AI Terms, You Should Know

Photo by Usama on Pexels

AI is not a new phenomenon by means. Since the 1950's, Artificial Intelligence has been used in different industries. With the inception of Mainstream AI tools in 2022, the term is not exclusive to industry specialists and professionals. Let's understand this buzzword that has come to mainstream for the past year, thanks to breakthroughs and innovations in machine learning.

Here's a list of few technical terms related to Artificial Intelligence that you should know to expand your horizon. 

Machine Learning

The need to teach computers solve problems and identify solutions on their own is very challenging. Teaching them to learn patterns is Machine Learning(ML), a very important subfield of Artificial Intelligence.  Our daily lives are touched by the fruits of Machine Learning in the form of Speech Recognition on Google Assistant, Apple's Siri and self driving cars. Machine Learning Systems currently require rigorous effort and deep knowledge of machine learning techniques. In the recent years, general public is brought in contact with the machine learning applications. Billions of images are scanned by Google's machine learning tools so that when you use Google lens in your camera app, it recognize almost anything. For example, you see a clothing item, and you want to search where you can find it, you take a picture of its tag, it will most likely find the stores nearby you, where you can buy it. 

Machine Learning can be supervised, semi-supervised, unsupervised and reinforcement.

In Machine learning, the decisions are made by dividing the information into numerous sets of similar objects and each set is separated with other by a boundary which is called decision boundary. Machine Learning systems compares the query with every set and reach at the verdict by using the decision boundary. 

Traditional AI solves specific problems which are done by the help of handcrafted rules.

Generative AI

Humans are not the soul creators of Art any more, Generative AI can harness the power of Artificial Intelligence to create text, music, images and audio.    

Imagine a paintbrush that doesn't just mimic reality, but births entirely new colors and textures. That's the magic of Generative AI. This branch of AI, powered by machine learning, devours vast data landscapes to learn the language of creation. Freed from mere imitation, it paints with the brushstrokes of originality, crafting hyper-real video game worlds, composing symphonies worthy of Beethoven's envy, and much more. From art to science, Generative AI rewrites the rules of creativity, shaping our future with unimaginable possibilities. 

 

Deep learning

Human brains are constantly matching patterns and finding similarities between obscure concepts. When a physic numerical problem is posed to a higher secondary student, his brain goes through a lot of layers of concepts and previous knowledge bases to bring out the necessary mathematical techniques, which would assist him solving the said problem. Those concepts and mathematical rules were not specifically learned to solve this particular problem, but they were acquired by him so that it could be used when needed. 

Similarly, in Artificial Intelligence, Deep learning is the process of incorporating multiple layers of knowledge bases which would help to bring out answer when the input is received. Generally, more than one layer between input and output referred as the Deep Learning neural networks.

Deep Learning refers to neural networks which use more than one layer between input and output.  In deep neural network learning (or, simply, deep learning), most model parameters are learned not directly from the features of the training examples, but from the outputs of the preceding layers.

 

Artificial General Intelligence (AGI)

Recent Development in Artificial Intelligence have led the researchers to believe that it is possible to see the Human-like artificial Intelligence in the future. We can't be sure about the timeline of the future events but it's looking promising at the moment. Cognitive ability of Humans in computers would be some sight to behold. 

AGI is an Artificial Intelligence systems which is comparable with human intelligence in a wide array of cognitive abilities and it is able to perform logical, rational decision making after considering benevolence of humans.

  

Natural Language Processing (NLP)

NLP is the process of adding cognitive abilities to the computers to bridge the Gap Between Machines and Humans. Natural language processing (NLP) is a crucial subfield of artificial intelligence (AI) focused on enabling computers to understand, interpret, and manipulate human language. 

It's essence is bring human creativity, ambiguity and ambivalence close to a machine's logical, rational and mathematical abilities. 

NLP performs these following tasks

Understanding Text 

Text understanding, which forms the very core of NLP, refers to the ability of computers to extract meaning and context from written language. It's not merely about deciphering individual words, but rather about grasping the deeper intent, relationships, and nuances within a text.

Text Generation

Writing coherent and cohesive paragraphs of text to convey meaning is one of the finer human skills. In Natural Language Processing, text generation refers to producing human understandable passages of text that are proper in meaning and context. Text Generation under NLP enables computers to compose poems, short stories and new articles.   

