AI4E V3 Module 3

Summary

This video script from 'AI for Everyone' explores the diverse applications of AI, showcasing projects by AI Singapore. It highlights AI's capabilities in object detection, post estimation for tasks like social distancing, and NLP for document understanding and spam detection. The script also discusses AI's role in Singapore's multilingual society, the importance of labeled data, and AI Singapore's programs to accelerate AI adoption in businesses, including the 100 Experiments and AI Makerspace initiatives.

Takeaways

• 🧠 AI has made significant advancements in object detection and recognition, making it easier to identify people, cars, bicycles, and animals with the help of correctly labeled datasets.
• 🏭 The success of AI systems relies heavily on the labor of workers who label images, often from countries like India, Vietnam, and Africa, highlighting the human effort behind AI's 'intelligence'.
• 📈 AI Singapore's computer vision team has enhanced AI models with heuristics to tackle more complex tasks, such as pose estimation which can understand a person's interaction with the environment.
• 👥 Pose estimation has been applied for social distancing measures during the COVID-19 pandemic, helping companies like HP maintain safety protocols in their factories.
• 💪 The AI system's capabilities have been extended to detect groups of people, supporting businesses in adhering to safety measures and classifying activities for exercise monitoring in local gyms.
• 📧 Natural Language Processing (NLP) is a prevalent AI task, with applications ranging from spam detection in emails to document understanding and automated code generation.
• 🌐 Singapore's multilingual environment poses a challenge for AI programs, but AI Singapore, in partnership with NUS and NTU, has developed an engine capable of understanding mixed-language speech, including Singlish.
• 🔍 AI Singapore's core programs, such as the 100 Experiments and AI Makerspace, assist companies in adopting AI by building custom solutions or providing near-ready AI solutions.
• 🛠️ The 100 Experiments program collaborates with companies to develop AI solutions when off-the-shelf options are not available, and the company intends to hire trained AI engineers post-project.
• 🛒 AI Makerspace focuses on creating AI solutions based on research, helping businesses, especially SMEs, to quickly adopt AI with minimal R&D investment.
• 🚀 AI Singapore's initiatives have facilitated various AI use cases, including predicting device return rates, fraud detection in insurance claims, and document classification for analysts.
• 📊 The script concludes by emphasizing the importance of exploring AI's potential responsibly, inviting viewers to engage with AI Singapore's resources and community.

Q & A

What is the main focus of Module 3 of AI for Everyone?

The main focus of Module 3 is to explore the capabilities of AI, showcasing examples of AI projects done by AI Singapore, particularly for companies in Singapore.

How has AI been utilized for object detection and recognition in the past decade?

AI has been successfully used for object detection and recognition in various scenarios, including identifying people, cars, bicycles, motorbikes, cats, and dogs, with the help of correctly labeled datasets.

What is the importance of having labeled images for training AI models?

Labeled images are crucial for training AI models as they provide the necessary data for the AI to learn and accurately identify objects such as people, cars, and animals.

Who are typically responsible for labeling images for AI training datasets?

The images are often labeled by low-paid workers in countries like India, Vietnam, and Africa, who manually draw boxes around objects and indicate what each box represents.

What is the role of heuristics in extending the capabilities of AI models in computer vision tasks?

Heuristics are added to AI systems to tackle more challenging computer vision tasks, such as pose estimation, which allows the AI to understand not only the poses a person is in but also their interaction with the environment.

How was pose estimation utilized for social distancing during the COVID-19 pandemic?

Pose estimation was used to develop an AI system for HP to ensure social distancing in their factory during the COVID-19 pandemic. The system was deployed on an edge device with a standard USB camera.

What is the purpose of the AI system developed for document understanding in NLP tasks?

The AI system for document understanding in NLP tasks is used to sort through documents, such as tax documents, and perform tasks like relation extraction, text summarization, sentiment analysis, and automated news generation.

How has AI impacted the productivity of handling spam emails?

AI has significantly reduced the time spent on handling spam emails by replacing the manual task with a modern spam detection engine, making the process more efficient and allowing users to focus on important emails.

What challenges do AI programs face in understanding multilingual speech in Singapore?

AI programs often face challenges in understanding speech that includes a mix of languages commonly spoken in Singapore, such as English, Tamil, Mandarin, Cantonese, Malay, Hokkien, and Singlish.

What are the two core programs by AI Singapore that help companies adopt AI?

The two core programs are the 100 Experiments program, which works with companies to build AI solutions, and the AI Makerspace program, which creates near-ready AI solutions called AI Breaks for quick adoption.

