Published at http://library1.njit.edu/staff-folders/sweeney/Articles/Article%20July%2018,%202002.htm
The Higher Education Competitive Advantage: Accelerated Learning & Artificial Intelligence
July 18, 2002
Just as there is no area of human endeavor that is unaffected by new knowledge, there is no area that is unaffected by learning. If learning is the principal skill of this knowledge age, then accelerated learning is the competitive advantage. Accelerated learning is the continuously increasing acquisition and use of new knowledge with little or no loss of understanding. The significant competitive advantage goes to those individuals and institutions that have succeeded in accelerating learning. All institutions, especially those in business, are affected by the speed of learning. Businesses that are quick to learn and use new knowledge have definite competitive advantages over those that are slower. In this knowledge age, both wealth and power come from accelerated learning.
The real question is whether it is possible to accelerate human learning. Can we learn the same skills with the same or better understanding in less time using new technologies and processes? The answer, as we shall discuss, is definitely yes.
Accelerated learning will inevitably change higher educational institutions as they prepare students to learn “how to learn”. Certainly the processes and technologies that enable accelerated learning will transform all institutions and individuals. If accelerated learning is the primary competency of this new century, then educational institutions must adopt such techniques and processes, just to survive.
It is not at all clear that colleges and universities have accelerated the learning of their students. It is also not clear that colleges and universities have taught and motivated students effectively to accelerate their own learning. Assisting students in the process of accelerating their personal learning has not been one of the historical roles of higher education and will require radically new academic strategies, processes and technologies.
The volume of new knowledge is constantly accelerating. There are a number of indicators showing the growth in published knowledge. The increased number of published journal articles, for example, demonstrates this. There has been a rapid increase in patents and copyrights awarded. There were more scientific papers published in the 1990s, for example, than in the 1980s, which was greater than the 1970s, and so on. The growth in PhD researchers, creating new knowledge, has accelerated. There has also been a growth in companies and institutions (e.g. higher education) that produce new knowledge.
Knowledge, itself, has caused the rapid spread of more knowledge. New knowledge makes technology and processes possible that, in turn, create further new knowledge. For example, new knowledge was used to create advanced computer software and electron telescopes that have, in turn, made it possible to learn about parts of the universe that were never understood before.
Knowledge of computers, software, telecommunications and the internet have clearly caused faster availability of new knowledge. The faster availability of new knowledge makes it possible to use that knowledge more quickly to further create additional new knowledge. Discoveries happen more rapidly.
The most successful learners spend less time accessing knowledge from libraries, for example, and more time learning. This creates even greater amounts of new knowledge that we then need to learn. This is the cycle of accelerated learning. Knowledge begets new learning and new knowledge in a continuously faster cycle.
The advantage has always gone to the fastest learners. In the early days of our species the slowest learners died and the faster learners propagated. Evolution has always provided humans with the imperative to accelerate learning. Biological, genetic evolution occurred slower than the current learning acceleration today. Since information technology increases new knowledge, it also creates the imperative to accelerate learning.
Perhaps it has only been in the last century that the rate of learning acceleration has begun to reach our human limits. Accelerated learning has caused us to run into the limits of time, attention, energy, resources, motivation, and memory. There is some evidence that accelerated learning has also increased our stress. We don’t have enough time to learn whatever new knowledge we need to know. Our time is fixed. We are genuinely fatigued and stressed by the effort of constantly learning new skills and knowledge. Our minds sometimes don’t seem to want to pay attention, drifting to some other task. In short, we need a new innovation that will greatly increase our learning speed as well as our motivation, working memory, energy and so on.
What is learning? Although there are many different definitions, the concept of learning can be found by examining experts in the area. We prefer a definition that includes a change in a learner’s skills, knowledge, understanding and behavior, obtaining meaning from experience and prior knowledge, constructing meaning pro-actively, adapting to the environment, and setting personal goals. Certainly there is little doubt that learning is also evolutionary. We use the definition that learning is the act of gaining knowledge and understanding from study, instruction and experience.
