Previously, we discussed the underlying factors that are responsible for ushering what Bill Gates describes as the 'the age of AI". We also prophesied about what this looks like for Pakistan's middle-class workforce and, in particular, the knowledge workers.
OpenAI, the organisation behind the wildly popular ChatGPT platform has estimated that this new revolution of 'Generative AI" will impact 10% work processes of 80% of the people and about 50% work processes of 19% of the people.
We also hypothesised that, perhaps, this ratio may be reversed in Pakistan with 80% of the people seeing 50% of their jobs affected and the remaining 20% might see 10% of their jobs affected (or taken over) by AI. In Pakistan's case the 80% will comprise the vast majority of Pakistan's white collar knowledge economy workforce who might see their jobs being taken over by machines, or another individual aided by AI working twice/thrice as smart as them. The remaining 20% might be at the higher end or the lowest end of the knowledge economy workforce who might migrate to higher level or may be doing irreplaceable physical and mental labour, respectively, but they too shall see disruption in ways that are difficult to assess at this time.
In this follow-up article we dig deeper into this crisis of education and skills and look at how universities, which are supposed to be the “factories” for producing knowledge workers, might be affected or aided (positively) by the Generative AI revolution.
The short answer is that these are challenging times for universities, already battered by the growing shift towards self-paced remote learning in the aftermath of COVID19, with seriously adverse implications for the value proposition of universities. In addition to this, universities are suffering badly from the very questioning of what they teach and from the declining economic returns to education - particularly at the lower tiers of university system. Universities in Pakistan produce a vast number of poorly trained 'professionals' who start very low in the knowledge economy, must learn on-the-job, and barely survive in the jobs marketplace. There is a visible shift towards skills training that deliver near-instant returns and away from a broad-based university education that, if badly delivered, hardly makes a difference.
ChatGPT in the University:
Before ChatGPT came along, there was already a growing realisation that universities will either reform themselves or largely perish as a viable option for education - with the exception, perhaps, of the elite in the society who may still value the experience even if it doesn't teach them much.
But the Generative AI revolution and the growing popularity of ChatGPT and similar products changes everything.
It creates both a negative and a positive effect. On the negative side, it could further hasten the decline of the universities that do not quickly get their act together as AI promises to automate most of the tasks for which these universities currently train their students. On the positive side, however, it promises to create the means and tools necessary to fix or bypass some of the most significant challenges faced by our university education.
We will spend the rest of this article looking at the positives and analysing how Generative AI and ChatGPT could be used to support our universities leapfrog many stages in their evolution from largely dysfunctional entities creating poor quality output to ones producing smart and market-ready graduates.
Academia in Pakistan suffers from a number of different self-imposed handicaps that, by and large, can be summarised as taking in poor quality inputs, running them through poor quality processes, and thus producing poor quality outputs. Structurally, these handicaps (as well as their solutions) can be summed into 4 broad categories: reading, writing, judging, and learning. Let us explain each in some detail.
Writing:
Clear and concise writing, particularly in the English language, is a major challenge for most students entering Pakistani universities—not to mention the faculty, staff, and leaders of these universities. A vast majority of high school and college graduates do not study most subjects in English and thus are barely able to develop enough facility with the language to get by. This is not enough for a university level education where, unfortunately, most of the good quality textual material is in English and where the lack of facility with language seriously hampers their learning experience. The result is an overall diminution of quality in academia as well as other domains of social and economic life.
From the perspective of university leadership, faculty, researchers and staff as well, writing plays a critical part in defining higher quality outcomes. And here, too, an inability to comprehend and express oneself in English leads to a frustrating experience.
