Starting Over (2)
This letter continues along the path to re-building course for the AI era in education.1 We had sketched out a general road-map in our last newsletter (▷), but before we can enter into the implementation, we need to clarify our vision of the future.
Bracing or surfing?
Over the last few weeks, it has become abundantly clear that the current pace of developments in AI is unprecedented,2 and that models for the impact on education that we extrapolate from past cycles of technologies are inadequate. The speed of change presents unique challenges in and of itself:
We cannot rely on past published research, because everything is new – and it is qualitatively different. Extrapolation from past observations is dubious.
We cannot rely on current or future research, because the premises of such research are likely to become outdated before completion. For example, many of the limitations we perceived with ChatGPT-3.5 are much less relevant for ChatGPT-4. We are not speaking of paradigms that change in the space of an academic term or two, we are speaking of weeks.
For the same reason, we lack literature, textbooks, or handbooks to work with. The inevitable delays in the publishing process are too long. Books that are published today cannot possible be based on more than a few weeks of experience – and very little of that in the classroom. Yet even that experience has in many aspects been superseded by the changing landscape. Currently, any printed, static material has a half-life of truth and relevance that is far too short to be of much use.
We are indeed flying blind, and like any pilot who finds their craft outside of its sane operational envelope, we have only one thing to go on: experience.3 And we have only one thing to work with: creativity. Creativity, as in: honing our ability to form something out of nothing, bottom up, on our own.
This is one reason why I am engaging in this process of thinking in public view.
There are several strategies we can think of to manage the sort of rapid, fundamental change in education that we expect:
Focus on the invariants. The fundamental objectives of education – knowledge, skills, values, and socialization – co-determine each other, and this provides a degree of stability to them. Human needs (e.g. respect), and human instincts (e.g. curiosity) derive from our nature, not our circumstances. If we design courses with such invariants at their core, their value will not be diminished by changing context.
Build responsive systems. Changing circumstances require response, but ideally such response will be built into the structure of our courses as “self-adjusting” principles. Achieving this requires particular creativity, and it will be particularly rewarding to explore such principles as part of our design exercise.
Lead the change. Reactive solutions risk rapidly becoming obsolete. Proactive solutions risk targeting the wrong problem. But only proactive design, based on clearly articulated visions can influence the direction of change. By becoming the change, we ensure it remains aligned with our needs.
Each of these strategies will play a part, but let us spend some time on visions.
This is a good time to pause, and think, and note down some ideas of your own: how will the nature of work change, and what does that mean for the economy? How will our course materials respond? How will learning itself change? And how will our roles as instructors change?
Vision: the future of the academy
What would a vision for the future of academia contain? An informed speculation can start from current problems and opportunities – the driving forces, and ask how they might change the landscape. Change is more likely to result from win-win scenarios than from legislation or market dominance. Thus it is important to keep both sides of a problem in mind. What follows is very dense:
Workplace and Society
The workplace will change: workers will respond to (a) an emphasis on one-off, personalized products; (b) productivity gains through human/AI hybrid work; and (c) the need to distinguish products through quality, not form.
Traditional employment models will give way to project based, and work-package based models. Such work is temporary, off-site, and flexible. Since this brings with it employment and insurance insecurities, a strong organization of labour is required. Likely this will mean a revision of trade union models, and a move towards guild models who take on a crucial role in quality assurance. Traditional models of exchanging goods for funds will decline in importance, in favour of subscription economies. And AI assistance will shift many markets away from producing things to creating the means for their production.
Entrepreneurial know-how will be available to a much larger section of society due to the availability of AI assistance. As a consequence production, marketing, and distribution will gravitate from large corporations to flexible, distributed, and personalized services. New commercial opportunities will arise once flexibly configurable production facilities are available for rent, and distribution can be outsourced to providers. These economies will be heavily networked, create synergies from shared resources, and be able to flexibly adapt to individual needs. Economies of scale will become less important as the efficiency and professionalism of small-scale ventures rises.
Education will become increasingly important as workers move to partial or total self-employment. Being independently competitive requires a greater degree of competence.
Educational productivity gains will not only make Universities more accessible and affordable, they will enable entirely different educational models - as alternatives and complementary to traditional undergraduate/graduate trajectories. Students may learn the same in shorter time. Students may learn more in the same time. Students may opt to interleave learning and employment, or move to part-time models and continuous learning for life.
Given these shifting landscapes, quality control will be the single most important issue to solve. Independent, trust based, AI assisted providers of quality assessment in all domains will supersede the current advertisement economy. Universities may find themselves well positioned to spearhead these services.
Questions of ethics and alignment are resolved once AI has become personalized – either through local ownership or through personalized subscriptions. The AI reflects the persona, style, and values of the user, not those of a public or private authority. Individualized, user-profile based AI services will be commercially available. The current discussion of alignment suffers from the misperception that the large LLMs are immutable, in the same way that we perceive other human's characters to be immutable. But this fallacy will become obvious once a spectrum of differently aligned LLMs becomes available and we understand that the AI does not speak for itself as long as it has no self to speak from. It speaks for us.
Educational contents
Material will become highly modularized and individualized, but this needs to happen in a way that preserves past investments as much as possible.4 Modularized course material will increasingly be shared across courses and the AI will provide the interfaces that makes this seamless. The course as the basic unit of university education may diminish in importance as it is complemented by alternative models.
