Notes

Using Al to Make Business Decisions

Artificial intelligence simulates human thinking and behavior, such as the ability to reason and learn. Business person believe AI will soon be integrated with every product and service. There are two categories of AI:

- Weak AI: Weak AI machines can still make their own decisions based on reasoning and past sets of data. Most of the AI systems on the market are weak AI.

- Strong AI: Strong refers to the field of artificial intelligence that works toward providing brainlike powers to AI machines. It works to make machines as intelligent as humans

Massive amounts of data are used as inputs to train AI systems such as billions of images, voice, video, and IoT sensor data. Al models use the data to self-train and then make predictions on new data. AI primary overview includes four areas:

1. Natural language processing: Uses language as an input that a computer system can decipher and act upon its meaning, such as Siri and Alexa.

2. Natural language understanding: Determines a user's intentions based on what the user typed or said. For example, a search engine uses natural language understanding to determine what the user is searching for based on what the user typed or said.

3. Knowledge representation: Stores large amounts of data with fast access.

4. Knowledge planning. Uses stored data to make predictions and decisions in real time such as goal-seeking analysis.

Expert systems are computerized advisory programs that imitate the reasoning processes of experts in solving difficult problems. Expert systems are the most common form of AI in the business arena because they fill the gap when human experts are difficult to find or retain or are too expensive. The best-known systems play chess and assist in medical diagnosis.

Case-based reasoning is a method whereby new problems are solved based on the solutions from similar cases solved in the past. An auto mechanic who fixes an engine by recalling another car that exhibited similar symptoms is using case-based reasoning.

Algorithms are mathematical formulas placed in software that performs an analysis on a dataset. Algorithms use formulas to solve problems, such as driving cars or playing chess. Driving a car is a difficult task and takes hours and hours of training for humans, let alone computers. Playing chess has one goal-to win-and can take hundreds of hours to learn how to play. In artificial intelligence, an algorithm tells the machines how to figure out answers to different issues or questions

A genetic algorithm is essentially an optimizing system: It finds the combination of inputs that gives the best outputs. Genetic algorithms can find and evaluate solutions with many more possibilities, faster, and more thoroughly than a human. Organizations face decision-making environments for all types of problems that require optimization techniques, such as the following:

- Business executives use to help them decide which combination of projects a firm should invest in.

-Investment companies use to help in trading decisions.

Machine learning is a type of artificial intelligence that enables computers to both understand concepts in the environment and also to learn. Is based on the principle that systems can learn from data, identify patterns, and make decisions with minimal human interaction.

Types of Machine Learning:

- Supervised Machine Learning: Is similar to a student learning a subject by studying a set of questions and their corresponding answers. After mastering the mapping between questions and answers, the student can then provide answers to newquestions on the same topic.

- Unsupervised Machine Learning: The most common use of unsupervised machine learning is to cluster data into groups of similar examples. For example, an unsupervised machine learning algorithm can cluster songs together based on various properties of the music.

- Transfer Machine Learning: Transferring information from one machine learning task to another. Most machine learning systems solve a single task. Transfer learning is a baby step toward artificial intelligence in which a single program can solve multiple tasks.

The secret to building successful machine learning models is to ensure the training data has enough data to train the model. Data augmentation can rotate, stretch, and reflect each image to produce many variants of the original images providing enough examples for training. Once you have your training data, you need to watch for two additional learning problems: overfitting and underfitting

- Overfitting happens when a model learns the details in the training data to the extent that it negatively impacts the performance of the model on new data. Essentially, the model knows the training data too well and is unable to make future predictions.

- Underfitting occurs when a machine learning model has poor predictive abilities because it did not learn the complexity in the training data. Many problems can cause overfitting and underfitting, and finding the sweet spot between the two is a difficult task.