What are some examples of AI projects done by AI Singapore for companies?

Examples include predicting device return rates for IBM, building a fraud detection model for Sample Insurance, creating an end-to-end AI system for GIC to classify documents, and developing a speech-to-text transcription engine for multilingual Singaporean contexts.

Outlines

00:00

🤖 AI Applications and Object Detection in Singapore

This paragraph introduces Module 3 of 'AI for Everyone' and delves into the capabilities of AI, particularly highlighting AI projects by AI Singapore. It discusses the successful implementation of AI for object detection and recognition in various scenarios over the last decade. The script emphasizes the importance of correctly labeled datasets for training AI models, often involving low-paid workers in countries like India, Vietnam, and Africa. It also touches on the integration of heuristics in AI systems to enhance capabilities, such as post-estimation for understanding human poses and interactions, with specific use cases like social distancing during the COVID-19 pandemic and activity classification in gyms. The paragraph concludes by mentioning the role of AI in natural language processing, including spam detection and document understanding, showcasing AI's ability to augment human productivity without job displacement.

05:09

🌐 Multilingual AI and Speech Recognition in Singapore

The second paragraph focuses on the advancements in AI speech recognition, especially in multilingual societies like Singapore. It points out the challenges AI faces in understanding mixed-language speech and introduces the AI Speech Lab, a collaboration between AI Singapore and local universities that has developed an engine capable of deciphering English, Singlish, and other languages. The technology has been implemented in call centers for automatic transcription of conversations. The paragraph also outlines AI Singapore's core programs, the 100 Experiments and AI Makerspace, which assist companies in adopting AI solutions. Examples of these collaborations include predictive models for device return rates, fraud detection in insurance claims, and document classification systems. The paragraph also mentions other AI projects, such as predicting maintenance needs, analyzing social media for credit risk, and diagnosing chronic wounds using AI, emphasizing the wide range of AI applications in various industries.

10:12

🔍 Encouraging AI Adoption in Singapore

The final paragraph summarizes the impact of AI Singapore's initiatives on local companies' AI adoption. It invites viewers to explore the AI Singapore website and the AI community page for more information. The paragraph concludes Module 3 by setting the stage for the next module, which will discuss the responsible use of AI. It emphasizes the importance of understanding AI's role in society and preparing for ethical considerations in AI deployment.

Keywords

💡AI for Everyone

AI for Everyone is a module or course designed to educate individuals about the capabilities and applications of artificial intelligence. In the video, it serves as the overarching theme, exploring various AI projects and their impact on Singaporean companies, illustrating how AI is becoming accessible and beneficial to a wide range of people.

💡Object Detection

Object detection is a computer vision technique used to identify and locate objects within an image or video. The video script mentions that recognizing objects like cars, bicycles, and animals has become relatively easy with AI, highlighting its practical applications in everyday scenarios.

💡AI Singapore

AI Singapore is an organization that promotes the use of AI in the country. The script discusses various projects done by AI Singapore, demonstrating how it has been instrumental in advancing AI applications for companies in Singapore, such as in object detection and post-estimation.

💡Post Estimation

Post estimation is a more advanced computer vision task that goes beyond object detection to understand the poses and interactions of individuals within an environment. The video script provides an example of how post estimation was used for social distancing measures during the COVID-19 pandemic, showcasing its practical applications in safety and health.

💡Heuristics

Heuristics in AI refers to the use of practical methods or rules of thumb to solve complex problems. The script mentions that AI Singapore's computer vision team added heuristics to their AI systems to extend capabilities, allowing for more challenging tasks to be tackled, such as post estimation.

💡Natural Language Processing (NLP)

NLP is a branch of AI that focuses on the interaction between computers and human language. The video script discusses various NLP tasks, such as spam detection, document understanding, and automated code generation, emphasizing the wide range of applications of NLP in processing and understanding text data.

💡Spam Detection

Spam detection is an NLP task that involves identifying and filtering out unwanted emails. The script provides a personal anecdote about the evolution of spam detection engines, illustrating how AI has improved productivity by automating this specific work task.

💡Relation Extraction

Relation extraction is an NLP task that involves identifying relationships between entities in text. The script gives an example of how AI Singapore's SG NLP AI model was used to extract relations from the 'Mak Han Sutra' text, demonstrating the capability of AI to understand and process unstructured data.

