AI Natural Language Processing (NLP)

• AI can analyze, process, and generate speech or text.
• Natural Language Processing (NLP) focuses on enabling computers to use human languages in written and spoken forms, similar to human capabilities.
• NLP techniques are applicable to coding and music.

Major NLP Capabilities of AI

• Named Entity Recognition (NER): Extracts and classifies names of persons, places, and companies into predefined categories.
• Topic Modelling: Identifies hidden topics within large document collections.
• Text Categorization: Sorts text into specific taxonomies.
• Text Clustering: Groups text or documents based on content similarities.
• Sentiment Analysis: Identifies, extracts, and analyzes affective states and subjective information.
• Summarization: Generates concise versions of input documents, retaining key points.
• Information Extraction: Locates meaningful information within unstructured text.
• Entity Resolution: Identifies records in data sources that refer to the same real-world entities and their relationships.
• Translation: Converts text from one language to another, preserving meaning.
• Speech Recognition: Converts spoken language into text.
• Speech Synthesis: Converts text into spoken language.
• Natural Language Generation (NLG): Transforms data into human language.

Named Entity Recognition (NER)

• Can be tested using the "displaCy Named Entity Visualizer" online demo.
• The demo allows users to input text, select a language, choose entity types to label, and then see how well the model identifies those entities.

Sentiment Analysis

• Involves detecting polarity (positive/negative), emotion (angry/happy/sad), urgency, and intention in text or speech.
• Brands use it to analyze customer feedback from surveys and social media to improve products and services.
• Online demos like Lexalytics highlight sentiment-indicating words in analyzed text, showing sentiment degrees and topics.
• Lettria’s Customer Sentiment Analysis tool on Hugging Face determines if sentiment is positive, neutral, or negative.

Summarization

• Extractive models select and combine existing phrases from the input document to create a summary.
• Abstractive models generate a summary using abstracted content.

Information Extraction

• AI NLP extracts useful information from large datasets (Big Data) that are too large for traditional analysis methods.

Entity Resolution

• Involves identifying records that refer to the same real-world entity, even if the records are different.
• Entity resolution can be used to detect fraud, improve risk assessment and compliance, improve customer insights, and reduce false positives/negatives.

Translation

• Traditional machine translation used rule-based approaches relying on dictionaries and grammars but requires significant manual effort.
• Neural networks provide superior translation performance compared to statistics-based models.
• Google Translate started using neural network-based translation models in 2016.

Speech Recognition

• AI Singapore, in collaboration with NUS and NTU, developed the Speech Lab engine, which recognizes conversations in multiple languages (e.g., Singlish).

Speech Synthesis

• Google's DeepMind created WaveNet in 2016, which generates realistic human-like voices.
• This tech can imitate a person’s voice.

Natural Language Generation (NLG)

• Transforms data into human language.
• Large language models (LLMs) are powerful NLG AIs that contain language information extracted from massive amounts of data.
• Transformer models are currently the most popular type of LLMs.

Further Applications of NLP

• Chatbots conduct written or spoken conversations in natural languages.
  • ChatGPT from OpenAI can adapt to the style and content of the prompt.
  • Microsoft integrated Copilot, powered by the GPT family of LLMs, into its 365 apps and Windows OS.
  • Google released Bard (now Gemini), Meta released LLaMA, Baidu released ERNIE bot, Alibaba released Tongyi Qianwen (Qwen), and Anthropic released Claude, all with ChatGPT-like capabilities.
• Virtual Voice Agents: Google Assistant, Apple's Siri, and Amazon's Alexa.
• Code generation: LLMs can generate and translate computer code.
• Music Generation: WaveNet can synthesize audio signals like music, and AI Duet by Google is a music "chatbot".

Current Challenges in NLP

• Voice recognition struggles with different languages and dialects.
• ASEAN languages lack sufficient corpus data.
• AI lacks true language understanding, limiting question answering.
• Sentiment analysis is affected by the multiple meanings and implications within text.
• Many NLP systems require additional training.
• LLMs require significant computer power and can be slow.
• Current LLMs may "hallucinate" and produce inaccurate responses.
• AI is capable of analyzing, processing, and generating images and videos.