The astonishing ability of Generative AI/ChatGPT to produce reasonably sophisticated text in response to simple queries and without any grammatical errors holds a lot of promise to assist Pakistani academics in overcoming the handicaps imposed by their weak communication skills. Deployed properly, this assistance can be transformative. On the one hand, it can carry Pakistani academics and students beyond their currently handicapped communication skills and allow them to focus on higher order learning. On the other hand, it can create a continuous learning experience for assessing and improving the quality of one’s own work. Similar benefits can accrue to university administrators and their staff, who could now be freed from the toil of (poorly written) drafting letters, memos, meeting minutes and other forms of office correspondence, and focus on higher order academic and administrative responsibilities. Faculty members could get support and help from Generative AI to help write first drafts of syllabuses, policies, announcements, quizzes, exam questions, or presentations.
Judging:
Like writing, judgment and assessment lie at the heart of the academic enterprise. Being an academic means spending one’s entire life separating wheat from chaff. University leadership and faculty have to evaluate their students, their peers, admissions applicants, classroom performance, assignments, tests, research papers, grant proposals, CVs, the list can go on. This can be an extremely time-consuming exercise, especially for those handicapped by poor writing and reading skills.
A weak academic system, such as Pakistan’s, tends to be lacking in in the area of judgment and the evaluation and assessment of quality. Degrees are awarded to undeserving persons through incompetent and inadequate exams. Grant proposals are assessed by stacked panels of incapable reviewers. Recruitment systems are based on archaic and indefensible assessment systems and procedures. Promotions and tenure decisions are made on the basis of laughable quantitative criteria.
In strong academic systems, university leaders and faculty develop their evaluative capabilities over years and years of experience. The experience enables them to develop an eye for detail, a mental model of what constitutes quality work, and a well-honed sense of what a good paper or proposal looks like. In the absence of such experience, academic systems tend to produce the kind of perverse outcomes identified above.
Restoring the integrity of the academic system is a long-term task, but technology can certainly help.
Generative AI and ChatGPT already has the capacity to summarise documents and query them in certain ways to assess their veracity and rigor. A system that can quickly learn the sum total of the years of learning and experience of the best of our reviewers and faculty and reliably replicate the results will significantly reduce the burden of evaluation on academics. This might require tools and plug-ins to be built on top of the ChatGPT's fundamental model and carry out supervised learning to ensure reliability and accuracy of output received. Once properly trained, the system can automatically perform the more mundane evaluation tasks, e.g., initial screening of papers, proposals, CVs, essays or exams, provision of draft comments, or preparation of summaries of submitted documents. This can free up the time of university leaders and faculty for final assessment or higher order evaluations. Once again, the continuous interaction with a professional evaluation system will also provide a learning environment for users.
Reading:
Given that the bulk of high-quality academic materials are in English, many Pakistani students and faculty face a severe handicap in accessing this material. Access to knowledge in the language of your choice is a major barrier to the academic process. Advanced academic texts— whether they are on philosophy, science, politics, history, medicine, or engineering—are predominantly in English. The quality of academic texts (other than poetry or fiction) in Urdu are far, far, inferior to those in English.
Although Pakistan is a multicultural society, with many languages, ethnicities, creeds, cuisines, customs, and cultural practices, the country has failed to create a massive translation movement that could provide its people with knowledge from around the world in their native language in a timely fashion. If one has ever ventured to a developed country or that is proud of its intellectual achievements and one visits a bookstore, one would find it almost next to impossible to find many many English titles. On the contrary, one would be pleasantly surprised to find every single book in the world translated in the local language of that country. This is what living nations do. Think of Turkey, or Finland, or China. They don't starve their people of access to global knowledge, they proudly translate all that is available into their local language.
Certainly not so in Pakistan where titles in Urdu and regional languages are hard to come by! And this where the knowledge divide starts for most at a very early age and remains throughout life. Similar disparities exist with respect to age as well as access to technology infrastructure and internet. The poor are systematically discriminated against because of their weak language skills, poor technological savvy, or lack of access to infrastructure and equipment.