Speech Processing

This narrows the gap between written words and spoken language. One of the key components of Speech Processing is Automatic Speech Recognition (ACR) which is responsible for understanding the spoken words using the acoustic models to decipher audio signals. It also uses language models to formulate word sequence in a sentence. 

Multiple speakers are differentiated and sorted out with the help of speech processing. AI can now recognize emotions and categories them from an audio by speech processing technologies. 

Dialogue Systems.

Dialogue Systems is a subfield of Artificial Intelligence, which can hold conversations with human in a truly natural and human like fashion. They are often known as chatbots and conversational AI as well. 

Large Language Models

Large Language Models rely on Machine Learning to adapt, improvise and overcome the nuances and intricacies of human speech and produce the output in the form of natural language. Progress in the development of LLM(Large Language Models) will be inflection point in diminishing the gap between human and computer intelligence. Large Language Models are intelligent enough to work with clues and layers of data sets so that they could compose a Rhyming Poem, do homework for you or even solve riddles for you. Although LLM's don't have any emotions or real thoughts, the imitate human behaviours through pattern recognition and deep learning through large sets of data. 

 

Hallucinations

Despite all the progress in the field of Artificial Intelligence, current Models are still prone to mistakes, which results because their inherent lack of differentiating between truth and lies. These are called Hallucinations by the researchers. Humans have the instinct of knowing something is wrong when they even don't know what is actually wrong in that situation. Machines are not capable of that yet. A model has to be trained on the most recent data to avoid these fabrications from happening and developers try to eliminate the hallucinations by providing extra information by trusted sources. This process is called Grounding and all the newer AI models are trained this way now. 

 

Responsible AI

A Responsible AI is an ethical, legal and transparent Artificial Intelligence that doesn't exploit privacy and data of its users. It's supposed to be centred around humans so that it could interpreted and understood by humans. Developers and Data scientists are in control of producing unsusceptible code that is not prone to any tempering by any harmful agents. Through responsible AI, society can move forward instead of it becoming a nuisance.    

Multimodal Models

Teachers often teach mathematics to little children by asking them to visualise the numbers. Once they can equate numbers with real objects, they can do addition, subtraction more easily by just counting the objects in their head, which is a lot easier than just numbers, which is just an abstract concept. 

Artificial Intelligence use multimodal models to incorporate various modes of information like images, numbers, audio, video and code to prepare an answer which is best suited for your query. 

Multimodal Models are fusion of neural network and each network is trained on a specific type of data like audio, video, images, text, code etc. For example, if you give it a video of a cute animal and ask him to describe the video in a creative fashion, at least two models will study the visual mannerisms of the animal and the sounds of the animal. They are both trained on different types of data but they work together to understand the whole picture. In the end, they transfer their understanding to the writing model which would describe the video for you. 

Copilots 

Copilots are now assisting the developers, writers, coders and animators in all sorts of niches and intricate professional tasks. 

 

No longer mere assistants, copilots rise as digital companions, weaving their intelligence into the fabric of applications. Large language models empower them to grasp your intent, craft content, and navigate tasks with uncanny ease. Built with ethical guardrails, they serve as trusted co-pilots, amplifying your prowess and propelling you towards greater digital mastery.

Prompts

When you type in any chatbot or AI model, you give the instructions in the human understandable language and the AI models generates responses on the bases of that. It's very different than typing commands in the terminal or clicking buttons in Graphical User Interface on your screen, because that requires a specific syntax in the case of Terminal/CLI(command line interface) and even in GUI, you are actually giving it tapping/clicking buttons. Prompts give you freedom and it's much more than searching an answer on the web. AI understands the prompts and then work on it instead of giving you just the website address and then you have to decide which article to read or which video to watch. 

 

A prompt is the question, suggestion, direction, or instruction that is given to the AI, it could be in any form but right now it's mostly in the form of written prompts. 

The effectiveness of a prompt depends on several factors, including:

• Clarity and specificity: The clearer and more specific the prompt is, the better the bot will understand what you want it to do.
• Context: Providing context in your prompt can help the bot generate a more relevant and accurate output.
• Bot's capabilities: Different bots have different capabilities, so it's important to choose a bot that is able to handle the type of prompt you want to give it.