“Agentic AI” refers to artificial intelligence systems that act with agency – meaning they can pursue goals, make decisions, and take actions autonomously in complex environments. These systems are not just passive tools responding to commands; they exhibit behavior that seems purposeful or intentional.
Key Characteristics of Agentic AI:
- Goal-oriented behavior: They pursue specific objectives, often over extended time frames.
- Autonomy: They operate with a degree of independence, making decisions without constant human oversight.
- Planning and reasoning: They can plan ahead, consider consequences, and adapt strategies.
- Interactivity: They interact with their environment (including humans) to gather information or achieve goals.
Possible Examples:
- An AI that manages a company’s operations with minimal input.
- A personal assistant that proactively schedules appointments, books flights, or even negotiates deals.
- AI agents in video games or simulations that behave like intelligent characters.
Example of Personal AI Assistant Who Prepares Breakfast
| 1. Ask about person’s choices for breakfast 2. Check the refrigerator for the ingredients. 3. If not all ingredients, prepare a list and send for order online. 3(a). If the order is arrived update stock list 4. Cook the breakfast 5. Serve the breakfast and call the person |
Agentic Algorithm: Smart Breakfast Assistant
START
1. ASK_USER_FOR_BREAKFAST_CHOICE()
→ breakfast_choice ← get_user_input()
2. CHECK_REFRIGERATOR_FOR_INGREDIENTS(breakfast_choice)
→ available_ingredients ← scan_fridge()
→ required_ingredients ← get_ingredients_for(breakfast_choice)
→ missing_ingredients ← required_ingredients – available_ingredients
3. IF missing_ingredients is NOT EMPTY THEN
CREATE_ORDER_LIST(missing_ingredients)
→ place_order(missing_ingredients)
WAIT_FOR_ORDER_DELIVERY()
ON_DELIVERY_ARRIVAL:
UPDATE_STOCK_LIST(missing_ingredients)
4. COOK_BREAKFAST(breakfast_choice)
→ follow_recipe(breakfast_choice)
→ start_cooking()
→ monitor_process()
5. SERVE_BREAKFAST()
→ plate_food()
→ place_on_table()
6. CALL_USER_TO_EAT()
→ notify_user(“Your breakfast is ready!”)
END
Agentic AI is a step closer to general intelligence. It raises exciting possibilities like fully autonomous scientific research agents.
There are major safety and alignment concerns also. If AI systems are acting independently, how do we ensure their goals align with ours?