Technology could be an important enabler but these avenues are distributed unequally. The simplicity and user-friendliness of ChatGPT and other technologies offer avenues of social advancement and the hope of producing solutions that can help bridge these divides at least partially. ChatGPT already has reasonable language translation ability and the ability to produce text in Urdu and other languages. Perhaps some work on refining the Urdu language capabilities of ChatGPT based model and we should have the means to providing every Pakistan with relevant content of their interest in the language of their forefathers in an automated, timely, and rather costless fashion.
No Pakistani should be at the receiving end of knowledge poverty just because they want to hang onto the language of their forefathers.
Learning:
Embracing this new technology will also require a significant learning effort. Universities are not known to change easily and quickly; and a vast majority of university leadership and staff is usually too old to fully embrace rapid technological change.
We should expect some resistance but be ready to demonstrate value and provide support and training for those who see the writing on the wall and who want to change. This could take the form of training and courses in understanding how ChatGPT works, the newly popular skill of "prompt engineering" or "AI whispering" as it is being called that will allow people to use AI's assistance properly, designing evaluation frameworks to work using ChatGPT, etc.
ChatGPT (and generative AI more generally) offers a number of solutions that could help Pakistani academia pull itself out of its rut and initiate progress towards the enhancement of quality and efficiency. Some of these solutions appear to be easy enough for the early adapters to begin using them without any further intermediation. Others will be more costly or difficult and may require additional work to build on existing platforms. Yet others might be in the relatively distant future and may have to wait for the construction of additional platforms and the investment of significant additional resources.
Sequencing the response:
We should classify use-cases to be developed for academia in 4 broad categories and get to work on these immediately in the following order:
Low-Hanging Fruits are some activities, such as drafting, that are already happening among the early adapters. However, some actions will need championing by a legitimate group of early adaptors and champions. These include hortatory or facilitating actions to broaden the group of adapters, building consensus around a framework for channeling AI actions towards socially beneficial areas.
Within Reach are other actions, such as some evaluative work has already in a mature form in advanced countries. ChatGPT and other solutions have brought it closer to the reach of potential users in lower income and less advanced countries. However, these will require some dedicated investment of resources to place them at the service of ultimate users—the so-called “last mile” challenge.
Pooled Actions: are some actions that would need further development such as developing a corpus of Urdu language to generally improve the translation skills of ChatGPT (which currently mixes a lot Hindi for Urdu) and this will require some funded effort on the part of a dedicated group of people and a technical team. This might also require collecting and training the models on Pakistani data.
Longer-Term: these are actions that are desirable but not within the current scope of ChatGPT or other Generative AI Tools. They might require a larger volume of investment or further development and maturity of AI platforms so will have to wait a bit before these can be put to use.
Enter PakGPT
The authors have proposed the formation of “PakGPT” - an informal association of scholars, academic leaders, AI researchers, and entrepreneurs – who have come together to provide thought leadership, guidance, and technological support (including, but not limited to, building simple software tools) to allow key stakeholders in Pakistan’s academia – university leaders, faculty, researchers, students, and parents – to navigate the challenges and opportunities inherent in the ‘Age of AI’ through the use of tools such as Large Language Models (such as ChatGPT and its clones) as well as other types of Generative AI Technologies.
In an earlier article, we raised the issue of the potential dangers of Generative AI and other futuristic AI technologies. Those debates are important and need to be addressed. The founding of the PakGPT group is premised on the idea that participation in the lingering debate over the dangers of technology cannot be done from the outside. The technology is already here and in the hands of a vast number of people, including those associated with academia and education in Pakistan. The faculty and leaders of our universities must take the initiative to learn, adapt, and use these technologies in order to be able to guide their proper use for the welfare of our citizens.
The Age of AI is upon us; it will either destroy us or make us stronger. Whatever qualms and misgiving some persons might have about this, the question is not whether the ‘genie’ can be put back into the bottle but whether it can be deployed, intelligently, for the service of all stakeholders in the Pakistani education system.
History will not forgive us if we remain laggards in this bold new race.
