What is AI? - An Introduction Article to AI world

Introduction

Artificial Intelligence (AI) is simply defined as mimicking human interaction through computer machines. AI may require special computer hardware or software to run, although there is no specific language for AI programming. This can be programmed through Python, Java, C++, etc.! ?(Laskowski and Tucci, 2022)

Hyperscale Google defines AI as the ability of computer to mimic human advanced actions as seeing, understanding, and translating languages to make recommendation and decisions. Optical Character Recognition, known as ‘OCR’, is an example of seeing and extracting data from different documents and images which is unstructured data to structure valuable data that can be used for different applications to make appropriate decisions or recommendations which are human like. (Google Cloud, 2023)

Other hyperscale players as Microsoft and AWS also define the AI in a similar way to Google as the ability to mimic human cognitive characteristics including learning and problem solving through collecting and analysis of large datasets. (AWS), (“What Is Artificial Intelligence? | Microsoft Azure”)

Cognitive-like skills developed by AI are (S and Raja) and (Laskowski and Tucci, 2022):

·Learning – Ability of data acquisition and analysis to create rules known as ‘Algorithms’. These “Algorithms’ shall provide detailed actions for computing different tasks.

·Reasoning – selection of the right algorithm to achieve the right outcome.

·Self-Correction – Algorithms continuous learning and improvements to provide more accurate outcome.

·Creativity – Using AI techniques to generate Human like outputs as Images, speech, or ideas. This may involve Neural Networks or rule-based systems.

Types of AI

Different players have similar definitions for AI types based on business usage and cognitive level or stages of readiness and development, in the following section we will get introduced to these definitions.

Microsoft has the following definitions (“What Is Artificial Intelligence? | Microsoft Azure”)

·???Narrow AI (NAI)

NAI is also referred as ‘Weak AI’ and it is the most developed and advanced AI by Human. NAI coordinates and processes several data in almost real time under pre-defined frameworks without involvement of continuousness or emotions.

NAI applications examples can involve Autonomous driving, digital assistants (ALEXA), recommendation engines (NETFIX), Chatbots, Voice and text recognition. (Marr)

Gen AI is considered as Narrow AI as it is still missing the common understanding and can’t handle situations outside of trained data. (Mucci)?

·??General AI (General AI)

General AI is refereed also as “Strong AI”. Gen AI attempts to act as Human like intelligence that can self-learn. (“What Is AGI? - Artificial General Intelligence Explained - AWS”).

General AI shall be able to solve complex problems and make judgments through incorporating previous knowledge into its current problem-solving reasoning. General AI can create and imagine, not limited to specific scenario which is different than Narrow AI. (“What Is Artificial Intelligence? | Microsoft Azure”)

Examples of AGI (Artificial General Intelligence) can include advanced customer service including emotional intelligence and response, also code generation for business functions and autonomous driving using connected vehicles and objects data analysis. (Mucci)

·???Artificial Super Intelligence (ASI)

ASI is a hypothetical concept where AI (software programming) can go beyond human min capabilities where it has cutting edge cognitive functions and developed intellectual thinking than human mind. ASI can help on innovations and advances related to drugs, energy resources and weapons (“What Is Artificial-Superintelligence? | IBM”)

Google defines its AI types based on stages of development as (“What Is Artificial Intelligence (AI)?”)

·????Reactive Machines – where AI only reacts based on predefined rules and engines.

·????Limited Memory – This is where most of AI applications exists today. It uses memory to improve performance over time through data training models as deep learning.

·????Theory of Mind – it simulates the human brain which can make decisions, remember situations, emotions and reacts accordingly. Such advanced stage doesn’t exist today.

·???Self-aware – it is an advanced stage over “Theory of mind” where machines recognize its existence and has intellectual and emotional intelligence. This also doesn’t exist as of today.

?

AI Training Models

Before Jumping into training models, it is important to understand the difference between AI, Machine learning (ML), Deep learning (DL) and Neural networks (NLP).

?

AI is the overarching technology for mimicking human activities or tasks including Neural Language Processing, Vision, Text to speech or Motion. While ML is using datasets to identify patterns, predict and perform tasks. ?ML models may include either Supervised, Unsupervised or reinforcement models. Deep Learning is a subset of ML where it involves Neural networks involved nodes that have statistical relationship mapping human mindset. (“AI vs. Machine Learning vs. Deep Learning vs. Neural Networks | IBM”)

Supervised Model involved more human involvement to investigate data training while labelled data is mapped to known output. Supervised Models can include:

·???Classification Models – Where dataset is grouped into predefined categorized. E.g. Customer Retention based on historical data.

·???Regression Models – Where different data inputs to the equation that are weighted based on outcome impact. The output is usually numerical value. E.g. Airfare ticket pricing which depends on schedule, destination, Forex, weekday…etc.!

Unsupervised training model involves running models and discover outputs that were not explicitly stated. Unsupervised Models include (Salian, “NVIDIA Blog: Supervised vs. Unsupervised Learning”):

·???Clustering Model – where datasets are grouped together based on trend. E.g. Customer Segmentation

·???Dimensionality Reduction – where model reduces number of input parameters into a dataset minimizing redundant parameters that may over impact on the output. ?

·???Anomaly Detection – This is to detect unusual behaviors or patterns in dataset. E.g. Detection fraud cases in banking systems.

·???Association – Where a dataset feature may correlate with other features and give recommendations accordingly. E.g. Shopping carts in Mobile Applications recommending certain products based on your in-cart products.

·???Autoencoders – Where data is input to generate code then try to recreate the data from this code. This can be used to remove noise and ambiguities in data as in images, videos, and medical CT.

Semi-Supervised Training Model where a training dataset can include labelled or unlabeled datasets. This is used when it is time consuming to label all dataset or extracting features from all datasets is not possible.

