers.

• This algorithm calculates gradient descent concerning weights across neuron connections by comparing desired outputs with actual outputs.

Mechanism of Backpropagation

• The unique aspect of backpropagation is its weight updates occurring from output to input, considering each neuron's impact on subsequent calculations.

• Backpropagation is utilized in various neural network architectures for tasks like pattern recognition and image segmentation.

Recurrent Neural Networks (RNN)

• In 1986, Michael Jordan introduced recurrent neural networks (RNN), which utilize supervised learning for sequential data analysis.

• RNN are particularly effective for deep learning applications such as video analysis and natural language processing due to their ability to handle sequential dependencies.

Challenges Faced by AI Systems

• Despite advancements, a second winter in AI occurred when ambitious projects failed to meet expectations regarding conversational capabilities.

• Funding cuts followed this realization; however, research continued at a reduced scale from 1987 to 1993.

Critique of Expert Systems

• Researchers faced challenges reconciling expectations with the actual capabilities of expert systems.

• John McCarthy criticized these systems for lacking common sense and self-awareness regarding their limitations.

Milestones in Chess AI: Deep Blue

• A significant milestone was achieved when Deep Blue defeated chess champion Garry Kasparov in 1996 after extensive development over three years.

• This victory marked a pivotal moment where software outperformed human expertise in chess strategy.

Improvements Post-Kasparov Match

• Following its match against Kasparov, improvements were made to Deep Blue's architecture leading up to a rematch in 1997.

Evolution of Robotics and AI

Advancements in AI and Robotics

• The analysis of moves attacking pieces in chess reflects the evolution of AI, showcasing improvements in efficiency and speed.

• Interactive robotic pets emerge, indicating a shift towards more complex behavioral interactions; MIT researchers develop robots capable of expressing emotions.

• Autonomous vacuum cleaners represent significant advancements, with features like obstacle avoidance and self-mapping capabilities.

Impact of Deep Learning and Big Data

• The 21st century sees exponential growth in computing power and storage, enabling neural networks to achieve remarkable results through massive datasets.

• Cloud computing becomes essential for handling big data, highlighting the interconnectedness of various digital transformation technologies.

DARPA Challenges and Autonomous Vehicles

• In 2004-2005, DARPA launches a challenge offering prizes for functional autonomous vehicles, igniting rapid development in this field.

• Sebastian Thrun leads the development of "Stanley," an autonomous vehicle that wins DARPA's Grand Challenge; he later joins Google to further advance self-driving technology.

NASA's Use of Autonomous Systems

• NASA employs autonomous systems for Mars exploration vehicles, capturing public interest and emphasizing robotics' growing prominence.

Asimov's Three Laws of Robotics

• Isaac Asimov’s three laws outline ethical guidelines for robot behavior:

• A robot must not harm humans or allow them to come to harm through inaction.

• Robots must obey human orders unless they conflict with the first law.

• Robots should protect their own existence as long as it does not conflict with the first two laws.