How Pakistan is slowly progressing in Machine Learning
What do humans do to improve their accuracy and speed of work? They get the knowledge their brain uses to respond, calculate, and perform a task. What do the computers do to make themselves more efficient or more accurate? They use the method called Machine Learning (ML). As the world is progressing and companies are putting effort into increasing their production more efficiently, they are going towards using Artificial Intelligence (AI). Pakistan, which is also advancing in the field of IT, is also making efforts to stay with the world in the field of AI and ML with the help of brilliant minds. Still, most of them are working for foreign companies due to the limitation of resources in Pakistan.
AI is everywhere today, but there was a time when the whole field wasn't considered valuable. After initial advances and much hype in the mid-to-late 1950s and 1960s, breakthroughs stalled and did not meet the expectations. There just wasn't enough computing power to realize that potential. Also, operating such a system was prohibitively expensive. As a result, both interest and funds dried up. Then, increased research funding and an expanding algorithmic toolkit revived this pursuit in the 1980s. But it didn't take long for the decade-long AI winter to return.
Then came two significant changes that directly enabled AI as we know it today. AI efforts have moved from rule-based systems to ML techniques that can learn using data without external programming. At the same time, the World Wide Web has become ubiquitous in the hands of billions of people worldwide, leading to an explosion of data and data sharing on which ML relies.
AI (AI) is the ability of computers or computer-controlled robots to perform tasks that require human intelligence and judgment. Siri and Google Translate are examples of AI use cases, and they use an AI system based on the learning process of human neural networks.
Many say that AI will improve the quality of our daily lives by being more efficient than humans at basic and complex activities, making life easier, safer and more efficient. Others argue that AI threatens people's privacy, classifying people to exacerbate racism, displacing employees and increasing unemployment.
Machine learning
ML is a subcategory of AI that allows software applications to predict outcomes more accurately without explicitly being programmed. ML algorithms use historical data as input to predict new output values. We have made significant progress in the last ten years.
Although AI and ML are often used interchangeably, the two have essential differences. AI is a collective term for a set of technologies that enable computers to learn and act like humans. In short, AI makes the computer smart; however, ML is responsible for how computers become intelligent.
Unlike traditional programming, a handwritten program that takes input data, runs it on a computer, and produces output, in ML or augmented analytics, input data and output are passed to an algorithm to create a program. This leads to meaningful insights that can be used to predict future outcomes.
ML algorithms use statistics to find patterns in vast amounts of data, including images, numbers, and words. If the data can be stored in digital form, it can be input into ML algorithms to solve specific problems.
ML Engineer Abdul Raheem told The Express Tribune that the first method came out of his pure statistics in the 1950s. "They solved formal math problems by looking for patterns in numbers, evaluating the proximity of data points, and calculating vector directions. Today, half the internet is working on these algorithms. If you see a list of articles to read next or your bank blocks your card at a gas station middle of nowhere, that's probably one of those little guys' jobs." he said.
"Big tech companies are big fans of neural networks. Obviously, 2% accuracy is a $2 billion increase in sales for them, but it doesn't make sense when they're small. I've heard stories of a team that spent a year working on a new recommendation algorithm for an e-commerce site before discovering that 99% of traffic comes from search engines. Their algorithm was useless, and most users didn't even open the main page," said Raheem adding that the only goal of ML is to predict results based on incoming data. If a task is not represented this way, it wasn't an ML problem from the beginning.
Types of ML
There are different methods of training ML algorithms, each with its strengths and weaknesses. To understand the strengths and weaknesses of each type of ML, we first need to examine what kinds of data they ingest. ML has two types of data: labelled data and unlabeled data. Labelled data has input and output parameters in fully machine-readable patterns but requires a lot of human effort to label the data in the first place. Unlabeled data has only one or no machine-readable parameters. This eliminates human effort but requires a more complex solution. There are also several types of ML algorithms used for particular cases, but there are three main methods today; supervised learning, unsupervised learning, and reinforcement learning.
Supervised learning is like a student with a teacher. It's one of the most basic types of ML, where you label your data to tell machines the exact patterns you're looking for. Although the data must be labelled accurately, supervised learning is compelling and gives excellent results when used in the proper context.
"When we press play on a YouTube, we're informing the ML algorithm to find similar videos based on our preference. They are then shown as the next recommended video," explained Raheem, a software engineer turned ML engineer.