(Dr. Tariq Banuri is Professor of Economics at the University of Utah, and a former Chairman of the Higher Education Commission. Email: Tariq.Banuri@Gmail.com)
OpenAI, the organisation behind the wildly popular ChatGPT platform has estimated that this new revolution of 'Generative AI" will impact 10% work processes of 80% of the people and about 50% work processes of 19% of the people.
We also hypothesised that, perhaps, this ratio may be reversed in Pakistan with 80% of the people seeing 50% of their jobs affected and the remaining 20% might see 10% of their jobs affected (or taken over) by AI. In Pakistan's case the 80% will comprise the vast majority of Pakistan's white collar knowledge economy workforce who might see their jobs being taken over by machines, or another individual aided by AI working twice/thrice as smart as them. The remaining 20% might be at the higher end or the lowest end of the knowledge economy workforce who might migrate to higher level or may be doing irreplaceable physical and mental labour, respectively, but they too shall see disruption in ways that are difficult to assess at this time.
In this follow-up article we dig deeper into this crisis of education and skills and look at how universities, which are supposed to be the “factories” for producing knowledge workers, might be affected or aided (positively) by the Generative AI revolution.
The short answer is that these are challenging times for universities, already battered by the growing shift towards self-paced remote learning in the aftermath of COVID19, with seriously adverse implications for the value proposition of universities. In addition to this, universities are suffering badly from the very questioning of what they teach and from the declining economic returns to education - particularly at the lower tiers of university system. Universities in Pakistan produce a vast number of poorly trained 'professionals' who start very low in the knowledge economy, must learn on-the-job, and barely survive in the jobs marketplace. There is a visible shift towards skills training that deliver near-instant returns and away from a broad-based university education that, if badly delivered, hardly makes a difference.
ChatGPT in the University:
Before ChatGPT came along, there was already a growing realisation that universities will either reform themselves or largely perish as a viable option for education - with the exception, perhaps, of the elite in the society who may still value the experience even if it doesn't teach them much.
But the Generative AI revolution and the growing popularity of ChatGPT and similar products changes everything.
It creates both a negative and a positive effect. On the negative side, it could further hasten the decline of the universities that do not quickly get their act together as AI promises to automate most of the tasks for which these universities currently train their students. On the positive side, however, it promises to create the means and tools necessary to fix or bypass some of the most significant challenges faced by our university education.
We will spend the rest of this article looking at the positives and analysing how Generative AI and ChatGPT could be used to support our universities leapfrog many stages in their evolution from largely dysfunctional entities creating poor quality output to ones producing smart and market-ready graduates.
Academia in Pakistan suffers from a number of different self-imposed handicaps that, by and large, can be summarised as taking in poor quality inputs, running them through poor quality processes, and thus producing poor quality outputs. Structurally, these handicaps (as well as their solutions) can be summed into 4 broad categories: reading, writing, judging, and learning. Let us explain each in some detail.
Writing:
Clear and concise writing, particularly in the English language, is a major challenge for most students entering Pakistani universities—not to mention the faculty, staff, and leaders of these universities. A vast majority of high school and college graduates do not study most subjects in English and thus are barely able to develop enough facility with the language to get by. This is not enough for a university level education where, unfortunately, most of the good quality textual material is in English and where the lack of facility with language seriously hampers their learning experience. The result is an overall diminution of quality in academia as well as other domains of social and economic life.
From the perspective of university leadership, faculty, researchers and staff as well, writing plays a critical part in defining higher quality outcomes. And here, too, an inability to comprehend and express oneself in English leads to a frustrating experience.
The astonishing ability of Generative AI/ChatGPT to produce reasonably sophisticated text in response to simple queries and without any grammatical errors holds a lot of promise to assist Pakistani academics in overcoming the handicaps imposed by their weak communication skills. Deployed properly, this assistance can be transformative. On the one hand, it can carry Pakistani academics and students beyond their currently handicapped communication skills and allow them to focus on higher order learning. On the other hand, it can create a continuous learning experience for assessing and improving the quality of one’s own work. Similar benefits can accrue to university administrators and their staff, who could now be freed from the toil of (poorly written) drafting letters, memos, meeting minutes and other forms of office correspondence, and focus on higher order academic and administrative responsibilities. Faculty members could get support and help from Generative AI to help write first drafts of syllabuses, policies, announcements, quizzes, exam questions, or presentations.
