The document outlines the three main steps needed to take over the world with artificial intelligence: 1) Perception - using techniques like Gaussian filters and pattern recognition to understand the state of the world. 2) Planning - deciding on a course of action using searches and Markov decision processes. 3) Control - executing the plan using controllers like PID and dealing with randomness. It provides examples of how these steps are implemented in applications like self-driving cars, robot helicopters, and optimizing data center resource utilization to reduce costs.

1. How to take over the World
with Artificial Intelligence

2. Taking over the World
is not an easy problem
1. You have to figure out the state
of the world (maybe it is already
taken by someone)

3. Taking over the World
is not an easy problem
2. You have to decide what to do
with it
I need your boots clothes and motorcycle

4. Taking over the World
is not an easy problem
You forgot to say please
3. You have to act according to your
plan (while keeping in mind that
it may be wrong)

5. Taking over the World
is not an easy problem
1. Perception
2. Planning
3. Control

6. Taking over the World
is not an easy problem
if failure in any step…

7. 1. Perception
2. Planning
3. Control
World
Intelligent
agent

8. Perception
Problem statement: censors are inaccurate
Where am I?
What is around me?
Solution:
- Gaussian filter
- Particle filter
- Pattern Recognition
- others

9. Perception
Problem statement: censors are inaccurate
Google uses Simultaneous
Localization and Mapping (SLAM)
for solving perception problem in
self-driving cars

10. SLAM Demo

11. Planning
Problem statement: choose one action among
them all
What should I do to
achieve the goal?
Solution:
- Searches
- Markov Decision
Process (MDP)

12. A* Search Demo

13. Control
Problem statement: the result of an action
is random
How should I act to
follow the plan?
Solution:
- PID controller
- Markov Decision
Process (MDP)

14. PID Controller Demo

15. Flying a robotic helicopter

16. Flying a robotic helicopter

17. 1. Perception
2. Planning
3. Control
World
Intelligent
agent
- Gaussian filter
- Particle filter
- Pattern Recognition
- Searches
- Markov Decision
Process (MDP)
- PID controller
- Markov Decision
Process (MDP)

18. Markov Decision Process
State
1
State
2
State
3
The helicopter is
close to the ground
The helicopter is
crashed
Cost = $1000
The helicopter even
closer to the ground
Cost = $0
Go down
p = 10%
p = 90%
Go up
Do nothing

19. Stanford robotic helicopter

20. Making better data center
Application
VM VM VM VM VM
Load Balancer

21. Making better data center
Capacity
VM utilization

22. Making better data center
AI-based capacity manager saves up to 25% costs
comparing to threshold-based capacity managers

23. Making better data center
In US only:
 3 million data centers
 12 million machines
 30 percent of them are drawing power without actually doing anything
• $3.8 billion and 39 billion kilowatt-hours annually are wasted
• Wasted electricity is enough to power 3.5 million American homes
© National Geographic

24. Power Supply
Physical Hardware
Virtual Machines
QOS Security
AI Data Center Manager

25. 1. Perception
2. Planning
3. Control
World
Intelligent
agent
- Gaussian filter
- Particle filter
- Pattern Recognition
- Searches
- Markov Decision
Process (MDP)
- PID controller
- Markov Decision
Process (MDP)

26. You can do with AI

28. Thank you 