"Don't spook the fishes with your noisy gestures, it'll make our exercise futile", the father said to his son. The father gave his son a few more guidelines on how to carry out this activity successfully. He showed him the part of the lake where he could easily find fish, the condition of weather that suits fishing, and the gear needed for fishing. Just like the father gives tips to his son on how to carry out effective fishing is the same application we give to machines but instead of verbal instructions, we load them with data for them to learn from. The valid interpretation of this data is termed intelligence. The interpretation of the data loaded or programmed in a machine is Artificial intelligence. It is "artificial" because it is not natural and it requires data sequencing and preprocessing to give it an unbiased output. AI is still evolving, and as we proceed, I'll open your minds to a few more things that you should know about AI using the fishing illustration, so you can grab its basics, stay with me.
Let's talk about Machine Learning before I show you a few more interesting things about AI, even though one cannot do without the other. Machine learning with the illustration of fishing is literally teaching the machine how to fish than giving it fish. Instead of you taking the machine to the lake and placing it on one spot to fish on that spot, you load it with numerous training data on fishing and let it explore its options, supervised, unsupervised, semi-supervised, or in the reinforcement way(where it's been rewarded for its achievement or given a penalty for its failure). The more exposure to fishing style data it gets, the better it gets at fishing. Isn't this interesting? They have the tendency to become better than humans logically but it's a bit far-fetched to come to this conclusion.
TYPES OF AI WEAK OR NARROW AI STRONG OR GENERALIZED AI
Let me break down each type of AI to you, the description of each AI could take thousands of pages to explain but I'll explain this in the simplest form that could guarantee your understanding of the basic knowledge of AI and how they function.
WEAK/NARROW AI: Weak or narrow AI can also be termed as reactive machine/AI. Like Newton's third law of motion states, for every action, there's a reaction. This also applies to the philosophy of weak/narrow AI, as its output is solely dependent on its input. It doesn't evolve nor deviate from its programming. Like the fishing illustration I made earlier, where a machine/robot is taken to a lake to fish at one spot and it stays there just throwing the hook in and drawing it out without caring about the semantics of how effective fishing works, so either there's fish there or not, it continues the same procedure day in, day out. An example is the self-driving car and also the IBM chess robot. We can also include the Chat bot.
STRONG/GENERALIZED AI: Strong AI or AGI as popularly called are simulations of human intelligence that have a range of improvements. They are given data to train them with and they improve on this data with time and as data evolves and expands, their abilities expand with it. In relation to fishing, AGIs would be trained on how to study patterns, and weather conditions via sensors to fish movement and waves connecting to neural networks with the data given to it carries out good fishing, and with frequent fishing, it develops better ways of fishing and gets better results. That's how AGIs function. They're still limited to humans as they do not understand emotions, a sense of belonging, and basically the soulful expression of humans.
SUPERINTELLIGENT AI: This type of AI is more like a budding prototype and this might take years to achieve. This is because it requires the soulful part of humans to carry out its function. It's of two types, the theory of mind and self-awareness AI. A superintelligent AI can carry out fishing with expressions, like "Wow!, I did that!!" or the machine can name the kind of fish it caught and how many pounds it weighs. I believe this level of invention is attainable only time is its limit but can't be for too long.
HOW AI WORKS
There are stages AI passes through for it to be effective and termed an AI. I will share with you these simple steps to enable you to create an AI with little to no stress that comes with tons of research when it could just be narrowed to these few steps. These steps include:
PROBLEM-SOLVING: Identify the problem you want to solve with AI.
DATA COLLECTION: Collect data of different sizes in correlation to the problems you want it to solve and train the AI with. DATA PREPROCESSING: Remove the noise from the data and all the inconsistencies in the data before training the model.
MODEL SELECTION: Once the data has been collected and preprocessed, the next step is to select an appropriate model that can solve the problem at hand. This involves choosing a suitable algorithm, architecture, and hyperparameters.