Judging:
Like writing, judgment and assessment lie at the heart of the academic enterprise. Being an academic means spending one’s entire life separating wheat from chaff. University leadership and faculty have to evaluate their students, their peers, admissions applicants, classroom performance, assignments, tests, research papers, grant proposals, CVs, the list can go on. This can be an extremely time-consuming exercise, especially for those handicapped by poor writing and reading skills.
A weak academic system, such as Pakistan’s, tends to be lacking in in the area of judgment and the evaluation and assessment of quality. Degrees are awarded to undeserving persons through incompetent and inadequate exams. Grant proposals are assessed by stacked panels of incapable reviewers. Recruitment systems are based on archaic and indefensible assessment systems and procedures. Promotions and tenure decisions are made on the basis of laughable quantitative criteria.
In strong academic systems, university leaders and faculty develop their evaluative capabilities over years and years of experience. The experience enables them to develop an eye for detail, a mental model of what constitutes quality work, and a well-honed sense of what a good paper or proposal looks like. In the absence of such experience, academic systems tend to produce the kind of perverse outcomes identified above.
Restoring the integrity of the academic system is a long-term task, but technology can certainly help.
Generative AI and ChatGPT already has the capacity to summarise documents and query them in certain ways to assess their veracity and rigor. A system that can quickly learn the sum total of the years of learning and experience of the best of our reviewers and faculty and reliably replicate the results will significantly reduce the burden of evaluation on academics. This might require tools and plug-ins to be built on top of the ChatGPT's fundamental model and carry out supervised learning to ensure reliability and accuracy of output received. Once properly trained, the system can automatically perform the more mundane evaluation tasks, e.g., initial screening of papers, proposals, CVs, essays or exams, provision of draft comments, or preparation of summaries of submitted documents. This can free up the time of university leaders and faculty for final assessment or higher order evaluations. Once again, the continuous interaction with a professional evaluation system will also provide a learning environment for users.
Reading:
Given that the bulk of high-quality academic materials are in English, many Pakistani students and faculty face a severe handicap in accessing this material. Access to knowledge in the language of your choice is a major barrier to the academic process. Advanced academic texts— whether they are on philosophy, science, politics, history, medicine, or engineering—are predominantly in English. The quality of academic texts (other than poetry or fiction) in Urdu are far, far, inferior to those in English.
Although Pakistan is a multicultural society, with many languages, ethnicities, creeds, cuisines, customs, and cultural practices, the country has failed to create a massive translation movement that could provide its people with knowledge from around the world in their native language in a timely fashion. If one has ever ventured to a developed country or that is proud of its intellectual achievements and one visits a bookstore, one would find it almost next to impossible to find many many English titles. On the contrary, one would be pleasantly surprised to find every single book in the world translated in the local language of that country. This is what living nations do. Think of Turkey, or Finland, or China. They don't starve their people of access to global knowledge, they proudly translate all that is available into their local language.
Certainly not so in Pakistan where titles in Urdu and regional languages are hard to come by! And this where the knowledge divide starts for most at a very early age and remains throughout life. Similar disparities exist with respect to age as well as access to technology infrastructure and internet. The poor are systematically discriminated against because of their weak language skills, poor technological savvy, or lack of access to infrastructure and equipment.
Technology could be an important enabler but these avenues are distributed unequally. The simplicity and user-friendliness of ChatGPT and other technologies offer avenues of social advancement and the hope of producing solutions that can help bridge these divides at least partially. ChatGPT already has reasonable language translation ability and the ability to produce text in Urdu and other languages. Perhaps some work on refining the Urdu language capabilities of ChatGPT based model and we should have the means to providing every Pakistan with relevant content of their interest in the language of their forefathers in an automated, timely, and rather costless fashion.
