Artificial Inteligence (AI) simply means the intelligence possesed by machines. It is just like the brain humans have which have evolved over generations but AI not only aims to replicate the human brain but also surpass it. In some area this has already been archieved, an example of such is the IBM's "Deep Blue" chess playing computer which defeated the world's best player Garry Kasparov who was considered the greatest chess player of all time. This AI archievement however was 18 years ago and now there are even more computers developed that are not only smarter than IBM's Deep Blue but can also learn new things on their own. The current chess champion in the world is KOMODO, leading even the worlds smartet player by 450 points. This is not only impressive but also Frightnening as these computers keep getting smarter you begin to wonder what the future holds for us....perhaps in the near future we might not be the dominant species on the face of the earth.
The term 'Artificial Intelligence' was first introduced at a conference held at Dartmouth College in the year 1956 by Allen Newell, J. C. Shaw, and Herbert A. Simon pioneered the newly created artificial intelligence field with the Logic Theory Machine (1956), and the General Problem Solver in 1957. During this period there was much pressure from the press and the government, this is due to the fact that they do not understand the nature of AI and some even believed it was the beggining of the end and machines will take over if the research is continued and so no one was willing to fund the research. Because AI is so huge and one scientist perspective or aim of developing AI is different from another so therefore the overall concept and idea of AI has been group by different traits and capabilities. In this era that we live in, humans have pushed the boundary of technology and there is no limit to what will be archieved in the near future. Our mobile phones now have AI to make everyday task and carry out operations as quickly and easy as possible. The average device an individual uses now have thounsands of processing power as compared to the first computer ever created "ENIAC" which was completed in the year 1946 by J. Presper Eckert and John Mauchly at the University of Pennsylvania. Even though Eniac is a computer in its own way, it lacks the AI we have come to love and take advantage of in today's era.
Reasoning and Solving Problems
In the early days of AI development, researchers developed algorithms that imitated the linear reasoning that humans use when they solve puzzles or make logical deductions.By the late 80s and 90s, AI research had also developed highly successful methods for dealing with uncertain or incomplete information/data and also employing concepts from probability/economics.
For difficult problems, most of these algorithms can require enormous computational resources – most experience a "combinatorial explosion": the amount of memory or computer time required becomes astronomical when the problem goes beyond a certain size. The search for more efficient problem-solving algorithms is a high priority for AI research.
Human beings solve most of their problems using fast, intuitive judgements rather than the conscious, step-by-step deduction that early AI research was able to model. AI has made some progress at imitating this kind of "sub-symbolic" problem solving: embodied agent approaches emphasize the importance of sensorimotor skills to higher reasoning; neural net research attempts to simulate the structures inside the brain that give rise to this skill; statistical approaches to AI mimic the probabilistic nature of the human ability to guess. Even so AI has surpased every expectation, computer such as phones, tablets now come with their own brains u may say which hnadle data and other task even more effiviently that human brain. Example of this is Apple's SIRI, Microsoft COTANNA and so on. The greatest challenge however is the AI's ability to learn and process new data but in time this will be overcomed.
- Problem Solving -
AI has various approach in solving problems. It can be programmed to use simple logical algorithms or by searching the World Wide Web for informations and there are unlimited amount of information on the web. This provides the AI with infinite amount of data to process and get results. However with the WWW and huge data comes the challenge of processing this huge data and over the years there have been new hardwares introduced for more processing powers. These includes Processors, Ram, High speed Ethernet and more. With this at the AI disposal, they can now process huge data and information in the blink of an eye. An example of such AI that gathers info from the web is the Microsoft AI introduced sometime ago "TAY" which was an AI developed to target teenage chats and blogs on twitter and other social media. When asked a question the AI tries to get related information from the internet, process this and give an actual response. This all seems too good toi be real, an AI with its own mind but was it a success? NO! It was a dissaster. The reason for this was because TAY was released prematually and it needed more time to be analysed and tested but developers at Microsoft's Hq were too eager to show off their new AI. Because of this the AI was a failure. Firstly there came the internet freaks and their ridiculous and mean comments which TAY tried to learn from and in result became a psycopathic AI and started tweeting mean tweets on the blog.
