Artificial Intelligence Reviewer

Typical Applications of AI

- Artificial Creativity - Computer vision, virtual reality, and image processing - Diagnosis - Face recognition - Game AI - Handwriting recognition - NLP - Nonlinear control and robotics - Optical character recognition - Speech recognition

ML vs Statistics

- Both can infer and predict - Inference: the act or process of reaching a conclusion about something from known facts or evidence. - Prediction: the act or process of stating what will happen in the future. - Statistics is focused more on inference - ML is focused more on prediction

Data Analysis

- Can be used in ML to predict future, classifying information, and effectively making decisions - Data analysis and machine learning enable people to push data usage beyond previous limits to develop a smarter AI. - In the context of the statement above, data analysis is the 'basic chemical transformations' that the oil undergoes in refinery before becoming valuable fuel or plastic products.

AI uses in computer applications There are two essential ways in which AI contributes to human needs:

- Correction: With the new data that it collects, it corrects its inside mechanism to adjust with new conditions. - Suggestion: With its newly calibrated mechanism, it may now suggest possible likelihood of success (e.g., shortest path/route).

Why Data Analysis Important

- Data analysis is essential to AI - AI needs the processes discussed earlier before its advanced algorithm do its work. - Various sources already produce data in such large quantities that you can find what you need without having to create special data for the task. - Provides information about files obtained from the internet. Example is an image. With data analysis, one can obtain image sizes, variety, number of colors, words used in the image titles etc.

What are Deep Learning Applications

- Online Learning - One can setup an AI that's built to recognize an uploaded picture on any social mediia if it's a cat or not. - Transfer learning - This is when two or more objects now are being classified. No need to do learning from scratch for new objects, just transfer the 'knowledge' from the learned objects - End-to-end learning - Facial recognition in the earlier days require splitting the problem into sub-problems to achieve an acceptable result in reasonable amount of time. Today, you can just guide the deep neural networks to find faces and classify them. And what's more is you can apply the same concept to language translation, or speech recognition

Application of Algorithm

- Scheduling problems - Searching routes (e.g., fastest route) - Recognizing patterns - Processing language - Bots

Types of Data

- Structured: You know exactly what it contains and how to access every data there is in the dataset. (Parang Database) - Unstructured: You have the idea of what it contains but accessing data is hard because it's arranged in a manner that's not distinguishable.

Making Suggestions

- Suggestions ≠ Command - Suggestions based on past actions: AIso can save past scenarios and use it to make new and better suggestions - Suggestions based on groups: Different individuals or objects share the same feature. An AI can easily spot a pattern when an emergent occurred.

AI based Errors

- There are some scenarios that even though, a well-designed algorithm is introduced to an AI, it still provides unreasonable incorrect response. AI still have high error rate in some circumstances and sometime, developers are clueless about what happened. - AI reduces human errors and help humans decide effectively thru human intervention.

How Machine Learning Works

- We are used as programmers creating a function that returns a result. - In ML, it's the other way around. You have the inputs, and you the output should be, but the function to obtain that output is undefined. - In order to obtain the function, the learner algorithm 'trains' itself with all sorts of examples of desired inputs and results expected from those inputs. This process is called training. - The learner algorithm then maps a flexible function to data. - Analogy: Child learning to distinguish between trees and non-tree objects

Too many factors, more time to consume. ML could help by classifying common features and use only the factors inside the chosen class.

Complex Analysis

Unfamiliar spending pattern.

Fraud Detection

For efficient scheduling. E.g. patients, based on their needs, needs to be assigned to experts

Resource Scheduling

Field where AI is commonly applied

- Artificial Life - Automated Reasoning - Automation - Biologically Inspired Computing - Concept Mining - Data Mining - Email Spam Filtering - Robotics - Semantic Web

AI with Deep Learning

- Human brains have millions of neurons, which are cells that receive, process, and transmit electric and chemical signals. Each neuron possesses a nucleus with filaments that act as inputs, dendrites that receive signals from other neurons, and a single output filament, the axon, that terminates with synapses devoted to outside communication. - Group of connected neurons via layers is a neural network:

Role of Algorithm in AI Processes

- More data = 'right' algorithm (in a general sense) - Algorithms still need human intervention

Limitations of boredom

- Other boredom is the lack of idea to know what to do next. This one cannot be addressed correctly by AIs. - AI can possibly read your facial expressions but humans have inter and intra personal knowledge thus it can't think for you. - It can help you of your idea but now with creating the idea itself.

