ELEMENTS OF AI

What is AI?
HOW SHOULD WE DEFINE AI?
- Three applications of AI that illustrate different aspects of AI:
+ Self-driving cars
+ Content recommendation
+ Image and video processing
- How it works?
+ Self-driving cars:
# A combination of many kinds of AI techniques: search and planning the
most convenient route, computer vision to identify obstacles, and decision-making to cope
with complex and dynamic environments  All of these things must work accurately to
avoid accidents.
# These technologies are also being used in other autonomous systems
such as delivery robots, autonomous ships, and flying drones
# Meanings: Road safety should be improved using systems surpassing
the human level. The efficiency of logistic chains when moving goods should be improved.
Humans move into the supervisory role and keep an eye on what’s going on while the
machine is driving  There are still some facets that humans haven’t even thought about it
yet.
+ Content recommendation:
# Information humans bumped into when using apps daily is
personalized.  The front-page of the printed version (newspaper) is the same for all
readers, but the front-page of the online version is different for each user  The algorithms
that determine the content humans see are based on AI.
# Meanings: Some companies don’t want to reveal the algorithms, but
some principals help understand the potential implications: filter bubbles, echo chambers,
troll factories, fake news, and new forms of propaganda.
+ Image and video processing:
# Face recognition is commonly used in applications  Some similar
techniques can be used to recognize other cars or obstacles around an autonomous car, or to
estimate wildlife populations.
# AI can be used to generate or alter visual content. Ex: Style transfer 
help adjust humans' photos to look like they were painted, and computer-generated
characters in motion pictures such as Avatar, … and animations where the animated
characters replicate gestures made by real human actors.
# Meanings: when techniques advanced and become widely available,
it’ll be easy to create fake natural-looking videos of an event, and difficult to distinguish
from real footage.
What is, and what isn’t AI? Not an easy question!
- The popularity of AI in the media is because people start using things that are
called by other names. Almost anything from statistics and business analytics to
manually encoded if-the rules called AI. Why is this so? Why is the public perception of
AI so nebulous?
+ Reason 1: no officially agreed definition
# The field is rather being constantly redefined when some topics are
classified as non-AI and new topics emerge
 # AI was defined as “cool things that computers can’t do”  If humans
can find a way to do something cool with a computer, it stops being an AI problem.
# Automatic methods for search and planning were considered to belong
to the domain of AI, and those methods are taught to every computer science nowadays.
# Certain methods for processing uncertain information becoming so well
understood that likely to be moved from AI to statistics or probability
+ Reason 2: the legacy of science fiction
# The confusion about the meaning of AI is made worst by the vision of
AI present in literary and cinematic works of science fiction
# The robothood of those creatures often only a thin cover on top of a
very human-like agent, which is understandable because it needs to be relatable by readers
 Otherwise, it will be shunned because of its difference and strange.
# Most science fiction is compared as a metaphor for the human
condition, and robots could be seen as temporary for blocked sections of society, or human
search for the meaning of life.
+ Reason 3: what seems easy is actually hard, and what seems hard is actually
easy.
WHAT SEEMS EASY IS ACTUALLY HARD
# It’s hard to know which tasks are easy and which ones are hard.
# Look around and pick up an object, then think about what you did:
* You use your eyes to scan your surroundings
* Figured out where are some suitable objects for picking up
* Chose one of those subjects and plan a trajectory for your hand to
reach that one
* Moved your hand by contracting various muscles in sequence
* Managed to squeeze the object with the right amount to keep it
between the fingers
 It can be hard to appreciate how complicated all this is, sometimes it
becomes visible when something goes wrong (the subject is lighter or heavier than expected;
someone grabs the handle and you out of balance  These tasks usually feel effortless but it
has been contradicted since several years of childhood practice.
# While it’s easy for humans, grasping objects by a robot is extremely
hard and an area of active study.
WHAT SEEMS HARD IS ACTUALLY EASY
# The tasks of playing chess and solving mathematical exercises can
seem to be difficult, but some initial AI research concentrated on these kinds of tasks.
# In-depth mastery of mathematics requires human intuition and tact,
many exercises of a typical high-school or college course can be solved by applying a
calculator and a simple set of rules.
What would be a more useful definition?
- A definition would be listed in properties that are characteristic of AI, in this case,
autonomy and adaptivity:
+ Autonomy: The ability to perform tasks in complex environments without
constant guidance by a user.
+ Adaptivity: The ability to improve performance by learning from experience.
Words can be misleading
 - When defining and talking about AI, many words used can be misleading, common
examples are learning, understanding, and intelligence.
- We define a machine as intelligent because it can perform tasks that an intelligent
person can perform
- The word ‘understand’ can be known that the system also understands what humans
can understand. Ex: If a person wearing a shirt that has a road image on it, then it’s not okay
to drive on that road (and over the person)
Note: Watch out for ‘suitcase words’
- Called ‘suitcase word’ for the term that carries a whole bunch of different meanings
that come along even if we intend only one of them
Why you can say ‘a pinch of AI’ but not ‘an AI’
- Some methods are AI or non-AI, but some involve a pinch of AI.
- It would be more appropriate to talk about the ‘AIness’ rather than arguing whether
something is AI or not
‘AI’ is not a countable noun
- AI is a specific discipline, like math or biology (a collection of concepts, problems,
and methods to solve them).
RELATED FIELDS
Machine learning
- A subfield of AI, also a subfile of computer science.
- Enables AI solutions that are adaptive.
- Systems that improve their performance in a given task with more and more
experience or data.
Deep learning
- A subfield of machine learning, AI, and computer science.
- Refers to the complexity of a mathematical model.
- The increased computing ability of modern computers allowed researchers to
increase the complexity of a mathematical model to reach levels that not only quantitatively
but also qualitatively, appear different from before.
- Enables researchers to zoom into a particular topic to catch up with the amount of
knowledge that accrued over the years, produce new knowledge on a topic, or correct earlier
knowledge to become more accurate
Data science
- A recent umbrella term (covering several subdisciplines), including machine learning
and statistics, and some aspects of computer science (algorithms, data storage, web
application development).
- A practical discipline requires understanding of the domain in which it is applied:
business or science: its purpose, basic assumptions, and constraints.
- Involve at least a pinch of AI.
Robotics
- Building and programming robots so that they can operate in complex, real-world
scenarios.
- The ultimate challenge of AI since it requires a combination of virtually all areas of
AI:
+ Computer science and speech recognition: sensing the environment.
+ Natural language processing, information retrieval, reasoning under
uncertainty: processing instructions and predicting consequences of potential actions.
 + Cognitive modeling, affective computing (systems that respond to human
feelings expressions or mimic feelings): interacting and working together with humans
- Many robotics-related AI problems are approached by machine learning  Makes
machine learning a central branch of AI for robotics.