It's something that I had in my mind for a long time but never got the time to execute it but finally, I decided to get out of my comfort zone to learn new concepts and techniques that would enable me to solve new problems. Hence, I chose to study Artificial Intelligence.
I did some online research and found out a really good book named Artificial Intelligence, A Modern Approach (3rd Edition) by the Stuart Russell and one of my favorite computer scientists, Peter Norvig (Director of Research at Google) to learn about it's concepts and techniques. The book has 1000+ pages and it's a book used for undergraduate and graduate level courses in university.
My current knowledge of Artificial Intelligence is pretty basic (e.g: write game AI) and I want to learn more about it and be able absorb any information related to it and build toy AI projects.
I've completed the first chapter of the book, so let's dive in because this going to be a long read.
What is Artificial Intelligence?
We read about it in the news, it's being deployed in our mobile applications that we use everyday such as Facebook, Instagram, Twitter, Reddit and so on to filter out graphic content, fake information and insensitive political content. It's also being used in games such as chess, scientific research, diagnosis of several diseases and self-driving cars.
But do we know what is it? According to Google Search, it means:
The theory and development of computer systems able to perform tasks normally requiring human intelligence such as visual perception, speech recognition, decision-making and translation between languages.
It encompasses a huge number of fields and sub-fields and AI is already the next big thing that it's shaping our everyday life.
Approaches towards AI
The book states that there are four types of approaches when it comes to creating an AI:
Acting Humanly
Proposed by British computer scientist Alan Turing, the Turing Test approach was designed to provide a functional definition of Artificial Intelligence. The test is proved positive only when a human is unable to tell the difference between the results of a computer or a human. In order to think like a human, it should possess the following capabilities:
- Natural Language Processing to enable communication in any language with the human
- Knowledge Representation to store what it knows or hears
- Automated Reasoning to make conclusions based on the repository of information to answer questions
- Machine Learning to learn and adapt to new patterns and extrapolations
It is to be noted that the test deliberately avoided interaction with the human because physical interaction with the human wasn't necessary for intelligence.
Then came another test called Total Turing Test, this was made to test the computer's ability of visual perception. The computer passes this test when it's possess the following abilities:
- Computer Vision to be able to perceive and identify objects
- Robotics to be able to manipulate physical objects
The abilities mentioned above, composes most of what modern AI is today and Turing deserves a huge credit for designing this test that still remains relevant for more than 60+ years.
Thinking Humanly
Do we know how humans think? Maybe, but for us to be able to determine that, we would need to achieve a deep understanding of the human mind works. There are a few ways such as:
- Introspection by catching our own thoughts as they pass by
- Psychological experiments by observing the actions or behavior of a human
- Brain imaging by observing the brain in action
Once we have sufficient information, it's possible to theorize that a computer program behaves like a human. Cognitive Science enables you to combine both computational models of an AI and psychological experimentation techniques to provide testable theories as to how the human mind works.
Thinking Rationally
Greek Philosopher Aristotle attempted to arrange information based on irrefutable evidence based on the process of reasoning. His rules of inference a.k.a syllogisms (a form of reasoning in which conclusions are drawn from various propositions or a set of premises) provided patterns that yielded correct conclusions from correct premises. For example: "Socrates is a man; All men are mortal; therefore, Socrates is a mortal being". These laws of thought initiated the study of logic, which gave hope to 19th century logicians to help create intelligent systems.
However, there are two main obstacles to this logical approach. Firstly, it's difficult to convert informal information into formal terms required by logical notations especially when the information isn't 100% certain. Secondly, being able to solve a problem in theory vs. solving a problem in practice are two different things. You can have a computer that can solve a problem with a few hundred facts yet use up all of it's resources.
Acting Rationally
This is focused on creating intelligent agents that can perform various tasks like being able to operate autonomously, perceive objects, adapt to change, create new goals and pursue them. A rational agent is an agent that acts to achieve the best expected outcome.
Making the right conclusions based on evidences i.e. correct inferences is part of a rational agent because to act rationally, an agent must be able to reason with logic to reach to a conclusion for a given action to achieve one's goals.
However, it doesn't necessarily that it's always "correct", sometimes, it has there's no such thing as the "right" thing to do but something must be done.
The skills needed for a Turing Test allows an agent to act rationally especially on making good decisions using Knowledge Representation and Automated Reasoning, generating intelligible sentences using Natural Language Processing for a complex society, adapting to change and generating effective behavior using Machine Learning.
But, there are some advantages to this approach. Firstly, it's more general in terms of the logical approach (mentioned in Thinking Rationally). Secondly, it's more open to scientific development compared to human behavior (mentioned in Acting Humanly) and human thought (mentioned in Thinking Humanly). The standard rationality of an agent is purely mathematically defined and completely general whereas human behavior adapts to a specific environment.
Later, the book states that it's focus is going to be based on the general principles of rational agents and on components for constructing them.
Is AI a science, or is it engineering?
As I was reading the book, it was fascinating to see how various disciplines have contributed ideas, techniques and viewpoints to the field of Artificial Intelligence. The following disciplines are:
- Philosophy
- Neuroscience
- Mathematics
- Economics
- Linguistics
- Psychology
- Computer engineering
- Control theory and cybernetics
Each disciplines had thoughtful questions like How does a human brain work? How are valid conclusions drawn from formal rules? How can we build an efficient computer? How to think and communicate in one's language? How does the brain process large amounts of information? How do humans and other living things think and act? How does language relate to thought?
This part of the book is really long but it was a good way to understand about it's early foundations.
How is it useful today?
Well, that's not very easy to answer because it's being used in multiple fields and sub-fields. There are so many applications such as:
- Self Driving Cars
- Speech Recognition
- Facial Recognition
- Fighting Malware and Spam bots
- Filtering graphic content and fake information from social media
- Game playing AIs for different board games like Checkers, Go and Chess
All of this used to be science fiction but thanks to the advancements of Mathematics, Science and Engineering, it's become a reality in today's era.
Conclusion
Well, I don't know if this is one of the longest articles I have ever written but I really did enjoy writing this because this made me read the chapter again and gained a better understanding of the concepts.
I will be writing more articles about it, write algorithms and build toy implementations of Artificial Intelligence applications.
In fact, I wrote this article to answer all, if not, most of the questions from the exercises section of this chapter.
Hope you liked reading this article!
Stay tuned for more!