💡Multilingual Speech Recognition

Multilingual speech recognition is the ability of AI systems to understand and process speech in multiple languages. The script discusses the challenges of understanding mixed-language speech, such as Singlish, and how AI Singapore partnered with researchers to develop an engine capable of deciphering such speech, highlighting advancements in speech recognition technology.

💡AI Breaks

AI Breaks, as mentioned in the script, refer to near-ready AI solutions developed by AI Singapore, based on AI research work. These solutions are designed to help companies, especially SMEs, quickly adopt AI with minimal research and development, showcasing the practical approach to integrating AI into various business operations.

💡100 Experiments Program

The 100 Experiments Program is an initiative by AI Singapore that collaborates with companies to build AI solutions where off-the-shelf solutions are not available. The script provides examples of projects from this program, such as predicting device return rates for IBM and building a fraud detection model for Sample Insurance, illustrating the program's role in fostering AI adoption and innovation.

💡AI Makerspace Program

The AI Makerspace Program is another initiative by AI Singapore that focuses on building AI solutions based on research. The script mentions how this program helps companies, particularly SMEs, to adopt AI solutions with minimal R&D, providing an example of how it was used to develop a delivery aggregation algorithm for a digital savvy SME.

Highlights

AI has been successfully used for object detection and recognition in various use cases over the last decade.

Correct datasets make it relatively easy to recognize people, cars, bicycles, animals, etc., with AI.

AI systems require a lot of carefully labeled images for training, often done by low-paid workers in countries like India, Vietnam, and Africa.

Heuristics added to AI systems extend their capabilities, allowing for more challenging computer vision tasks.

Post estimation in AI can understand poses and interactions of people with their environment.

AI was used for social distancing measures in HP's factory during the COVID-19 circuit breaker period.

AI can detect groups of people to support companies in ensuring COVID-19 safety protocols are followed.

Pose estimation can classify activities such as push-ups and sit-ups, aiding in exercise guidance for local gyms.

AI systems are used for sorting through documents, including spam detection in email systems.

Document understanding in NLP can extract relations from texts, as demonstrated with the 'Makhan Sutra' example.

AI Singapore's SG NLP AI model can decipher mixed-language sentences, including Singlish and other languages.

AI speech lab launched in 2019 to improve AI's understanding of multilingual speech.

AI Singapore's 100 Experiments program helps companies build AI solutions when off-the-shelf solutions are not available.

AI Makerspace program creates AI solutions based on research to help companies, especially SMEs, adopt AI quickly.

AI has been used for various projects, including predicting device return rates, fraud detection, and document classification.

AI Singapore has helped companies adopt AI for diverse applications, from speech-to-text transcription to suspicious vessel detection.

Popular AI use cases today include computer vision, natural language processing, and speech recognition.

AI Singapore's initiatives have significantly contributed to the adoption of AI by companies in Singapore.

Transcripts

welcome back to module 3 of ai for everyone  in this section we'll explore what ai can do  

and in particular we'll show you some  examples of ai projects done by ai singapore  

for companies here in singapore ai has been  successfully used for many object detection  

and recognition use cases in the last 10 years  recognizing people cars bicycles motorbikes  

cats and dogs is relatively easy now with  the correct data sets let's watch this video

as you can see this ai system is able to  pick up cars the bicycles the person and  

so on this video was taken outside my house what  is important to understand is that identifying  

people cars bicycles animals etc requires a lot  of carefully labeled images to train the ai model  

and someone has been paid 10 cents or  50 cents to tag or label these images  

the person will look at a picture and  then draw a box around a person or card  

and indicate that the box represents a  person or car this is a very laborious  

task and is often done by lowly paid workers in  countries such as india vietnam and africa today  

without these workers the ai system would  not have the labeled data sets to train on  

so intelligence still come from us human the ai  system is merely replicating what we tell them

besides object detection ai systems can do a lot  more today in ai singapore computer vision team  

we have added heuristics to the ai system  to extend the capabilities of the ai model  

this has allowed us to tackle more  challenging computer vision tasks  

in particular post estimation with post  estimation we can not only understand  

the poses the person is in but also his  or her interaction with the environment

let's look at some interesting use cases

one such use of post estimation is for social  distancing as part of kovit 19 safety measures hp  

approached ai singapore back in june  2021 during the circuit breaker period  

they were preparing to reopen their factory  and needed a way to ensure social distancing  

was maintained between  their workers in the factory  

we enhanced our post-estimation algorithm fine  posts for hp we implemented fine posts on an edge  

device mounted with a standard usb camera this ai  on the edge was deployed throughout hp factory in  