TRAINING: After selecting the model, the next step is to train it using the training data i.e. the preprocessed data that's been denoised solely for the purpose of problem-solving. This involves optimizing the model created to bring out the best output after a series of trials and testing.
EVALUATION: Once the model has been trained, the next step is to evaluate its performance using the test data. This involves calculating metrics such as accuracy, precision, recall, and F1-score.
DEPLOYMENT: Finally, the trained model is taken out on the field to express its function and to let people know how it operates. It can be used to make predictions or decisions depending on the area of problems, it's designed to solve. Now with these few things I've shared about AI, I'd like us to dive into the world of Machine Learning not too deep but let's go about 3 to 4 ft deep so we don't drown, just stay with me or wear a life jacket either way, I won't leave you.
MACHINE LEARNING
I believe I've explained the meaning of Machine learning from the introduction but let me juggle your memory with the expression "Teach how to fish rather than giving it fish", that's what machine learning happens. Letting the machine understand things from its perspective rather than giving it one. I will explain to you the various types of machine learning which include:
SUPERVISED LEARNING: This type of machine learning is stereotyped, it has predetermined results. So let's say a machine or robot is programmed to catch 10 fishes from a pond, no matter how many fishes are there in that pond, it won't deviate from the goal of 10 fishes if it catches 9 and probably other fishes have gone far away from its reach, it won't stop until it gets the 10th one and there's nothing in the world that can make it go for an 11th except an anomaly or malfunction which happens like none of the time because it is supervised. It's absolutely linear in its execution, it learns solely from the generally accepted answer and doesn't function outside of its frame. If A is for APPLE in supervised learning, there's no A for Hanger, it'll give an error. If 9+9 =18, there's no 15+2= 18, it'll give an error. Comprehende?
UNSUPERVISED LEARNING: This type of learning gives room for exploration. It doesn't settle for predetermined answers, it's given the avenue to explore and analyze different possible outcomes based on the data it's been given or "fed" with. It doesn't require a training dataset to get an outcome, it handles data blindly without any prior knowledge of what it's about to implement. In this type of machine learning, the machine learns itself, with all the noise and inconsistencies in the data. In other words, there's no predetermined 2+2=4, there are just random numbers given to it, to help figure out how it can achieve 4, and with A for APPLE, it's given different letters to figure out which could represent APPLE. Capiche? I hope so.
SEMI-SUPERVISED LEARNING: SSL as it is fondly called is achieved between supervised and unsupervised machine learning. It is more like the inference of both machine learnings( the supervised and unsupervised). Noisy data is given for a supervised machine to fill in the gap with its trained data. It reduces human bias, I feel it's the most accurate with its outcome because there's a bridge between the worlds of AI and natural intelligence. If 2+2=4 with supervised learning, with semi-supervised learning, 2 and the result 4 is given, then it's left for the machine to fill in the gap by either achieving 4 by multiplying by 2 or adding 2 to achieve 4 or it could come up another inference to achieve the result 4.
REINFORCEMENT LEARNING: Reinforcement learning is a type of programming in machine learning that trains algorithms using the style of reward and punishment. In reinforcement learning algorithm, or agent(you'd understand this the further you go, don't fret), learns by interacting with its environment. It is rewarded when it does well and penalized when it doesn't. So it learns without having to be directly taught by a human, it just wants all the goodies, so it tries to minimize being wrong. So when an unsupervised machine gives a correct result, it is rewarded but when it doesn't, it is given a penalty, that's how reinforcement learning works. This learning is dependent on context because what may lead to tremendous or accurate results in one situation may get a penalty in another. This type of learning consists of three components: the agent (the AI learner/decision maker), the environment (everything the agent has interaction with), and actions (what the agent can do). So it's like when 2+2=4 is achieved in the context of elementary mathematics, it is awarded a "candy". I hope I've been able to enlighten you a little bit with my perspective on elementary AI and ML, pretty sure it's been insightful. If you want to learn more about this, I'll recommend that you check these links out :
https://builtin.com/artificial-intelligence/types-of-artificial-intelligence https://www.edureka.co/blog/types-of-artificial-intelligence/