MIT Sloan: Intro to Machine Learning (in 360/VR)

The main types are supervised learning (requires human-labeled data), semi-supervised learning (uses a mix of labeled and unlabeled data), and unsupervised learning (learns patterns from unlabeled data without human guidance). Reinforcement learning is also mentioned, where agents learn through trial and error.

Neural networks are inspired by the human brain's structure. They consist of interconnected simple computational units (neurons) that process inputs through weights and biases. Their power comes from stacking many layers (deep learning), enabling them to learn complex representations and functions from data.

Deep learning is a subset of machine learning focused on representation learning. Unlike traditional methods that require manually encoded features, deep learning models automatically learn complex features directly from raw data, enabling them to handle massive datasets and discover intricate patterns.

Key challenges include the need for vast amounts of labeled data, the susceptibility of algorithms to noise and adversarial attacks (robustness issues), the difficulty of achieving true reasoning beyond pattern recognition, and the limitations in robotics related to perception and manipulation.

Computer vision faces challenges like viewpoint variation, scale changes, complex lighting, occlusions, and the sheer difficulty of interpreting raw pixel data. Humans overcome these by bringing prior knowledge, which AI systems currently lack, and they can be easily fooled by subtle changes or adversarial examples.

Advancements in compute power, particularly through parallel processing with GPUs and specialized hardware like TPUs, have been crucial for the recent breakthroughs in machine learning and deep learning, enabling the training of complex models on large datasets.

The global community of programmers and engineers, facilitated by platforms like GitHub for code sharing, plays a vital role. Collaboration, open-source development, and competitions accelerate innovation by allowing rapid iteration and shared problem-solving.

Ethical dilemmas arise when systems must make trade-offs, such as choosing between different unavoidable accident outcomes (the 'trolley problem'). Designing objective functions that balance safety, efficiency, and ethical considerations is a major challenge for engineers.

While machine learning excels at specific tasks and can exhibit aspects of general intelligence (like in the Pong game example), achieving human-level general intelligence, including complex reasoning, imagination, and subjective experience, remains an open and significant challenge.

Current machine learning models are often criticized for merely memorizing patterns from training data rather than truly understanding the underlying concepts. A key research goal is to move beyond memorization to genuine understanding, enabling systems to reason and generalize more effectively.

Robotics faces immense challenges in actuation, localization in dynamic environments, dealing with uncertainty and noise, and manipulating objects. These 'mundane' human tasks require a complex interplay of perception, reasoning, and physical interaction that is difficult to replicate with current AI.