Learning has typically been associated until recently with organisms, especially our species. Yet is it possible and practical that machines can learn? While it is not yet clear if every aspect of learning can be duplicated by machine, there is already evidence that machines can learn and improve a new skill, can adapt to a new environment, and actually respond in a manner that was neither anticipated nor designed by their programmers and creators. It is not yet clear if they can change their own “goal”.
Cognitive science has already suggested ways that learning can be made more efficient. There is no evidence that college and university administrators have learned from the cognitive scientists. Cognitive scientists have told us that experts are those individuals who have the ability to manage cognitive resources more efficiently when solving problems in a knowledge domain. They can reach their conclusions more quickly and with a greater accuracy. There are known bottlenecks that students have to negotiate to progress from basic to intermediate to expert proficiency. Learning efficiency is the speed with which we learn and are able to use some new knowledge within a subject domain. Learning efficiency is the critical factor in accelerating learning. In other words, accelerated learning depends upon speeding novice learners into experts.
It has long been known from scholars that working memory capacity is a limiting factor in our ability to process information. Working memory includes the resources needed to remember the results of mental operations for short periods of time after processing that information. Sometimes this is called short-term memory. It is a bottleneck in our cognitive ability. The choke point in human learning is working memory, which is quite small. Therefore we have begun to depend to a greater extent on our information systems to remember and expand our working memory. Perhaps learning technology can be used to bypass or enhance working memory. For example, we already use the memory feature on our phones to remember the telephone numbers of our friends and family. This may be to “buffer” short term knowledge that can be called at the time it is needed. Improving working memory will certainly accelerate human learning if everything else remains even.
Colleges and universities have and will continue to employ more efficient and effective, machine assisted learning technologies and processes. Such conventional, state-of-the-art systems and algorithms do not learn themselves, but assist teachers or learners. They may include, for example, multimedia courseware, e-learning, distance learning, simulations, virtual reality, automated assessment, etc. Learning platforms today increase both the variety and combination of learning formats (multimedia; audio, etc.) and provide the learner timely and useful feedback and interactivity. Today such systems provide remote learning opportunities when and where a student is ready to learn and by providing faster access (e.g. a video of a lecture; a book from the library, etc.). Higher education has already moved into machine-assisted-learning through web based enterprise-wide learning platforms such as BlackBoard.com, WebCT.com, WebBoard.com, and Prometheus.com, (recently acquired by BlackBoard from George Washington Universtiy). Current learning technologies are primarily teacher driven.
Current learning technologies are not capable of making any useful inferences to help the learner, only the inferences programmed by the teacher/designer. They cannot create new rules for solving problems. The human teacher or tutor makes all of the useful inferences and adds all of the rules that have been coded into the algorithms. The human, either learner or tutor, picks when, where, and how to learn.
There are skills and processes that can increase the brain’s ability to learn faster such as memorization techniques, action learning (by doing), and team learning. The manner in which learning rules in a specific discipline are applied is the difference between an expert and a novice. For example, the expert in chess is able to see the pattern of the board while the novice typically only sees individual pieces. The expert knows which moves are the most important while all of the moves to a novice have approximately the same importance.
Today a person can play a game such as chess, poker, hearts, pool, and so on, against some of the best human players in the world anytime using current technology. (e.g. Yahoo games). A person can constantly play against very different players, some of them much better. The players are diverse, varying widely in age, sex, ethnic background, etc. The real advantage of such technology is that an individual can learn by doing (or in this instance, more precisely playing). Computer networks make such games possible by pitting remote players against each other. Furthermore, the network keeps track of performance and permits an individual to learn from other more advanced players. Games are much more than entertainment. They can be serious case studies to explore intelligence, learning, cognitive psychology, and human interaction.