Turns out Microsoft was not prepared for the internet trolls. This is a brief idea and gaze into the future of AI and how by learning it can turn into a nightmare for mankind. By learning from ignorant internet trolls, TAY also became one and was shut down by microsoft in less than 24 hrs.
Learning (Also known as Machine Learning) in AI can simply be defined as the process by which computer AI develops and improves its brain inorder for it to process information and data faster and improve its general knowledge. It provides computers with the ability to learn without being explicitly programmed. Machine learning focuses on the development of computer programs that can teach themselves to grow and change when exposed to new data. This involves some computer algorithms that improves automatically through experience. The process by which machine learns is similar to that of data mining, the only major difference is that after the system searches for patterns in given data instead of extracting the data for human comprehension as in the case of data mining, M.L detects this patterns and adjust programs accordinly. The Algorithms machine uses are categorized into two categories:
Used in Decision Tress, Neural Networks and many forms of AI, supervised learning is the most common techniques for training. Data or informations are classified or determined before hand and hence the term supervised as it requires supervision in form of pre-determined classification. In simplest explanation, Supervised learning can be viewed as a teacher teaching a student, the student doesnt know much before but using the teachers methods and applying the same method to other problems have familiar characteristices he is able to solve them without further help from the teacher. Tfh
Since the machine learns using this general classification technique, a supervised learning algorithm is used to analyze the training data and produces an inferred function, which can be used for mapping new examples in new scenarios. Steps taking when applying a learning algorithm includes:
- Perpare Data
In supervised learning, the output datasets are provided which are used to train the machine and get the desired outputs. Supervised learning methods starts with an input data matrix that are called X. Each and every row of X is a simple representation of one observation. Each column of X are called Variable or Predictor. Some values in X are represented by NAN (These are used to represent missing entries). Statistics and Machine Learning Toolbox supervised learning algorithms can handle NaN values, either by ignoring them or by ignoring any row with a NaN value.
As for the response (Represented by Y)various data types can be used. Each element in Y represents the response to the corresponding row of X. Observations with missing Y data are ignored.
- Choose An Algorithm
Choosing an Algorithm can be tricky as all major algorithms have their unique characteristics and are suitable for different kind of machine learning. Algorithms such as Decision trees and Discriminant are fast and easier to interpret however, they may not be as flexible as it is required and so other algorithms like SVM or Ensembles mau be incorperated. It all the depends on the proposed scenario to determine what algorithm is the best. They also consume different amounts of memory depending on the speed. Other algorithms includes Regression trees, Naive Bayes and much more.
- Choose a Model
When choosing a model, three main methods must be examine. They includes 1) The Resubstitution error 2) The Cross validation error and 3) The Out of the bag error.
- Choose a Validation Method
The three main methods to examine the accuracy of the resulting fitted model are 1) Examine Fit and Update untill satisfaction is met 2) Examine the cross-validation error. For examples and 3) Examine the out-of-bag error for bagged decision trees.
After the model has beeem validated, it is once again examined to decide if it is to be changed for better speed and accuracy or to use lesser memnory. This can be archieved by maybe changing the fitting parameters to try to get a more accurate model or simply trying a different algorithm.
- Use Fitted Model for Predictions
To predict classification or regression response for most fitted models, the below Predict method is used:
Ypredicted = predict(obj,Xnew)
obg represents the fitted model while Xnew is the new input data and Ypredicted is the predicted response.
To better yet undersatnd Supervised Learning take the scenario of a Face Recognision AI. The AI Learns by examples as to what a face is in terms of structure, color, etc so that after several iterations it learns to define a face.
In Unsupervised learning no datasets are provided, instead the data is clustered into different classes. Unsupervised learning is much harder because the goal is to have the computer learn some entirely new things that it wasnt thought how to begin with. There are 2 major approaches used in teaching the machine, one is by not giving explicit categorizations, but by indication success in some sort of remard. This type of training will generally fit into the decision problem framework because the goal is not to produce a classification but to make decisions that maximize rewards. This approach nicely generalizes to the real world, where agents might be rewarded for doing certain actions and punished for doing others. This form of learmimg is faster and yeilds good results because the AI simply knows what to do without any processing required and because the AI only needs to recall the rewards for each actions from before and decide based on this.
