Game AI - Introduction to Artificial Intelligence

Artificial Intelligence

Lecture 01 - Introduction

Edirlei Soares de Lima

<edirlei.slima@gmail.com>

What is Artificial Intelligence?

•

Artificial intelligence is about making computers able to

perform the thinking tasks that humans and animals are

capable of.

o Computers are very good at:

arithmetic, sorting, searching, play

some board games better than

humans, ...

o Computers are not very good at:

recognizing familiar faces, speaking

our own language, deciding what to

do next, being creative, ...

What is Artificial Intelligence?

•

AI researchers are motivated by:

–

Philosophy: understanding the nature of thought and the nature of

intelligence and building software to model how thinking might work.

Psychology: understanding the mechanics of the human brain and

mental processes.

Engineering: building algorithms to perform human-like tasks.

Academic AI vs Game AI:

–

Academic AI: solve problems optimally, less emphasis on hardware or

time limitations;

–

Game AI: entertain player, have to work with limited time and

hardware resources.

Academic AI – History

•

Early Days (time before computers):

–

Philosophical questions:

•

What produces thought?

•

Could we give life to an inanimate object?

–

First programmable computers (1940s): war simulation, break enemy

codes, …

Symbolic Era (1950s – 1980s):

–

Symbolic systems: set of knowledge (symbols) + reasoning algorithm;

Expert systems: large database of knowledge + expert rules;

Trade-off: when solving a problem, the more knowledge you have, the

less work you need to do in reasoning.

Academic AI – History

•

Modern Era:

–

Increasing frustration with symbolic approaches (scalability problem);

–

Move towards natural computing (inspired by biology or other natural

systems):

•

Neural networks (first suggested in 1943);

•

Genetic algorithms.

–

Key ingredient: ability to handle uncertainty.

Current research:

•

Machine learning;

Big data;

Deep learning;

Current AI Advancements

•

Google & Uber Driverless Cars

Personal Assistants

Autonomous Robots

Game AI – History

•

Pac-Man (1979):

–

Very simple AI technique (finite state machine);

–

Semi-random decisions;

scatter_time >= 5 (sec)

Chase

Scatter

chase_time >= 20 (sec)

player_energy

time >= 10

sec)

player_got

_energy ==

true

_

(

Frightened

Game AI – History

•

Goldeneye 007 (1997):

–

Sense simulation system: characters could

see their colleagues and would notice if

they were killed;

–

Still relying on finite state machines with a

small number of well-defined states;

–

Sense simulation was the topic of the

moment:

•

Metal Gear Solid (1998);

•

Thief: The Dark Project (1998);

Game AI – History

•

Warcraft (1994):

–

One of the first times pathfinding

was widely noticed in action;

Warhammer: Dark Omen (1998):

–

Robust formation motion;

–

Emotional models of soldiers;

Game AI – History

•

Creatures (1996) & Black and White (2001):

–

The first time neural networks were used in

a game;

The neural network-based brain of each

creature allowed them to learn what to do;

Made AI the selling point of the game;

Game AI – Model

Complexity Fallacy

•

It is a common mistake to think that complex AI equals better

character behavior.

When simple things look good: Pac-Man

–

Semi-randomly decisions at junctions;

–

Player comments:

•

“To give the game some tension, some clever AI was programmed into the game.

The ghosts would group up, attack the player, then disperse. Each ghost had its own

AI.”

“The four of them are programmed to set a trap, with Blinky leading the player into

an ambush where the other three lie in wait.”

Complexity Fallacy

•

It is a common mistake to think that complex AI equals better

character behavior.

When complex things look bad: Black and White [2001]

–

Neural Networks and Decision Trees allowed creatures to learn.

–

When many people first play the game, they often end up

inadvertently teaching the creature bad habits, and it ends up being

unable to carry out even the most basic actions.

Perception Window

•

Most players will only come across some characters and

enemies for a short time, which might not be enough for the

player to understand the AI.

–

Make sure that a character’s AI matches its purpose in the game and

the attention it will get from the player.

–

A change in behavior is far more noticeable than the behavior itself.

Illusion of Intelligence

•

“If it looks like a fish and smells like a fish, it’s probably a fish.”

–

if the player believes an agent is intelligent, then it is intelligent.

For game AI the nature of the human mind is not the key

point.

–

The AI characters must look right and demonstrate intelligent

behavior.

•

Sometimes, simple solutions are enough to create a good

illusion of intelligence.

–

Halo [2001] – increasing the number of hit points required to kill

enemies made testers thought the AI was very intelligent.

Illusion of Intelligence

•

Player’s perception of intelligence can be enhanced by

providing visual and/or auditory clues about what the agent is

“thinking”.

Animation is an excellent way to create a good illusion of

intelligence.

–

The Sims [2000] – although it uses a complex emotional model for

characters, most part the characters’ behaviors is communicated with

animations.

–

Triggering animations at the right moment is the key point.

Illusion of Intelligence

•

The goal of game developers is to design agents that provide

the illusion of intelligence, nothing more.

Game developers rarely create great new algorithms and then

ask themselves, “So what can I do with this?”

–

Instead, they start with a design for a character and apply the most

relevant tool to get the result.

•

Be careful to never break the illusion of intelligence:

–

Running into walls, getting stuck in corners, not reacting to obvious

stimulus, seeing through walls, hearing a pin drop at 500 meters, …

Most Common Techniques

•

Pathfinding

Steering behaviours

Finite state machines

Automated planning

Behaviour trees

Randomness

Sensor systems

Machine learning

Most Common Techniques

•

Pathfinding

Steering behaviours

Finite state machines

Automated planning

Behaviour trees

Randomness

Sensor systems

Machine learning