When people start learning machine learning, they often focus on different things. But in practice, the most important question is much simpler:
What kind of work do you want to do?
What problems do you want to solve using ML?
When you look at machine learning from this perspective, the entire field can be cleanly divided into five main work domains.
1. Predictive Modeling
This domain works primarily with structured, tabular, or numerical data. The objective is to predict an outcome based on historical patterns.
Examples:
- Helping businesses detect and prevent fraud and avoid financial losses
- Analyzing effectiveness of past promotional activity
- Forecasting patient admissions and readmissions
- Predicting energy consumption
- Regression, classification, and time-series forecasting all belong here. This is the oldest and most battle-tested area of machine learning, and where most practitioners build their foundation.
2. Perception AI (Vision / Audio)
In this domain, machines learn to interpret visual and auditory signals from the real world.
Examples:
- Counting people in public spaces using camera feeds
- Detecting cracks or defects in infrastructure from drone images
- Recognizing spoken commands in voice-controlled systems
- Identifying equipment malfunctions from vibration or sound patterns
- Identifying pedestrians and traffic signs in a self-driven car
- Models such as CNNs and Vision Transformers (ViT) are heavily used. Perception AI enables machines to convert raw sensory data into meaningful signals.
3. Language Intelligence (NLP)
This domain focuses on understanding and processing written or spoken language.
Examples:
- Extracting key clauses from contracts and legal documents
- Automatically generating meeting notes from transcripts
- Detecting abusive content in online communities
- Grouping news articles by topic at scale
- Speech recognition, Spell check, autocomplete etc
- Transformer-based models like BERT and GPT-style architectures power this domain. The aim is to help machines understand nuance, context, and intent in language.
4. Generative AI (Multimodal)
This domain is about creation, not just prediction. Models here can generate entirely new outputs across multiple data types.
Examples:
- Creating a short story based on the style of a particular author,
- Generate a realistic image of a person who doesn't exist
- Composing a symphony in the style of a famous composer
- Generating test cases or documentation from source code
- Large Language Models (LLMs), diffusion models, and multimodal systems define this space, where language, vision, and reasoning intersect.
5. Decision & Control Systems (Reinforcement Learning)
What kind of work this is: Learning what action to take in an environment to maximize long-term reward.
Examples:
- Dynamic pricing and bidding systems
- Recommendation systems with long-term user engagement
- Robotics control and motion planning
- Resource allocation and scheduling
Agentic AI: The Extension
Agentic AI is an extension, not a standalone ML domain. It builds on Generative AI to enable systems that can reason, plan, and act.
Examples:
- An AI assistant that schedules meetings by checking calendars and availability
- A system that monitors metrics, detects anomalies, and triggers alerts
- An AI agent that runs experiments, compares results, and selects the best approach
- Multiple agents collaborating to complete a complex workflow
- In short, Agentic AI represents AI systems that don’t just respond but take action.
Top comments (0)