On the other hand, unsupervised learning is a student without a teacher with no data labels. The machine looks for patterns randomly; this type does not require human intervention to make the records machine-readable. This means that you can programmatically work with much larger data sets. Unlike supervised learning, unsupervised ML services are less popular due to less application in daily life.
"With unsupervised learning, the machine is left with a stack of photos and the task of identifying things in the photos. There is no teacher, and the machine tries to find patterns independently," said Raheem. "Of course, machines learn faster with the teacher, so they are used more often for real-world tasks. Such tasks include classification to predict categories of objects and regression to predict specific points on a numerical axis.
Reinforcement learning mainly describes a class of ML problems in which agents operate in environments without a fixed training data set. Agents should know how to handle feedback. This ML algorithm reinforces or promotes favourable outcomes and suppresses unfavourable results.
Use in daily life
ML has many applications, including external (customer-facing) applications such as product recommendations, customer service, and demand forecasting, as well as internal applications that help companies improve their products or speed up time-taking processes.
ML algorithms are typically used in areas where solutions need to be continuously improved after deployment. Customizable ML solutions are highly dynamic and used by companies across all industries.
If we see it around us, we will see the use of ML everywhere. ML is used in a variety of applications today. Perhaps one of the most famous examples of ML is the recommendation engine that powers social media news feeds.
Social media uses ML to personalize how each member's feed is served. If members read posts in a particular group less frequently, the recommendation engine will show that group's activity earlier in the feed.
Behind the scenes, the engine tries to reinforce known patterns of members' online behaviour. If members change their behaviour and stop reading posts from this group in the coming weeks, their news feed will adjust accordingly.
Similarly, it is used in the field of medicine. For example, Deep Patient is an AI-powered tool that helps doctors identify high-risk patients before they are diagnosed with the disease. According to inside BIGDATA, the technology uses a patient's medical history to predict nearly 80 illnesses up to a year before they will develop them.
In contrast, PathAI's ML algorithms help pathologists analyze tissue samples to diagnose accurately. Not only is diagnostic accuracy improved, but treatment is also improved. PathAI's algorithms can also find suitable participants in clinical trials.
Virtual assistants like Ideal use AI to support the hiring process. This is a new type of HR technology aimed at eliminating or reducing time-consuming tasks such as resume reviews, candidate identification and selection, and other repeating tasks. Data and predictive analytics are used in HR screening software. As a result, it helps optimize HR department time and break down prejudices through superior recruiting skills. AI supports decision-making by transforming data from various sources into better and sharper insights.
All of the AI software and platforms are powered by ML. Without this, they will not perform as efficiently as they do. The more data fed into the system, the better and more accurate results will be.
The most common areas where ML is used include; Identifying Spam, Making Product Recommendations, Customer Segmentation, Image & Video Recognition, Fraudulent Transactions, Demand Forecasting, Virtual Personal Assistant, Sentiment Analysis and Customer Service Automation.
Scope of ML in Pakistan
Pakistan's progress has been somewhat slower, with new achievements in specific areas such as AI and ML, but there are some notable achievements. Pakistan has developed AI (AI) technology to calculate COVID risk. The National University of Science and Technology (NUST) is now a top university in Pakistan for providing competitive and quality AI education.
In recent years, the Koshish Foundation Institute has proven to be a rising star in exploring and applying every aspect of AI research. One of his main goals is to use AI and ML to raise the bar and transform Pakistan's agricultural sector. Another initiative the company is working on is improving weather stations around the country, and its importance is evident given recent severe weather events and natural disasters in Pakistan.
A computer science graduate, Sabih Sheikh, now a data scientist, says that AI and ML are a big part of how the world works, continuing to evolve in various industries and improving the quality of life. "The scope of AI and ML is vast. Not only the local market but also international IT giants such as Google and Microsoft are hiring AI developers from Pakistan. Talented IA professionals are often welcomed by well-paid Pakistani software companies. In addition, professionals in this field can earn more money as freelancers. Many Pakistani companies have deployed AI and ML. Still, not much due to the limitation of data," he said, adding that projects like PIAIC will make the reach of AI in Pakistan greater than before.
The use of AI in medicine is increasing worldwide, but developing countries like Pakistan lag in adopting AI-based healthcare solutions. AI can help with diagnosis, monitoring, and resource allocation. Surveys show that most Pakistani doctors and medical students are unaware of AI and its uses but have positive opinions about it in healthcare and are open to its adoption.