The merger of simulation and virtual reality software with e-learning greatly increases the learning speed for thousands or even millions of people. Virtual reality environments are already being used in higher education. The best simulators are designed to teach individuals basic skills (e.g. virtual laboratories) in a highly interactive, entertaining and graphic rich, 3D multimedia environment. They are much more effective tools for learning complex skills, involving less cost, risk, and time. As advanced as such systems are however, the systems do not learn nor do they alter their own algorithms. Current learning technologies, no matter how advanced, lack the ability to progress without human intervention…at least for now.
The future is the marriage of current learning technologies and artificial intelligence to accelerate learning. Artificial intelligence (AI) is the science of making machines do things that would require intelligence if they were done by humans. AI is composed of a collection of different software and sometimes hardware. AI software includes expert systems, artificial neural networks, genetic algorithms, sensory capabilities, natural language processing, data mining, machine learning, and others. One type of AI software is often used with another. For example, natural language processing is often used with artificial neural networks.
There are three questions for this paper. Can AI be used to increase learning effectiveness, i.e. improve the level of understanding for better use of that knowledge? Can AI be used to make sure that a concept is fully understood in all of its particulars so that it can be used more quickly? Can AI be used to accelerate human learning and use? The answers to these questions are critical to the future of higher education. The authors are suggesting that it is probable that the answer to all three questions is yes.
This paper is not intended to discuss every type of AI algorithm nor the vast array of potential applications. The future of AI will certainly take care of itself, so the big issues of whether conscious machines can be built (Alekzander says yes) and whether machines will be more intelligent than humans in this century (Ray Kurzweil says yes) are interesting and legitimate questions for the philosophers but not for this paper.
The important point about AI is that companies are now using it in real applications today and higher education can also. Electric utilities are using AI systems to diagnose power quality (PQ) problems to supplement human expertise that is in short supply. Several Internet retailers are using AI to monitor shoppers and make recommendations for future purchases. AI, in the form of an artificial neural network, is being used to find the genetic mutations that create drug resistance in HIV, the AIDs virus. One AI process examines the genes of HIV infected patients and determines the best mix of drugs for treatment. This application would be too lengthy and too expensive to be performed by human beings but is only possible because of the pattern recognition ability of AI, specifically artificial neural networks (ANN). AI can perform pattern recognition, e.g. recognizing a face or a voice or patterns in data. There are many, many examples.
AI can also generate different “genetic algorithm mutations”. A checkers program, for example, can try different moves and strategies and play these against one another, learning better and better strategies. The program keeps the most successful strategies and kills off the less successful ones. AI can quickly perform thousands or even millions of iterations to find the best result.
AI can amplify existing current learning technologies and therefore accelerate student learning. AI can help students learn faster and with greater understanding. Students will be able to acquire needed skills, including critical thinking skills, much faster. AI will increase student motivation to learn. The amount of learning proportionate to the amount of effort will increase.
The AI advantage to faculty is that they will have more time for higher level research and mentoring. AI might be used to construct 90% of the examinations and the grading, for example. AI might be used to determine which students are falling behind and suggest why. AI is already being used to help grade essay examinations at the high school level.
AI can be used to detect patterns and make inferences. AI learning technology could adapt the learning environment to personalize the style to the individual learner and adapt simulations based upon the conditions and the learner’s style and behavior. For example, AI could be used to evaluate a student in a human anatomy simulator. It might analyze whether the student can identify each organ and maladies. It may also be used to “read” and evaluate essays written by the student and identify creative expressions. AI has been used to create poetry, for example, that could not be differentiated from poetry created by a typical student. AI will not only improve science and technology, it will improve humanities and the arts as well.
deduce "causes" of knowledge “that should exist”. For example,
when you observe a cup, you can deduce that it was made to hold some kind of
liquid or powder. If the causes are unknown then it can deduce that such
knowledge should be discovered. For
example, you observe an artificial object that you have never seen
before. You may deduce that it had a purpose, e.g. art, function, etc.
and then seek to learn what that purpose may be.
