Deepdive in AI & AI agents

1. AI Types, AI Levels, AI Subsets
2. AI Models, Platforms, Frameworks
3. AI Workflow
4. AI Ethics: Responsible, Trustworthy, Compliant

---

1. AI types, AI Levels, AI Subsets

AI Types

AI types are classified by capability:

AI Levels

Below a re the five levels of AI, as defined by OpenAI to track progress toward AGI. These levels represent a progression from AI that can only chat (Level 1) to increasingly sophisticated systems that can reason, act, invent, and eventually perform the work of an entire organization (Level 5).

1	Conversational AI	AI systems capable of interacting with people through conversational language.	- Modern chatbots used for customer service
2	Reasoners	AI that can perform complex problem-solving tasks at a human-level, such as those a doctorate-level professional might handle.	- Analyzing medical data to suggest diagnoses - Get key insights from resumes
3	Agents	Systems that can take actions	- AI frameworks that allow multiple agents with specific roles to collaborate on complex tasks.
4	Innovators	AI that can generate new ideas and new inventions	- AI that can self-select goals and create a path to achieve them.
5	Organizations	AI that can perform the work of an entire organization	- An organization that operates with AI managing and executing the majority of its functions.

AI Subsets

AI -> ML -> DL -> Gen AI & LLM   AI families

AI	AI is a family made up of smaller, specialized parts called subsets of artificial intelligence: The learner, the talker, the creator, the watcher — and together they weave the magic you see in apps, games, hospitals, and even cars. - Problem-solving - Decision-making - Natural Language Processing - Perception Learn -> Analyse -> Predict
ML	Machine Learning is a subset of AI that focuses on training machine to solve a specific AI problem Data -> Stats and Math Algorithms -> Prediction	- Supervised Learning Models - Unsupervised Learning Models - Selfsuper Learning Models - Reinforcement Learning Models
DL	Deep Learning is a subset of ML that uses artificial neural networks with multiple layers (hence "deep") to model and understand complex patterns in large datasets: - Image recognition - Speech recognition - NLP Natural Language Processing: understands and processes human language - Game playing Data -> Neural Networks -> Predictions	- CNN Convulution Neural Networks
Gen AI	Generative AI is a branch of AI that focuses on creating models capable of analyzing and generating new content that resemble human-created content, such as: - Text / Code - Images - Videos - Music Gen AI uses LLM to invent. Large Language Model are very large deep learning models that are pre-trained on vast amounts of data. The underlying transformer is a set of neural networks that consist of an encoder and a decoder with self-attention capabilities Data -> Neural Networks (Transformers) -> New Content	- GAN Generative Adversarial Networks - Transformer models LLM: - BERT Bidirectional Encoder Representations from Transformers - GPT Genertive Pre-trained Transformers
Other Subsets	Robotics: AI driven machines performing physical tasks Genetic Algorithms: AI takes lessons from nature. Genetic Algorithms are modeled on evolution itself. NASA designed antennas with evolution that had a new type of shape and outperformed any existing design.

2. AI Models, Platforms, Frameworks

AI Models

Google Deepmind	World Model: - Genie can generate interactive environments Generative Models: - Gemini - Gemma - Veo - Imagen - Lyra Gemini model Ecosytem: - Gemini Robotics - Gemini Diffusion - Med-Gemini Experiments - Project Astra - Project Mariner - SynthID
Google AI	- Gemini - Nano Bana - NotebookLM

AI Platforms

	GCP VerteX AI
	Azure	- Azure Machine Learning studio is a managed web-based environment for building, training, deploying, and operating machine learning models end-to-end documentation. - Azure AI Foundry portal is a unified platform for developing and deploying generative AI apps and Azure AI APIs documentation.
	AWS	- SageMaker - Bedrock
	Hugging Face	The platform where the machine learning community collaborates on models, datasets, and applications.
	Arxiv	arXiv is a free distribution service and an open-access archive for nearly 2.4 million scholarly articles in the fields of physics, mathematics, computer science / AI, quantitative biology, quantitative finance, statistics, electrical engineering and systems science, and economics.

AI frameworks

	PyTorch	A popular deep learning framework developed by Facebook's AI Research lab, favored for its flexibility and Pythonic feel.
	TensorFlow	An open-source framework developed by Google for large-scale machine learning and deep learning, known for its flexibility and ability to scale from small to large projects.

3. AI workflow

AI workflow is the sequence of tasks and processes followed to develop, train, deploy, and maintain artificial intelligence models.

4. AI Ethics: Responsible, Trustworthy, Compliant

AI ethics focus on ensuring that artificial intelligence systems are developed and deployed in ways that are fair, transparent, and beneficial to society. This includes addressing issues such as bias in algorithms, privacy concerns, accountability for decisions made by AI systems, and the potential societal impact of AI technologies. Ai solutions are: Responsible, Trustworthy, Compliant

Responsible	- Human-Centered Design - Awareness - Data Understanding, Privacy & Security - Fairness & Non-Discrimination - Social & Environmental Responsibility - Repeatability & Testing - Accountability
Trustworthy	(Guidelines for Trustworthy AI - European Commission) - Lawful: respecting all applicable laws and regulations - Ethical: respecting ethical principles and values - Robust: both from a technical perspective while taking into account its social environment Key requirements that AI systems should meet in order to be deemed trustworthy. - Human agency and oversight - Technical Robustness and safety - Privacy and data governance - Transparency - Diversity, non-discrimination and fairness - Societal and environmental well-being - Accountability
Compliant	The AI Risk Categories - Unacceptable risk is the highest risk category - High-risk AI is likely to constitute the majority of AI systems currently in practice. - Low-risk AI systems do not directly pose harm onto individuals or society but still require transparency and accountability mechanisms to ensure they remain at this risk level. - Minimal risk is the lowest risk category