Journalism is another field where AI and ML are used quite actively. ML can help news organizations improve their business models, for example, by fine-tuning flexible paywalls for subscribers. ML is already being used to enhance the skills of journalists throughout the journalism process.
With data stored in ever-greater amounts, investigative journalists can and have tried extraordinary analytical techniques to make sense of these massive datasets. It can also hold companies and governments accountable in the process. This is done using ML, which deepens data-driven reporting. In the age of big data, this technique is valuable and necessary.
The rule about when to use ML in reporting is pretty simple. When stakeholders can't reasonably analyze the data themselves (we're talking hundreds of thousands of rows in a spreadsheet), it's time to turn on the machine.
Institutes for ML in Pakistan
AI centres have been developed across the country to involve young people in the growth of the country's economy. Some of the top AI labs include; The national centre of AI at NUST Islamabad, the AI Research Lab (AIRL) at UET Lahore, Peshawar's UET holds Center of Intelligent Systems and Networks, and the AI Lab is present at IBA Karachi, who are offering bachelor's in AI (BSAI) degrees.
One game changer for the Pakistani market in AI and ML is the Presidential Initiative for AI & Computing (PIAIC). PIAIC was launched with a mission to reshape Pakistan by revolutionizing education, research, and business by adopting the latest, cutting-edge technologies. Experts are calling this the 4th industrial revolution. It aims to make Pakistan a global hub for AI, data science, cloud-native computing, edge computing, blockchain, augmented reality, and the internet of things.
The President launched the initiative in collaboration with the Sailani International Welfare Trust (SWIT) aims to empower youth through training in advanced online courses. Around 25,000 students from across Sindh participated in the Presidential Initiative for AI and Computing (PIAIC) 2022 admission test at the National Stadium in Karachi.
"What a pleasure it has been to motivate Pakistan's youth. 24,833 appeared in a test to start their mentored online course in Software, AI, Blockchain, IoT, Networking & Cloud computing. No other place could accommodate them but National Stadium. Karachiites will lead Pakistan," said President Arif Alvi after the first admission test.
This program is backed by the Digital Skill Program, an online platform for IT education. In contrast, Kamyab Jawan Program extends interest-free financial support of up to Rs. 1,000,000 to youth to launch their journey to become an entrepreneur.
Career opportunities in Pakistan
A vast number of Pakistani ML engineers and AI professionals are working in Pakistan, but significantly less for Pakistan-originated companies. Most professionals work for companies that provide services to foreign companies. "There are great minds in this field in Pakistan, but as Pakistan is lacking in the compiling a data, the professionals with the degree and experience are unable to utilize their skills for Pakistan. Except for some big companies, the reason for not adopting AI and ML is the lack of data available," said Sabih.
He added that for the AI and ML graduates/professionals, there are several career opportunities in robotics, computer vision, language processing, gaming, expert systems, speech recognition, and many more. "Anyone interested in this profession and willing to invest time and dedication in achieving a high level of education could easily be recruited into this fast-growing and challenging industry," he added.
This field is not only open for the one with a degree in the respective field, but the ones who know computers and probability can easily do wonders in this field.
Future of ML
ML is an incredible technology in the field of AI. Even with early applications, ML is already improving our daily lives and future.
As ML evolves, so makes the range of ML use cases and applications. To effectively address the business problems of this new decade, it is worth considering how ML applications can be used across all business areas to reduce costs, increase efficiency, and improve user experience.
If two technologies are combined, the potential to improve efficiency. For example, quantum algorithms could transform and revolutionize the field of ML. Quantum computing allows performing simultaneous multi-state operations, enabling faster data processing.
ML with Quantum can help you better analyze your data and gain deeper insights. Such improvements can help companies achieve better results than conventional ML methods.
So far, there are no commercially available quantum ML models. However, big tech companies have started to invest in this technology, and the rise of quantum ML systems is not far off.
Talking about the future in Pakistan, Raheem said that Pakistan has a long way to go to implement ML and improve productivity in the fields like healthcare and agriculture, which are in desperate need of technological breakthroughs. "The talent and skills are present, but they are yet to be used to make things better in Pakistan. Pakistan should have an AI policy to remain competitive in the global economy. The country is already lagging behind the rest of the world in using AI and is yet to benefit from the fourth industrial revolution. Missing the AI wave would be a disastrous setback for the nation's efforts," he concluded.