Some examples of processes addressing these boredoms are:

- Searching for needed items online - Ordering needed items automatically - Performing sensor and other data-acquisition monitoring - Managing data - Accomplishing mundane or repetitive tasks

How can AI achieve Acting Rationally

Based on given conditions, environmental factors, and existing data

2 Human Errors

Developer and Respondents

- Machine learning enables people to perform such tasks as predicting the future, classifying things in a meaningful way, and making the best rational decision in a given context. - You can represent reality by using a mathematical function that the algorithm doesn't know in advance, but which it can guess after seeing some data. - Purely mathematical

Machine Learning

Grasps the relationship between the elements present in data. One needs tools from statistics which are, but not limited to the following: t-tests, linear regression, and correlations.

Modeling

2 Data Sources

People and Sensors

a ML algorithm learns from fed data that lack labels and learn to react to an environment on its own. However, you can accompany an example with positive or negative feedback according to the solution the algorithm proposes

Reinforcement

Data

Universal resources for AI

a ML algorithm learns from fed data without any associated responses, leaving the algorithm to determine the data patterns on its own

Unsupervised

Intelligence Comprises of what?

· LEARNING TARGET - Having the ability to obtain information. · REASONING - Being able to manipulate information. · UNDERSTANDING - Considering result from information manipulation. · GRASPING TRUTH - Determining the validity from manipulated information. · SEEING RELATIONSHIP - Divining how validated data interacts with other data. · CONSIDERING MEANING - Applying truths to particular situations in a manner consistent with their relationship · SEPARTING FACTS FROM BELIEF - Determining whether the data is adequately supported by provable sources that can be demonstrated to be consistently valid.

Ai as solutions to boredom

- We humans become more emotional or tardy, which impacts our performance as workers - AI can improve this unsatisfying scenario by intervening thru music, or light etc. AIs can be configured to detect for example, movements, or facial expressions and adjust accordingly depending on the computed 'mood' level of a person. - Another way is to assign loads to AIs which are: Loads are repetitive Require intensive computation Prone to error

Classifications of AI

1. Reactive Machines: Designed to solve using pure computing power and does not require memory or experience. 2. Limited Memory: Designed to use less memory where experiential knowledge is stored. 3. Theory of mind: Designed to consider other agent's way of thinking in the same environment and develop a plan for its action. 4. Self-awareness: Designed to do their tasks remotely because of advanced technology. AIs in this category have awareness and consciousness. Also have the ability to infer the intent of other agents based on their experiential knowledge

Define AI in computer context

1. Set a goal based on needs or wants. 2. Assess the value of any currently known information in support of the goal. 3. Gather additional information that could support the goal. 4. Manipulate the data such that it achieves a form consistent with existing information 5. Define the relationships and truth values between existing and new. information. 6. Determine whether the goal is achieved 7. Having the ability to obtain and process new information. 8. Repeat Steps 2 through 7 as needed until the goal is achieved (found true) or the possibilities for achieving it are exhausted (found false).

Employing Machine Learning to AI

1. Symbolic Reasoning - symbols 2. Connections modelled on the brain's neurons - neurons 3. Evolutionary - fitness function, heuristics, genetic algorithms 4. Bayesian Inference - continuous updating of previous beliefs that grew more and more accurate

Automating common processes

A. AI act as processes e.g., do laborious tasks B. Because AIs are consistent doers. C. Benefits: a) do exactly what they're told; b) very high throughput for a short period of time. D. Unlike for us humans, we are limited beings, when we are emotionally occupied or bored, we tend to do the bare minimum efforts for finishing the workloads we're doing.