?

Reinforcement model where the model interacts with surroundings, collect feedback (reward or penalties) and feedback accordingly. It usually has agent have measure the impact of applying the model on surrounding environment and based on the output it feedback reward or penalty. E.g. self-driving cars and computer gaming.

?Deep Learning (DL) is a subset of Machine Learning where DL requires more resources, larger datasets, and higher costs consequently to run. DL is mainly used with unlabeled data where high level of abstraction and analysis is needed on dataset to extract its features and get relationships as in social media analysis for users’ sentiment. (“Deep Learning vs Machine Learning - Difference between Data Technologies - AWS”)

?DL architecture and design is replicating how human brain is working with neurons represented as Nodes. DL will include at least 3 layers of Nodes including input and output layers. ML will usually use Mathematics and statistics to predict output while DL will use Mathematics, Statistics and Neural Networks to solve the problem. Data flows progressively from input layers to following layers where each neural network layer assign weight to each feature. Examples of DL trainings models may include convolutional neural networks, GANs and autoencoders. (“Deep Learning vs Machine Learning - Difference between Data Technologies - AWS”)

??

Different types of Artificial Neural networks may include (“What Is Artificial Intelligence (AI)?”):

?·????? Feedforward Neural Networks (FF) – Where data flow in a forward manner from one layer to subsequent layer until output is achieved. Usually, the model is paired with error correction algorithm called “Backpropagation” where error value is then propagated backward to the neural network adjusting the weight per neuron.

?·????? Recurrent Neural Network (RNN) – RNN is different than FF model that it uses time related data or sequencing which is different than FF which depends on weighting. RNN keeps a memory or what happened in the previous layer for adjusting the output of current layer. Example of that is speech recognition and translation.

?·????? Long/Short Term Memory (LSTM) – it is a similar form of RNN but in advanced form where the model can memorize what happened in previous several layers. Same as RNN it is used for speech recognition and market predictions.

?·????? Convolutional Neural Networks (CNN) – it is a common neural networks model that is used for image recognition which uses different distinct layers including a convolutional layer followed by pooling layers that filter different parts of the image then connecting them back together in the connected layer. Convolutional layer will look into simple features in the image including colors and edges before analyzing other complex features of the image.?

·????? General Adversarial Networks (GAN) is a popular training model with a small dataset. GAN is an example of two deep learning networks competing to outsmart each other. One model will usually create new datasets mimicking training data. While other model, disclaimer, will evaluate these data if they are real or fake. As the disclaimer gets better in identifying the real vs fake data also generator gets better on generating the data. An example of that in medical CT scans where trained radiologist can go through small amount of data to label it but deep learning will benefit of this small data to improve its accuracy. (Salian, “NVIDIA Blog: Supervised vs. Unsupervised Learning”)

References

Laskowski, N. and Tucci, L. (2022).?What Is Artificial Intelligence (AI)??[online] TechTarget. Available at: https://www.techtarget.com/searchenterpriseai/definition/AI-Artificial-Intelligence .

Google Cloud (2023).?What Is Artificial Intelligence (AI)??[online] Google Cloud. Available at: https://cloud.google.com/learn/what-is-artificial-intelligence .

AWS. “What Is Artificial Intelligence? - Artificial Intelligence (AI) Explained - AWS.” Amazon Web Services, Inc., 2024, aws.amazon.com/what-is/artificial-intelligence/ .

“What Is Artificial Intelligence? | Microsoft Azure.” Azure.microsoft.com , 29 June 2024, azure.microsoft.com/en-us/resources/cloud-computing-dictionary/what-is-artificial-intelligence#self-driving-cars .

S, Sneha, and Beschi Raja. “A Conceptual Overview and Systematic Review on Artificial Intelligence and Its Approaches.” Social Science Research Network, 14 Dec. 2019, ssrn.com/abstract=3519180 .

Marr, Bernard . “What Is Weak (Narrow) AI? Here Are 8 Practical Examples.” Bernard Marr, 2 July 2021, bernardmarr.com/what-is-weak-narrow-ai-here-are-8-practical-examples/ .

“What Is AGI? - Artificial General Intelligence Explained - AWS.” Amazon Web Services, Inc., aws.amazon.com/what-is/artificial-general-intelligence/ .

“What Is Artificial-Superintelligence? | IBM.” Www.ibm.com , www.ibm.com/topics/artificial-superintelligence#:~:text=Artificial%20superintelligence%20(ASI)%20is%20a .

Mucci, Tim. “Getting Ready for Artificial General Intelligence with Examples.” IBM Blog, 18 Apr. 2024, www.ibm.com/blog/artificial-general-intelligence-examples/ .

“Discover the Differences between AI vs. Machine Learning vs. Deep Learning.” Simplilearn.com , www.simplilearn.com/tutorials/artificial-intelligence-tutorial/ai-vs-machine-learning-vs-deep-learning .

“AI vs. Machine Learning vs. Deep Learning vs. Neural Networks | IBM.” Www.ibm.com , 15 Apr. 2024, www.ibm.com/think/topics/ai-vs-machine-learning-vs-deep-learning-vs-neural-networks#:~:text=AI%20is%20the%20overarching%20system . Accessed 18 July 2024.

Salian, Isha. “NVIDIA Blog: Supervised vs. Unsupervised Learning.” NVIDIA Blog, 2 Aug. 2018, blogs.nvidia.com/blog/supervised-unsupervised-learning/ .

“Deep Learning vs Machine Learning - Difference between Data Technologies - AWS.” Amazon Web Services, Inc., aws.amazon.com/compare/the-difference-between-machine-learning-and-deep-learning/#:~:text=ML%20solves%20problems%20through%20statistics