No Pakistani should be at the receiving end of knowledge poverty just because they want to hang onto the language of their forefathers.
Learning:
Embracing this new technology will also require a significant learning effort. Universities are not known to change easily and quickly; and a vast majority of university leadership and staff is usually too old to fully embrace rapid technological change.
We should expect some resistance but be ready to demonstrate value and provide support and training for those who see the writing on the wall and who want to change. This could take the form of training and courses in understanding how ChatGPT works, the newly popular skill of "prompt engineering" or "AI whispering" as it is being called that will allow people to use AI's assistance properly, designing evaluation frameworks to work using ChatGPT, etc.
ChatGPT (and generative AI more generally) offers a number of solutions that could help Pakistani academia pull itself out of its rut and initiate progress towards the enhancement of quality and efficiency. Some of these solutions appear to be easy enough for the early adapters to begin using them without any further intermediation. Others will be more costly or difficult and may require additional work to build on existing platforms. Yet others might be in the relatively distant future and may have to wait for the construction of additional platforms and the investment of significant additional resources.
Sequencing the response:
We should classify use-cases to be developed for academia in 4 broad categories and get to work on these immediately in the following order:
Low-Hanging Fruits are some activities, such as drafting, that are already happening among the early adapters. However, some actions will need championing by a legitimate group of early adaptors and champions. These include hortatory or facilitating actions to broaden the group of adapters, building consensus around a framework for channeling AI actions towards socially beneficial areas.
Within Reach are other actions, such as some evaluative work has already in a mature form in advanced countries. ChatGPT and other solutions have brought it closer to the reach of potential users in lower income and less advanced countries. However, these will require some dedicated investment of resources to place them at the service of ultimate users—the so-called “last mile” challenge.
Pooled Actions: are some actions that would need further development such as developing a corpus of Urdu language to generally improve the translation skills of ChatGPT (which currently mixes a lot Hindi for Urdu) and this will require some funded effort on the part of a dedicated group of people and a technical team. This might also require collecting and training the models on Pakistani data.
Longer-Term: these are actions that are desirable but not within the current scope of ChatGPT or other Generative AI Tools. They might require a larger volume of investment or further development and maturity of AI platforms so will have to wait a bit before these can be put to use.
Enter PakGPT
The authors have proposed the formation of “PakGPT” - an informal association of scholars, academic leaders, AI researchers, and entrepreneurs – who have come together to provide thought leadership, guidance, and technological support (including, but not limited to, building simple software tools) to allow key stakeholders in Pakistan’s academia – university leaders, faculty, researchers, students, and parents – to navigate the challenges and opportunities inherent in the ‘Age of AI’ through the use of tools such as Large Language Models (such as ChatGPT and its clones) as well as other types of Generative AI Technologies.
In an earlier article, we raised the issue of the potential dangers of Generative AI and other futuristic AI technologies. Those debates are important and need to be addressed. The founding of the PakGPT group is premised on the idea that participation in the lingering debate over the dangers of technology cannot be done from the outside. The technology is already here and in the hands of a vast number of people, including those associated with academia and education in Pakistan. The faculty and leaders of our universities must take the initiative to learn, adapt, and use these technologies in order to be able to guide their proper use for the welfare of our citizens.
The Age of AI is upon us; it will either destroy us or make us stronger. Whatever qualms and misgiving some persons might have about this, the question is not whether the ‘genie’ can be put back into the bottle but whether it can be deployed, intelligently, for the service of all stakeholders in the Pakistani education system.
History will not forgive us if we remain laggards in this bold new race.
(Dr. Tariq Banuri is Professor of Economics at the University of Utah, and a former Chairman of the Higher Education Commission. Email: Tariq.Banuri@Gmail.com)