A Recap on Artificial Intelligence

Developing machines and algorithms that imitate the cognitive capabilities of the human brain has been a prime focus of researchers and scientists in the last decade.

This imitation is starting to emerge with the advent of applications in the field of artificial intelligence (AI). Many applications today replicate in some way or another the function of the biological human brain through the use of neural networks. These interconnected artificial neurons can process billions of input data in seconds, a mechanism that is almost impossible to replicate in the natural human brain. The beauty of these neural networks is that they are capable of continuously learning and adapting to new data, new environments, and new challenges. However, these neural networks still lack a general understanding of all subjects as they tend to focus on specific tasks, and they lack consciousness as well. Those two elements might be the factors that are preventing us from yet reaching artificial general intelligence or AGI.

Artificial intelligence (AI), neural networks, and machine learning represent interconnected but distinct aspects of the field of computer science. Artificial intelligence (AI) is the general and broader discipline that allows computers to perform tasks such as reasoning, learning, and understanding natural language. Machine learning is a subset of AI focused on algorithms that can learn and that can make predictions based on given data. Neural networks are a subset of machine learning that are inspired by biological human neural networks. They are composed of layers of interconnected nodes or neurons that process the input data through a system of connections.

While early programming was constrained to the usual “if” and “then” concepts—if the question is about the car dashboard, then list the items present on the dashboard—neural networks and machine learning algorithms do not abide by this concept. These networks and algorithms are capable of analyzing, connecting the dots, figuring out patterns, and giving detailed analyses.

The implementation of AI in various domains is moving at a fast pace, and its adoption is not limited to a couple of industries. Today, AI is utilized in many sectors, such as retail, health care, transportation, manufacturing, and finance. Businesses are taking advantage of AI, machine learning, and advanced predictive analytical systems to engage the consumer, enhance their consumer experience, and maximize profit.

Below are some examples of the use of AI in the various sectors and industries:

• Retail: This industry might represent one of the best sectors for AI. For example, AI is used to analyze customer behavior, habits, trends, preferences, and purchase history for online e-commerce websites. Chatbots are today common for any online website; these chatbots act as a customer service first point of contact. AI is also utilized for inventory management, product development, demand forecasting, and stock optimization.

• Health care: This industry is picking up fast and starting to implement AI in research and development, and in personalized applications. Many laboratories today utilize AI to detect viruses, tumors, and abnormalities in medical imaging scans. Laboratories can detect and advise on whether a patient is diabetic or not for example. At the same time, we are seeing an increase in the production of mobile applications that tackle health. Some applications are capable of monitoring heart patients, detecting anomalies in the heart, and sending signals to available emergency doctors. Other mobile applications give you tips and advice on how to stay healthy based on connected sensors.

• Manufacturing: The rise of AI-powered factories is on the horizon. Digital AI twins can replicate any factory and its functions on a digital platform and then analyze all the processes. By incorporating AI, digital twins are capable of predicting outcomes, diagnosing issues, and optimizing processes in a simulated environment. In addition, sensors installed inside the factory can monitor, alert, and optimize processes in real time. These advanced capabilities allow for a deeper analysis and understanding, enabling engineers to make informed decisions and to implement improvements without the risk and cost associated with real-world testing. To improve operations and maintenance procedures, General Electric’s Gas Turbine Power Plant in France installed a digital twin system. With the aid of this AI system, engineers are gathering real-time data from sensors installed in the turbines, analyzing it, and offering suggestions for proactive maintenance and performance improvement. AI-powered robots are being utilized more and more in factories as well. They are capable of working autonomously and they are also capable of lifting and displacing heavy material without having the risk of injury.

• Finance: Today, we can’t imagine the financial sector operating without AI tools. Banks are using sophisticated AI and machine learning algorithms to detect fraud. These algorithms are capable of detecting and noticing abnormalities and unusual patterns that the human eye can easily miss. AI can go through millions of documents and scrap the web to advise banks and consultants on investment opportunities. The use of forecasting tools in trading and in the stock market is gaining momentum.

• Transportation: Autonomous driving cars are already present and undergoing testing processes in many countries. AI-powered taxis might be the first service to take off, while autonomous copters and AI-powered drones are already being utilized and implemented worldwide. But AI doesn’t stop at autonomous driving and goes way beyond that. Optimizing routes for shipping and logistics, improving delivery lines and timings, and optimizing fuel consumption are but a few things that are possible today.

The basic concept of machine learning is learning by finding meaningful patterns and structures from the collected data through the use of mathematical algorithms; Algorithms are a sequence of instructions commanding computers to execute specific tasks using logic and arithmetic operations.