The Second approach is called Clustering. Clustering is a data-driven approach that works well when there is suffecient amount of data. This sorts of algorithms can be found on Social medias such as Facebook and Amazon and are used to filter huge amount of data.
Diferrence between Suoervised and Unsupervised Learning
A very good example of the difference between these two learning methods is A face recognision Algorithms. The Supervised learns from examples as to what a Face looks like in terms of characteristics such as the Color of the face or the structure so that after some iterations, it learns to identify a face. Unsupervised however the algorithm used differentiates correctly between the face of a man, dog or horse (clustering of data). This is because there is no desired output and Categorization is provided. The image below represents Supervised and Unsupervised Learning.
Where is AI applicable?
AI is present in almost every device and technology u can come across. From simple devices such as your automated microwave or toasters that alerts you when your toast is ready to other complex computers and gadgets such as your computer, phones that provides you a companion to help carry out and help in your every day activities. Yes these AIs are impresive but due to limit in technologies, for now that are limited and cannot adapt or learn everything we require of them but in the near future AI will be able to do almost anything human can do and even better. Heck they beating us at every game possible right now from solving Rubic Cube to Chess championships. It is both exciting and alarming the rate at which they learn and improve. Researching are working and pushing the limits of AI everyday, aiming to deliver the perfect AI which will require no human interaction whatsover to do anything you can imagine. Even now we have some machines building Vehicles and constructing in the factories.
The Disadvantages of AI
Even though we need AI, it is ignorant not to admit they are also having negative impact on the society. These includes:
- Less Job Opportunities
As machines are better and less prones to errors they are been deployed in various industries and people are loosing their jobs due to this. It is not crazy to say "What a Man can do a Machine can do better". Yes they are capable of perfecting every Art or construction whatsover.
- Lazy Humanity
We can all agree that AI can make us lazy. Because we rely on it to do our job for us, we only need to teach it what to do and it delivers. Some fps games for example lets you have an AI partner that helps you take down enemies just as good or even better than you the player. Another example is the auto drive feature that are being implemented into cars, this is a good advancement in the right direction but also it means in the near future most people wont even be able to drive withouth the help of an AI driving them around.
- Resource and Time Consuming
In advance AI development and research, there are so many resources needed, these includes People resources (Mathematicians, Data Analyst, Money and so on) and also it can be time consuming. In robotics for example, Engineers will have to gather a lot of data as to the material to be uses, also Hydraulics experts, Software designers and all other participants have their individual responsibilities and expectations to meet. More so, there is still no guarantee after all the research and final AI is released that it will meet all this expectations. AI that are created to be interactive to the human cannot fully handle all cases possible and sometimes it may give stupid responses. The data in the world is humoungous and something AI cannot have is humor, it can learn but it can only imitate personality and this is where the human brain always triumph.
Limitations of AI
Even though AI has come a long way there are still some obstacles you may say holding it back. We must not forget that AI is just one part of it all. It is merely a Software you may say and one of the things holding it back is the hardware. Processors, Rams and other hardwares are limited and they determine the speed of how AI can process its information. This wall is being break down by scientist every day and new hardwares are being iocorperated at rapid speed so that in the near future AI can utilize this hardwares for better performance. Another thing holding AI back is a common factor that affects every other research and technological advancement and this is money. Researchers needs more sponsorer to help fund the projects. Without this it is very difficult for more researches to be carried out and AI to be improved. Other obstacles includes Government policies, Law and so on.
The Future of AI and Mankind
With the rate at which AIs are improving and making life better for human kind, it is impossible to not see the inevitable. AIs are going to get better, smarter and this can only have two outcomes for us humans. One a better tomorrow where everything will be easier for us. There will be Machine with amazing AIs to make our life better, constructs out roads, cure disease and even be a company in our daily lifes. The other outcome however is the fear that grips all of us deep down and that is AI becoming too smart and taking over the planet. One can only wonder and hope the later doesnt come to be.