Major Vision Capabilities of AI

• Text recognition: Converts images of text into machine-readable text (OCR).
• Object recognition: Identifies the presence and location of objects in images/videos.
• Facial recognition: Identifies and verifies people using facial features.
• Action recognition: Identifies categorized actions and when they occur in videos.
• Visual question answering: Answers questions about images or videos.
• Image/video generation: Creates new images and videos.
• Image/video processing: Enhances resolution (super-resolution), adds colors to black-and-white images (colorization), and transfers styles.

Text Recognition

• AI text recognition is more flexible than traditional OCR.
• AI can pass traditional CAPTCHA tests.
• Text recognition is useful for digitization.

Object Recognition

• Types of object recognition: classification, semantic segmentation, object detection, and instance segmentation.
• YOLO (You Only Look Once) is a popular object recognition model known for speed and accuracy.

Facial Recognition

• Systems like Google's FaceNet accurately verify, recognize, verifying, and comparing faces.
• AI systems can recognize faces even when wearing masks.

Action Recognition

• One way to recognize action is to first recognize pose.

Visual Question Answering

• Systems like ChatGPT and Google’s Gemini can answer questions about images.

Image/Video Generation

• AI can generate high-quality images from text descriptions.

Image Processing

• AI can enhance the resolution of images, add colors to black and white photos, etc.

Further Applications of AI Vision

• Reverse image search
• Games, augmented reality (AR), virtual reality (VR), and the metaverse
• Deepfakes are synthetic videos that are digitally altered.

Vision Challenges

• Technical and ethical.
• A robot is a machine that can carry out complex tasks with little human intervention.

Major Capabilities of AI Robots

• Sensing: Detecting the environment using various sensors.
• Navigation: Finding location and planning routes.
• Physical Activities: Moving and handling objects.
• Interacting with Other Machines: Communicating with other machines.
• Interacting with Humans: Engaging with humans.
• Robotic Process Automation (RPA): Automating software processes.

Sensing

• Weight, contact, accelerometer, and gyrometer sensors
• Sound perception
• Depth cameras
• Hyperspectral sensors
• Radar, lidar, and ultrasonic sensors

Navigation

• GPS, Bluetooth, Wi-Fi, and mobile phone signals
• Google's Visual Positioning System uses cameras for more accurate indoor and urban navigation.

Physical Activities

• Autonomous vehicles use AI to process sensor data and determine actions.
• Robots walking on two or four legs.
• Drones.

Interacting with Other Machines

• Robots can communicate using local networks or the Internet.

Interacting with Humans

• Collaborative robots (cobots) work safely with humans in the workplace.

Robotic Process Automation (RPA)

• Automates mundane tasks in computer systems.
• AI allows RPA to handle semi-structured and unstructured data.

Further applications

• Robot cleaners.
• Robots can be used to chat with people.

Challenges

• Regulations on where robots can operate
• Behavior in unfamiliar situations may be hard to predict.
• Imitating emotions is still difficult.
• Key tasks for which AI is generally deployed: Perception, Notification, Suggestion, Automation, Prediction, Prevention, and Situational awareness.
• Adopting AI is key to the ongoing “Fourth Industrial Revolution”, aka “Industry 4.0”.

Data Analytics

• Converts raw data into actionable insights.
• Data analytics can be descriptive, diagnostic, predictive, and prescriptive.
• Personalization is a general application of data analytics.

Manufacturing

• Smart factories use connected devices, machinery, and production systems.

Agriculture

• Farming's labor crunch is seen to be a global problem.
• AI robotics allows farmers to control equipment remotely, optimize operations, and improve productivity.
• AI can be used to sort crops.

Education

• AI automation makes education more widely available.

Government

• Smart Nation projects in Singapore integrate technology into daily life.

Green

• AI can be used to optimize flight paths in order to save fuel.

Weather

• AI vision to detect the weather in specific locations.

Retail

• Facial recognition, speech analytics, and text analytics can be used to analyze customer behaviors.

Information Technology

• AI helps automate computer infrastructures.

Finance

• AI can extract information from complex documents.

Insurance

• Chatbots and NLG technology can help promote products.

Human Resources

• Chatbots can help, screen candidates, and help with the process of providing employee feedback.
• Key factors that enable AI: artificial neural networks, hardware acceleration, and Big Data.
• Key factors for AI availability today: open source code, low-code development, cloud computing, and edge computing.

Artificial Neural Networks

• The challenge
• Human programmers must give a computer each and every instruction precisely to get it to perform a task.
• Machine learning
• Using a (generally simpler) algorithm to find, within a specific class