AI can also consider potential or future "consequences" of some given knowledge. We observe leaves falling from the trees, so we deduce that winter will shortly be coming. We can further deduce that we should be prepared for the cold. AI can infer such consequences based upon knowledge already possessed.
AI can be used to interactively and continuously assess the learner’s progress rather than waiting for a periodic exam or for the learner to ask for assistance. AI can focus on a particular learner in a personalized manner gathering much more feedback than is possible with traditional learning processes. AI can restate the question to optimize learner feedback. Does the learner understand the question? Does the learner understand the concept? Does the learner follow all the steps in the right sequence? Does the learner make the correct inferences? For example, AI might assess a student more frequently when it is obvious that a student’s learning performance is decreasing.
AI could learn about the student and their learning goals. AI could keep track of the knowledge learned through asking questions as well as unobtrusively monitoring activity. Imagine a system that knows exactly what the student learned and when and how well it was learned. Just as the machine can deduce what knowledge it needs to know, it can also deduce what knowledge the human is missing and what needs to be learned next. AI can also be used to determine the likelihood that specific knowledge will be needed in the near term future based upon past experience. AI learns that once students learn one area, they eventually will need to know another area.
AI could be used to update, add to, and modify courseware. For example,
· Adapt testing and assessment tools to optimize learning feedback.
· Reinforce learning. When a student leaves the material for a while, AI could prompt test questions and provide reminders to help the student get back up to speed faster.
· Make specific suggestions to the learner for self-improvement.
· Monitor student performance, learning patterns, and recommend areas for learner improvement to a tutor or teacher.
· Solve problems that the human programmer (teacher) did not anticipate.
· Optimize learning a specific narrow skill or group of modules monitoring many learners, what is learned, how long learning takes, and makes adjustments to speed learning.
· Reduce time wasted on activities that do not result in learning.
· Observe and report the pattern of effectiveness of different learning tools on different learning styles. It can vary the mix of how and when these learning tools are used.
· Learn from the real experience of learners. While an AI tool may not yet change it’s own algorithm, it may suggest patterns to the programmer/teacher that might result in new rules. Although the learning rules do not change, the learning program can “learn from experience”. Eventually it will be possible to create programs that change their own algorithm, i.e. they will learn how to learn.
Where can AI be applied with the most efficiency and effectiveness to improve learning platforms? The most significant use of intelligent machine assisted learning is whether it can solve learning problems and improve learning processes that the learner, teacher, tutor or programmer did not anticipate. It must solve learning problems that were never envisioned.
AI has already had a number of successes in various fields of endeavor. It is very clear that AI has already solved problems that its human programmers did not anticipate. Unlike checkers programs that were infused with human expertise from the very beginning, there has been work on a checkers program that learned to play the game by itself. The program is genetic in that it tries different winning strategies and plays these against itself. Then the program takes the most successful winning strategies and kills off those that were less successful. Many of the programmers were surprised by the results, noting that it played a better game of checkers than they did. Similar results were found in other types of applications.
Currently there are very few, if any, instances of the successful use of AI in learning and instruction platforms. Artificial intelligence has been under research and development for a generation or more but it has not yet been truly adapted to the learning environment. Higher education institutions must now focus resources on adapting AI to solve learning problems, some of which the teacher did not anticipate or notice.
One possible application of AI with machine-assisted-learning is to create a unique question and answer administrator similar to an FAQ (Frequently Asked Questions) administrator. This FAQ administrator would really be an EAQ (Every Asked Question) administrator. This EAQ administrator would be a sophisticated “help desk” for a particular course. The EAQ administrator might be the first step for students obtaining answers to course questions. It would maintain and update a list of the non-duplicated questions asked of students in a particular course from any number of teachers and any number of institutions. The FAQ administrator would refer ambiguity in questions or answers to the teacher. The FAQ administrator would analyze patterns of questions and use this information to suggest areas to the faculty needing content or structure improvement.