under two months you can see it in action in the  first video we then further extended fine posts  

to include the ability to detect groups of people  between two to five or eight this can be used to  

support companies and fmv outlets who need to  ensure kovit 19 safety pro costs are followed

as pose estimation can detect  various joints in images  

we classify activities such as  push-ups and sit-ups and count them  

also this has been deployed for local gym  here to help their clients exercise correctly

natural language processing on nlp is a popular ai  task organizations and individuals generates a lot  

of tax documents and ai systems are often  used to help sort through these documents  

one of the earliest and most useful ai systems  for tech processing sits inside your email  

system the spam detection engine i remember  when i started working in the early 1990s  

i had to spend 10 to 20 minutes every  day clearing my inbox of spam messages  

today i spent zero minutes as a modern spam  engine is really very good this is a good example  

of how ai has replaced a specific work task and  made you productive without taking away your job  

in fact you now have more time to focus on the  important emails document understanding is an  

important nlp task in this example i copied  the about text from the machan sutra site  

into ai singapore's sg nlp ai model for relation  extraction the ai model correctly found relations  

such as makhan sutra was founded by ksito  makan sutra inception was in 1997. other nlp  

tasks include text summaries sentiment analysis  automated news generation and the latest automated  

code generation ai system from microsoft and open  ai that has caused a stir in the programming world

singapore is a multilingual country  and it is common for people to speak  

more than one language in the same sentence  

they woman paktor always makhan at kopityam1  this single sentence consists of a mix of english  

tamil mandarin cantonese malay hokkien and  singlish it can be challenging for the ai programs  

from google or microsoft to understand sdr model  is often trained only on u.s english speakers

ai singapore partnered with  researchers in nus and ntu  

to launch the ai speech lab in 2019 and  they now have an engine that can decipher  

english singlish mixed with mandarin or  malay let's listen to this demo hello

very loud

this technology has been  deployed to several call centers  

not to replace the operators but to automatically  transcribe the conversation once the audio  

is converted into text all the powerful ai  guardians for text analysis can then be used

in ai singapore we have two core programs which  directly help companies in singapore to adopt  

ai faster in the 100 experiment program we  work with companies to build ai solutions  

when there is no off-the-shelf solutions and where  the company wants to hire the trained ai engineers  

after the project to build up their own  capabilities some hundred e examples are  

shown here for ibm we built an ai model to  predict which devices they have manufactured  

may have high return rates and we also help  automate the whole analysis end-to-end for sample  

insurance we build a fraud detection model into  their claims pipeline to detect suspicious claims  

these claims are not rejected but instead flagged  to a employee to review with gic we built not  

just an ai model but a complete end-to-end ai  system to ingest documents and classify them  

into different categories and rank them so that  the gic analysts can focus on the important ones  

in the ai makerspace program as nupor builds  near-ready ai solutions what we call ai breaks  

based mainly on ai research work done in ai  singapore and other parts of the world to help  

companies especially our smes to quickly adopt  them with very little r d we had earlier mentioned  

hp use of our fine posts ai break to deploy a  social distancing application in the factory  

at dapple road upasser is a digital savvy  sme who approached us for help to build a  

simple delivery aggregation algorithm that is now  embedded into their online platform it is used  

to cluster similar routes to reduce the number of  delivery vehicles required per area some of these  

companies have agreed to do a video interview  for us and you can watch them here in learn ai

other projects we have done for companies include  prediction of when a leave in hdb fed will fail  

for preventive maintenance classifying the risk  of someone borrowing money based on his or her  

social media activities analysis of the face  to recommend the right type of cosmetic to use  

chronic wound diagnosis using infrared images  this application in particular helped to  

cut down the nurse investigation  time from 30 minutes to 3 minutes  

we have spoken about our singlish speech-to-text  transcription engine we have also built models to  

detect suspicious vessels in our surrounding  waters and to predict inventory levels for  

factories here and one interesting project with  a large mnc was to predict which phrases and  

words to use to get better hits from a search  engine in this case japanese phrases and words  

let's summarize popular ai use cases with deep  learning algorithms today include computer vision  

natural language processing and speech recognition  there are so many ai use cases leveraging  

traditional statistical and machine learning  algorithms especially those with tabular data  

ai singapore's 100 experiments and ai  makerspace have helped companies here in  

singapore to adopt ai i invite you to spend  time exploring not just the learn ai site  

but also our main home page at aisingapo.org and  the ai community page at community.aisingaport.org  

this concludes module 3. in the next module  we'll discuss the responsible use of ai