Machine Learning| Introduction First step to AI

Artificial Intelligence (AI) refers to the simulation of human intelligence in machines that are designed to think, learn, and solve problems like humans. AI encompasses a broad range of technologies and approaches aimed at making computers capable of performing tasks that typically require human intelligence, such as understanding language, recognizing patterns, and making decisions.

• ANI (Artificial Narrow Intelligence) : These AI do one things such as smart speakers, self-driving cars, AI web search etc. These AI systems address one type of scenario only.
• Generative AI : Things like chatgpt, Bard, Midjourney, Dall-E. These are works on general purpose. These helps as brainstorming, summarisation, works as assistant.
• AGI (Artificial general intelligence) : This is goal of building ai for intellectual tasks, that a human can do. Can be better than humans

We are going by baby steps toward AGI. We are more far away from AGI. Lets kepp it to future and now lets see how we are going to apply AI to our projects.

Machine Learning

In supervised learning what we try to map is Inputs to Outputs.

• Spam Filter : Input is email and output is spam(0/1)
• Speech recognition : Input is audio file and output text transcript
• machine translation : Input English language and Output Japan language
• Online Advertising : Input Add, User info and Output user click(0/1)
• Self driving car : Input image, radar info and Output Position of other cars
• Visual Inspection : Input Image of phone and Output defect product(0/1)

LLM(Large language models) like chatgpt, copilot is also at heart it use machine learning so if we give one or two words this can generate next words. This repeatedly predict what is the next word.

How LLM works

LLM built by using supervised learning to train a model to repeatedly predict the next word. For example if AI system read a sentence online like My favourite drink is Bubble tea. Then this single sentence is turned into lot of inputs and output data points for the model to learn to predict next word.

SO model is trained to predict the next word by giving few words. So if you have billion or trillion of words you could train a model like chatgpt 😈. But in here we have to use technical improvements to follow instruction, disclose inappropriate content and wrong informations

DATA

How Data amount VS Performance improve

How to acquire data

• Manual labeling : Collect data like images and label them(cat | Not cat)
• Observe behaviours(People, machines etc) : Sells in shopping center. Info — Userid, purchases, date, province, total amount
• Download from websites/Partners

Once you start with some data you can feed them to ML model. And you can get idea what is the data you need and what type of data model would perform better. You can do modification such as collect data of machine by 10 minutes time laps than 1 minute time gaps. But more data is not usually better. It would downgrade the model some times.

Data Problems : You may have

• Incorrect labels (human errors)
• Missing values
• Multiple data types (mp3, mp4, jpg, png etc)
• Unstructured data

Supervised Learning Models can work both of Structured | Unstructured Data.

AI Terminologies

Machine Learning : Field of Study that gives computers the ability to learn without being explicitly programmed. Arthur Samuel

Data Science : Science of extracting knowledge and insights from data.

Deep Learning : We have inputs and set of Neural Network layers and output

How to use AI in your Projects

If you have a shoping mall and you have a website. Does this enough to make profits. What is different between amazon and you. 🤔

So lets look at how to improve our shoping mall

Internet Era

• We can have 2 different websites at see which attract most customers and do changes time to time.
• We can do short iterations such as new product launches.
• Decision making push down to Engineers and Specialized roles.

AI Era

• Strategic data acquisition : We launch products to use for data acquisitions not only money products. Large companies have 2 or more products few for profit making and others for data acquisitions.
• Unified data warehouse : Rather than different data bases it is better to have data warehouse that connects each data.
• Pervasive automation : Better at using automation opportunities. Hey lets insert supervised loearning algorithm and have ATP mapping. So no need human involvement.

Steps you can apply in your company| Project

1. Execute pilot projects to gain momentum
2. Build an in-house AI Team
3. Provide broad AI training
4. Develop an AI strategy
5. Develop internal and external communications

Strengths and Weaknesses of Machine Learning

ML tends to work well :

• Simple concept
• Lot of data available

ML tend to work poorly :

• learning complex concept from small amount of data
• Perform poor on new types of data (X ray images are fliped or resized)

So as conclution ML is not working good if the prediction is more ambiguous. Like ML is not able to identify human intensions or human thoughts. AI is not able to acurate stock market predictions because it does not heavily depend on previous data.

Hope you would gain basic knowledge on how to apply AI to your organization.

References : Deeplearning.ai