As Pedro Domingos puts it in his book, "The Master Algorithm:" Machine learning is the scientific method on steroids.

Machine learning allows data scientists and machine learning engineers to quickly reiterate, refine, and test hypotheses that took weeks and months using traditional scientific methods.

While machine learning finds meaningful patterns in data, analytics and algorithms focus on finding correlations between the different variables in the data to understand how one variable might affect another in a positive or a negative way. Learning analytics use techniques like data visualization, tables, and graphs to help the viewer gain insight into the data.

The fast development of machine learning enabled the emergence of the field of natural language processing (NLP). NLP equips computers with the ability to understand text and spoken words in a natural way, the way we communicate with each other. The applications of NLP are vast and rapidly growing, and they are being integrated into business tools such as customer management platforms.

At the intersection of AI, NLP, and LLMs are chatbots which we will discuss and develop in this course. Developing a customer support chatbot for car owners is based on the principles of artificial intelligence and natural language processing with an emphasis on large language models (LLMs). LLMs enable machines to understand and generate human-like text, and this capability is essential for creating chatbots that can understand natural language input from users, interpret their queries, and generate relevant responses. The advantage of using LLMs in chatbots lies in their ability to process and understand huge amounts of data, enabling them to respond to queries in seconds. This allows LLMs to provide assistance in an intuitive manner, thus enhancing the user experience.

LLMs are often referred to as foundation models because of their versatility and adaptability. Foundation models are trained on vast amounts of data, which, once trained, can be fine-tuned for a variety of downstream tasks, such as text summarization, question-answering, translation, and much more. This method of fine-tuning is also referred to as transfer learning because the downstream task learns from the primary model, but it enhances and improves its own capabilities by specializing in a particular task. With the process of transfer learning, the specific tasks, such as text summarization, act as the learner who learns from the main LLM model that has been trained on more general tasks. This is the most efficient way of learning today.

Moreover, fine-tuning LLMs with domain-specific data is an essential step in tailoring the customer support chatbot to answer car-related inquiries. This process involves augmenting the database of the LLM with task-specific knowledge, which we refer to as retrieval-augmented generation (RAG). It also involves prompt engineering, a method that is utilized for instructing the LLM to behave in a certain manner to steer it in the right direction for generating the desired output. By augmenting the LLM with task-specific data, the chatbot becomes capable of recognizing the context and specifics of the chosen car automotive queries, leading to more accurate and relevant responses. The benefits of using RAG with LLMs are many, as they enable developers to enhance the LLM response, customize the style of the output, steer the conversation between the customer and the LLM in a certain manner, and provide a seamless experience for the customers.

NLP is heavily used in many industries, and we don’t even realize it. Some of the common NLP use cases are as follows:

• Speech-to-text: NLP enables users to convert spoken language into text, while text-to-speech transforms text into spoken words with voices that are almost natural. There are many benefits and uses for such services, including hands-free type applications that are present in many applications today, reading text for visually impaired users, and utilizing such services for automated bots to give them a human sense or approach.

• Machine translation: It helps with translating languages across cultures and dialects. The progress achieved in this field is huge. Today, we are way beyond Google Translate’s simplistic, noncontextual word translation. Large language models and neural networks are capable of deciphering and contextually translating languages with great accuracy.

• Text classification: NLP helps classify text and sentences into categories. A great example of text classification is sentiment analysis. The classification algorithm can detect the nuances and polarity of the text by classifying it into positive, neutral, or negative intent. Classification algorithms can also filter emails into spam or not spam, making sure that spam emails are discarded or blocked. Topic classification and intent recognition are other uses of text classification, whereby the algorithm classifies the text into categories or intents, such as fashion, technology, and education. Text classification can also identify and classify languages.

• Entity extraction: A large and important topic in NLP, the main task of entity extraction or named entity recognition (NER) is to identify and extract fields or predefined categories from text such as names, dates, locations, organization names, and name of persons. This is beneficial in tasks such as extracting text from invoices, receipts, and official documents.

• Text summarization: It helps generate a concise summary of the text. Its usage is simple but highly needed, as we can quickly generate summaries of large amounts of text to decide whether the text has the information we are looking for or not.

• Question answering: As the most demanded AI service today, this is the main task of chatbots that are currently utilized in e-commerce websites, retail venues, and educational websites. The chatbots are meant to respond to users in a natural language as if it is the customer service representative of the company or the institution. Chatbots also act as personal assistants by replying to users based on their profile, information, habits, and preferences.