AI makes applications friendlier

A. The user becomes more efficient by typing fewer characters B. The application receives fewer errant entries as the result of typos C. The user and application both engage in a higher level of communication by prompting the user with correct or enhanced terms that the user might not otherwise remember, avoiding alternative terms that the computer may not recognize.

Machine Learning?

AI algorithms in n this Discipline of AI contains a training set where features of training instance in this set are the basis of analysis.

Deep Learning

AI algorithms in this Discipline of AI mimics the way our brain works requiring an input which is then processed by networks of 'neurons' which are continuously adjusted through weights and bias to produce a well-trained network. Having a well-trained network produce more accurate results at for example, recognizing patterns in images.

Capabilities of Artificial Intelligence in order to mimic the human intelligence

Acting Humanly, Thinking Humanly, Thinking Rationally, Acting Rationally

Problem with automation today -- upon encountering unknown phenomena, automation could stop. AI e.g. ML can be used to handle unexpected events.

Automation

How can AI achieve Thinking Rationally

Based on recorded behaviors and create guidelines to aid its decision-making process.

Fixes imperfect data. Example subjects for cleansing is missing information, extremes in range, or simply wrong values

Cleansing

Quality of Data

Commission - the outright attempt to substitute truthful information for information. Omission - providing cherry-picked true data, neglecting other important facts for purpose that benefits the person Perspective - people where each provide data based on perception Bias - the quality of data from a person is affected by that source's belief or condition. Frame of reference - the person's understanding of the context is not sufficient so the data he/she provided might be ineffective

What is Data?

Data is the quantities, characters, or symbols on which operations are performed by a computer, being stored and transmitted in the form of electrical signals and recorded on magnetic,

Benefits of Machine Learning?

Fraud Detection - Unfamiliar spending pattern. Resource Scheduling - For efficient scheduling. E.g. patients, based on their needs, needs to be assigned to experts Complex Analysis - Too many factors, more time to consume. ML could help by classifying common features and use only the factors inside the chosen class. Automation - Problem with automation today -- upon encountering unknown phenomena, automation could stop. AI e.g. ML can be used to handle unexpected events.

Validates the data. Job of a human to recognize patterns and spot strange elements. Can be done using Visualization tools (e.g. graphs)

Inspecting

Two of the Discipline in AI

Machine Learning: AI algorithms in Machine learning contains a training set where features of training instance in this set are the basis of analysis. Deep Learning: AI algorithms in Deep learning mimics the way our brain works requiring an input which is then processed by networks of 'neurons' which are continuously adjusted through weights and bias to produce a well-trained network. Having a well-trained network produce more accurate results at for example, recognizing patterns in images.

AI uses on different field of studies

Monitoring AI: - AI which can track movements and heart rate used for statistical analysis for the purpose of providing advice to create better workout. Games with Ais - AI rewards a therapy patient when they achieved specified movements in a fun way thru games. - Farm simulation game Movement sensing Ais - AI has access to person's movement and turn it to a useful quantifiable data. - Exoskeletons for handicapped persons. Also used in military Medical - Records heart rate, sleeping schedules - Downside: lack strong security. Data could be used against you for unethical activities.

How can AI achieve Thinking Humanly

Observed thru different tests: introspection, psychological testing, and brain imaging

a ML algorithm learns from fed data and associated target responses

Supervised Learning

Machine Learning Types

Supervised Learning Unsupervised Reinforcement

Changes the data's current appearance. Example is transforming data into ordered rows, and columns -- a matrix.

Transforming

Processes During Data Analysis

Transforming - Changes the data's current appearance. Example is transforming data into ordered rows, and columns -- a matrix. Cleansing - Fixes imperfect data. Example subjects for cleansing is missing information, extremes in range, or simply wrong values Inspecting - Validates the data. Job of a human to recognize patterns and spot strange elements. Can be done using Visualization tools (e.g. graphs) Modeling - Grasps the relationship between the elements present in data. One needs tools from statistics which are, but not limited to the following: t-tests, linear regression, and correlations.

How can AI Achieve Acting Humanly

Turing Test