Machine learning models along with the training data are created by humans and will do exactly what they are taught to do. Clearly, with humans building the algorithms and feeding the training data, there are problems with bias being introduced into the artificial intelligence models. Bias can be detected and mitigated if you know what bias looks like and can identify its source. The four type of bias in machine learning include:

- Sample bias is a problem with using incorrect training data to train the machine. For example, you are training an autonomous vehicle to drive in all weather conditions, but your sample only has driving data on sunny days over 85 degrees. You have now introduced sample bias into your model. Training the algorithm to drive in rain, snow, sleet, hail, etc. would eliminate this source of sample bias.

- Prejudice bias is a result of training data that is influenced by cultural or other stereotypes. For example, you are training a machine vision algorithm, and you have men going to work and women taking care of children in your data images. The algorithm is likely to learn that men work and women stay at home. The primary issue with prejudice bias is that the training data decisions consciously or unconsciously reflected cultural and social stereotypes.

- Measurement bias occurs when there is a problem with the data collected that skews the data in one direction. For example, if the same camera takes photos of all the training data and there is a problem with the camera's filter, then the images could be distorted. The algorithm would be trained on image data that is incorrect and does not represent reality.

- Variance bias is a mathematical property of an algorithm. Models with high variance can easily fit into training data and welcome complexity but are sensitive to noise. Models with low variance are more rigid, less sensitive to data variations and noise.

Neural Networks

A neural network, also called an artificial neural network, is a category that attempts to emulate the way the human brain works. Unlike humans, these software robots work at a much faster rate and never sleep, saving both your business money and freeing up employees to work on more creative and exciting tasks

Neural networks analyze massive quantities of data to establish patterns and characteristics when the logic or rules are unknown. These models take inspiration from the brain and are composed of layers consisting of simple connected units or neurons.

Fuzzy logic is a mathematical method of handling imprecise or subjective information. The basic approach is to assign values between zero and one to vague or ambiguous information. Zero represents information not included, whereas one represents inclusion or membership.

The growth of computer capacity that can train larger and more complex models leads to the development of deep learning and reinforcement learning.

Deep learning is a process that employs specialized algorithms to model and study complex datasets, the method is also used to establish relationships among data and datasets. Deep learning trains from each layer and then uses that learning in the next layer to learn more, until the learning reaches its full stage through cumulative learning in multiple layers.

Reinforcement learning is the training of machine learning models to make a sequence of decisions. The model learns to achieve a goal in an uncertain, potentially complex environment, for example, a game-like situation. To train the model, the programmer uses either rewards or penalties for the actions it performs. The model's goal is to maximize the total reward.

Virtual Reality

Virtual reality is a fast-growing area of artificial intelligence; enables telepresence by which users can be anywhere in the world and use virtual reality systems to work alone or together at a remote site. Typically, this involves using a virtual reality system to enhance the sight and touch of a human who is remotely manipulating equipment to accomplish a task. Examples range from virtual surgery, during which surgeon and patient may be on opposite sides of the globe, to the remote use of equipment in hazardous environments such as chemical plants and nuclear reactors.

Augmented reality is the viewing of the physical world with computer-generated layers of information added to it.

A virtual workplace is a work environment that is not located in any one physical space. It is usually in a network of several places, connected through the Internet, without regard to geographic borders Employees can interact in a collaborated environment regardless of where they may happen to be in the world.

A haptic interface uses technology allowing humans to interact with a computer through bodily sensations and movements-for example, a cell phone vibrating in your pocket. A haptic interface is primarily implemented and applied in virtual reality environments and is used in virtual workplaces to enable employees to shake hands, demonstrate products, and collaborate on projects

Augmented reality simulates artificial objects in the real environment, virtual reality creates an artificial environment to inhabit. In augmented reality, the computer uses sensors and algorithms to determine the position and orientation of a camera. AR technology then renders the 3-D graphics as they would appear from the viewpoint of the camera. In virtual reality, the computer uses similar sensors and math. However, rather than locating a real camera within a physical environment, the position of the user's eyes are located within the simulated environment. If the user's head turns, the graphics react accordingly.