π AI Overview
Artificial intelligence (AI) is a branch of computer science that deals with the creation of intelligent agents, which are systems that can reason, learn, and act autonomously. AI has been around for decades, but it has only recently begun to make significant progress. This is due in part to the development of new machine learning algorithms, which are able to learn from data without being explicitly programmed.
What is Artificial Intelligence?
- AI is about machines acquiring and applying knowledge and skills β much like humans do.
- AI systems are engineered, not naturally occurring like human intelligence.
- AI is also the name of the field that studies how to create such systems.
The AI Effect
- Our understanding of what constitutes AI changes over time.
- Technology that is new and complex tends to be labeled as AI, but things we grow familiar with often lose that label.
AI in Pop Culture
- Movies often portray AI as humanoid robots, sometimes with negative intentions.
- This is not an accurate representation; much of AI exists in networked computers.
AI Agents
- AI can be implemented through intelligent agents, which are either:
- Robots (with a physical body)
- Software agents (limited to computers)
AI in Everyday Life
- AI is used in smartphones, search engines, Mars rovers, and even beating humans at games like Go and chess.
- AI has the potential to revolutionize many aspects of our lives. For example, it is already being used to develop self-driving cars, medical diagnosis systems, and personalized education tools.
- There are also concerns about the potential risks of AI, such as job displacement and the development of autonomous weapons.
- Today's AI is 'narrow' or 'weak' β itβs good at specific tasks, but can't generalize (like humans can).
Types of AI
- Narrow/Weak AI: Current AI is specialized to one task (playing chess, recommending songs, etc.)
- General/Strong AI: Future AI is theorized to think and reason like a human across many areas, even without previous training.
The Future of AI
- Experts predict 'general' or 'strong' AI (which can reason like humans across many areas) may happen in the coming decades.
- There's concern that surpassing human intelligence could lead to AI becoming very powerful and potentially dangerous. The point where AI could rapidly self-improve is called the 'technological singularity'.
AI Technologies
- Machine Learning A dominant approach where algorithms build models from data to make predictions without being explicitly programmed.
- Other AI technologies exist, like fuzzy logic, search algorithms, and reasoning techniques.
Here are the notes from the provided text, focused on key concepts and terms:
Developing AI Systems
- AI Development vs. Traditional Software: AI development focuses on automatic generation of machine learning models from training data, rather than manual coding.
- The Machine Learning Workflow:
- Gather requirements
- Collect training data
- Utilize a machine learning algorithm
- Train the model
- Deploy the generated model
- Data Scientists: AI specialists similar to traditional developers, using AI-specific frameworks.
AI Development Frameworks
- Open-Source Frameworks: Apache Net, CNTK, TensorFlow, Keras, PyTorch, Scikit-learn
- Commercial Frameworks: IBM Watson Studio
Machine Learning Hardware
- Training vs. Prediction: Hardware needs differ between training models (resource intensive) and running models.
- Processor Characteristics:
- AI systems often make approximate decisions, reducing the need for high-precision arithmetic.
- AI is data-centric, so processors optimized for large data structures are beneficial.
- Deep neural nets benefit from parallel processing, making GPUs suitable.
- Hardware Options:
- General-purpose CPUs: Good for learning AI basics and small models.
- GPUs: Ideal for many machine learning applications.
- AI-specific Processors: Optimized for large-scale, resource-intensive AI tasks.
- Neuromorphic Processors: Emerging architecture mimicking human brains.
AI as a Service (AIaaS)
- Outsourcing AI Development: Using pre-built AI services or access to specialized hardware.
- Benefits: Access to larger datasets, expertise, and cost-effectiveness compared to building in-house.
- Examples: Image recognition, translation services, chatbots.
- Considerations: AIaaS often has limited service level agreements and shouldn't be relied on for mission-critical systems.
AI Standardization
- Relevant Standards Bodies: ISO, IEC, ITU, national bodies, industry bodies
- AI in Non-Safety Systems:
- Few existing standards specifically for AI
- GDPR is relevant for AI systems handling personal data
- AI in Safety-Related Systems:
- Use of AI for safety-critical decisions is on the rise.
- ISO 26262 (for automotive systems) doesn't yet cover AI specifically.
- Efforts are underway to develop new AI-specific safety standards.