Artificial Intelligence
Machine Learning:
a subset of AI, behind the AI bot
trains systems to learn from data and make decisions/ predictions based on trends
4 Types of machine learning:
Supervised Learning
Unsupervised Learning
Reinforced Learning
Transfer Learning
Supervised Learning:
Task Driven - Train system on labelled data to identify images or determine values
A human trains and test the machine using labelled data
Unsupervised Learning:
Data Driven – Uses deep learning to find patterns in unlabeled data
Rather, machine identifies patterns in the unlabeled data.
Once clusters and patterns are identified e.g., cluster 1, cluster 2, cluster 3, the machine creates its own algorithms to evaluate and interpret new incoming data as e.g., cluster 1, cluster 2, cluster 3.
Used to determine similarities in groups (e.g., patients who may respond better to one treatment or another) or abnormalities in data (e.g., fraud detection)
Reinforcement Learning:
Learns from a system of trial and error with rewards and punishment – seeks maximum reward
Example of Reinforcement Learning:
A self-driving car, where the vehicle learns to navigate complex traffic situations by taking actions based on its current environment and receiving rewards (like avoiding collisions) or penalties (like hitting a curb), gradually optimizing its driving behavior through trial and error in a simulated environment
Transfer Learning:
Utilizes Neural Network that is trained on a large dataset and retains that knowledge for future use.
Example of Transfer Learning:
Google Teachable Material
With transfer learning, we can now add additional data to train the model to a new, specific task. This requires only a small change in the inner workings of the already trained model.
Deep Learning:
a subfield of machine learning that uses neural networks with multiple layers to learn and extract features from data.
AI
machine’s ability to perform some cognitive function we usually associate with human minds
AI Bot
software robot that has AI capabilities that is trained on vast amounts of unstructured data to generate new responses to user prompts
trained to understand what is being asked and respond in a natural language approach that may vary depending on the customer
Types of AI
Classification AI
Generative AI
Classification AI
Analyzes data and makes predictions. For example, a classification AI system can be trained to recognize and classify images of trees and flowers (or cats and dogs).
Generative AI:
Creates new content like images, text, music, and computer code.
Generative AI systems learn patterns from training data and use them to generate new content.
For example, a generative AI system can learn from images of trees and flowers and generate new images that resemble them.
Neural Network:
a type of machine learning algorithm that mimics the structure and function of the brain
Goal of Neural Network:
Neural Network allows AI systems to learn and process complex data.
Artificial Narrow Intelligence (ANI)
AI that specializes in 1 area and solves that 1 problem (extremely specific)
Example of ANI:
Chatbots and virtual assistants (Siri)
Artificial General Intelligence (AGI)
Compared to ANI, it incorporates more human behaviors (the ability to interpret tone and emotion) to the point that it can perform on par with humans
Artificial Super Intelligence (ASI)
Strong AI
Software-based AI system with intellectual scope beyond human intelligence in terms of cognitive functions and thinking skills
Limitations of AI:
Data is heavily dependent on quality that can be difficult to source and manage
Judgement and reasoning may limit the AI’s ability to handle unstructured & ambiguous situations
Versatility:
Lacks the flexibility and adaptability of human intelligence and can only perform tasks they have been trained for.
Steps of Supervised Learning:
Train the AI model through a corpus of pre-labelled images
Test the AI model and improve accuracy
Apply the AI model
Corpus:
training data that may be curated based on subject area experts
information - text, videos, images regarding to the subject area in concern
Large Language Models:
A specific application of AI
Deep learning models that utilizes extensive text datasets (internet/ books) to learn patterns & relationships between words and phrases
3 main challenge enterprise adoption of AI:
Data Problems - data is in different format/ varying quality which affects AI as AI demands real-time data.
Infrastructure - AI solutions are costly as they require higher computing power.
Hallucinations - outputs of AI that deviate from facts and contextual logic or are wrong
How to reduce hallucinations
Provide clear and specific prompts
Active mitigation strategies - settings that control parameters of how LLM works during generation
Multi-shot prompting - providing the LLM with multiple formats and examples
Small language models
smaller problem-focused datasets that are high quality - company managed.
Addresses privacy and security issues