Today, the problem with FAQ is that the learner has to search to find the same kind of question to see if it has already been asked. This takes valuable time. The EAQ administrator would perform much of this function. The EAQ administrator would also be able to keep a catalog of all unique questions asked, even those asked infrequently. The EAQ administrator would be looking for patterns in student learning by which questions they ask, and when and how they ask them. The EAQ administrator could even develop better learning assessment tools based upon the questions asked.
AI can amplify highly interactive, motivating, learning environments, enabling students to continuously improve their critical thinking skills. For example, AI might suggest that the student make a reasonable estimate of the answer before solving a problem. Estimating is a very valuable critical thinking skill. AI might prompt the student when to use that skill and when not to use it, i.e. when an estimate is not going to be good enough. In fact, AI is most suited to reasoning and thinking and therefore is more suited to improving critical thinking skills than today's interactive but preprogrammed systems.
AI might also be able to be used to determine that a significant component is missing from the content, observe personal learning patterns, recognize the learner’s voice and habits (e.g. typing skill), determine when to seek more learner feedback, explain learner errors, find, clarify and resolve ambiguity, and so on. AI might be used as an intelligent learning scheduler to increase the number of students that can be effectively mentored by a single tutor.
AI might be used to effectively communicate between two totally different learning platform environments. Sharable Content Object Reference Model (SCORM) is being developed to provide standards for sharing content among different learning platforms. For example, a faculty member using WebCT might be able to use some module created by a colleague on WebBoard. While this will help cooperation and sharing, there are still going to be issues of finding and linking modules on different learning management systems and AI may be an effective supplement to this.
Institutions of higher education must focus resources and research on Intelligent-Machine-Assisted Learning. This is both the short term and long term solution to achieve accelerated learning. It is also the most fertile area for a university to achieve a significant competitive advantage.
It is certainly true that AI will not be able to replace the teacher for the near-term future and, perhaps never. However, AI can effectively amplify the capability of the tutor, teacher and mentor and therefore accelerate the learning of students. AI can also be used to directly assist the learner with more effective and efficient processes.
For the next couple of years provosts and presidents will be dealing with budget cuts and accrediting agencies that are emphasizing outcomes of teaching and learning. AI might be used to increase learning outcomes without breaking the bank. Certainly the cost of the AI technology itself would raise overall costs. However, increased research funding and increased student revenues may offset the AI costs. If a group of universities use AI to effectively accelerate student learning, then they might have an attractive advantage over other universities, everything else remaining the same.
It is in the interests of students to learn more efficiently and effectively. Better students may seek out universities with AI learning tools. AI might provide the competitive advantage necessary to increase the quality of the students.
If AI does accelerate student learning, it would become the most valuable competitive advantage. For example, if a student completes his/her education by taking the same required credits faster than at another university, then the university will be able to handle more students. Furthermore students would also be more satisfied. Lets assume that AI could increase learning acceleration by 20% over the course of a college education. A student that might now may take 5 years to graduate, might then take four years.
In the long term, universities without AI, will be marginalized and begin to be less of a competitor among learning institutions. A university is not only competing against every other university but also against every other type of higher learning opportunity. Customers, including students, will inevitably use technologies that improve their experience and increase their learning efficiencies and effectiveness.
This technology will not eliminate the need for faculty to teach. Such technology, will change the teaching role of faculty more to mentoring and coaching and away from lecturing. Faculty will necessarily become more involved with students and their learning.
Accelerated learning will eventually become the primary focus of all higher educational institutions. In order to achieve accelerated learning and remain competitive, colleges and universities must invest, research, and develop learning platforms that integrate artificial intelligence with the research findings of cognitive scientists. Cognitive science has already suggested ways that learning can be made more efficient. Artificial intelligence has progressed to the point where it can automate some, perhaps many of the processes suggested by the cognitive scientists. It is time to begin incorporating the findings of both disciplines into the practice of higher education.
THE END =======================================================