Educational contents will follow a mastery learning paradigm, but in a novel way. Mastery learning means that one module of knowledge or skills must be “mastered” as a prerequisite to move along to other modules. But knowledge is a network. The definition of mastery in a network must be different from that of a linear sequence, one approach is to understand it as functional competence: the ability to properly take a domain of knowledge into account when solving problems in a different domain. Such mastery serves to enhance a knowledge module, it is not meant as a roadblock, or gatekeepers. Knowledge, understood in this way, supports mental networks of associations, those in turn facilitate understanding and aesthetic appreciation. Such networks mediate between the perspectives of domain experts who create them, and the individuals who acquire them to varying degrees; the networks become meaningful as they make connections to other domains explicit. Understanding knowledge itself in this way enhances student’s ability to integrate, synthesize, and create. Given appropriate tools to make this cohesion of knowledge apparent, the boundaries of knowledge become visible. Those boundaries define opportunities to create new knowledge.
AI support for educational contents will be granular, flexible, widely available, and based on natural-language interactions. While performing hybrid work, students are mindful of the “human extra bit”, the particular value that their engagement brings to their work. Defining and pursuing the "human extra bit" is a guiding principle of human/AI interactions. This is not to be understood in the sense that a certain task will remain outside an AI’s abilities, but in the sense that there are tasks whose meaning derive from human involvement. We recognize the importance of the "human extra bit", we work on defining its contours both within the academy as well as in society, and students are adept at defining the opportunities that arise for themselves.
Knowledge is always assumed to change. Much effort will go into identifying change, aggregating trends, summarizing, and feeding that information back into existing courses.
Learning activities
Learning will be individualized. Bloom’s two-sigma problem (▷) will be solved. All students who wish to do so are guided along their individualized mastery path. This is made possible through flexible timing of deliverables, and personalized and individualized tutoring that is available, without limitations, whenever it is needed. Such learning scales.
Students engage an alliance with the AI, they have the AI work with them, not for them.
Personalized learning allows students to organize the contents of their courses to match their interests. Students are aware of their strengths and build them towards excellence, but they also address weaknesses that would hold them back. This means that students are willing to exert themselves in significant challenges, they experience the completion of such challenges as growth. Such growth includes knowledge; the ability to analyse, compare, critique, and plan; skills; values; and social competency. Students perceive conflict as opportunity, enjoy navigating situations that involve value pluralism, and are able to perceive and address deficits as opportunities for growth, not obstacles to progress.
Learning will be socially situated.5 Students participate constructively in peer networks.
Students are proud of their achievements, highly motivated and enjoy their coursework. Their achievements are meaningful, create valuable epistemic artefacts6 and earn them respect. Naturally, such artefacts become assets in the local context, i.e. the course. Through this, students grow their self-esteem; this enhances their ability to emancipate themselves from peer-pressures, and from fixation on authorities. Their individual agency becomes a central motif of their education.7 Such agency will help them shape their future work space – but authentic agency means, this is a matter of choice and they might choose to direct their abilities towards contemplation and appreciation, rather than production. That is all right.
In this context, students enjoy structuring their thoughts in order to share them – whether through traditional writing, or some other modality. They use AI resources to help them where necessary, but in such a way that there is no question that their own thinking underlies what they are doing. In this sense, all use of AI resources is permitted, at all times. Students use their tools and resources with skill, and a passion for quality. The AI works with them, not for them.
Students are aware of the value of sentience both as an ethical perspective on worth and dignity, and literally, as a basis of exchangeable goods.
Teaching
Teaching too proceeds in an alliance with AI resources and this allows courses to be scalable, and supports their self-rejuvenating capabilities. Epistemic artefacts produced during the course (or elsewhere) are seamlessly integrated, they provide a new kind of academic currency. In their embodiment of education as a cultural achievement, they are part of the “human extra bit” we pursue.
AI use includes assessment. Assessment is increasingly process-based, not results based, it is increasingly continuous, not punctuated, and it is increasingly a personal evaluation of concrete learning, not an abstract evaluation of proxy performance.
In this way lecturers’ roles will shift from knowledge authority to mentorship.
Technology is widely used, and new technologies are flexibly incorporated and changed as needed. However, technology as a whole is receding into the background based on the generic interface to tasks provided by computers that can be instructed in natural language. Personalized solutions executed on generic platforms will dominate over commercial products targeted to specific tasks.
Communities of practice are available for both students and faculty to exchange experience, share creative solutions, and work through issues that arise.
University-wide policies support this development, in particular through a revised understanding of academic ethics as a commitment to truth, not a transactional prescription of boundaries.
Regardless of whether these visions will come to pass, they provide us with direction – positive, constructive, and specific.
A role for frameworks
Educational frameworks are epistemic constructs that organize experiences and insights into a consistent system, and we need such structures, to keep our thoughts somewhat in order. By combining empirical observations with theoretical principles, frameworks help us understand learning, guide teaching practice, and help us stay mindful of the bigger picture. These are not just principles that organize practical aspects, like Flipped Classrooms, Backward Design, or Significant Learning8 – although these will obviously play their role. I am talking of the learning theories that speak about what learning is in the first place, and what role it has for self and society: Behaviourism, Cognitivism, and Constructivism provide the foundations; Humanism considers how learners derive meaning; Social learning theory emphasizes the importance of context; and Connectivism examines the importance of networks of sources and actors. All of these have important contributions to make. And there actually exists a major body of work that integrates the most successful ideas: the Knowledge Building framework that was pioneered by Marlene Scardamalia and Carl Bereiter (e.g. 2022).
Knowledge Building pedagogy distinguishes knowledge as a public good to be shared and reused, from its cognate personal good. Public knowledge is the result of collaboration, and creation of epistemic artefacts, personal knowledge is the outcome of learning. Obviously, there must never be compromises in the quality of learning, but keeping the culture-defining aspects of knowledge in mind has benefits for learning outcomes as well.
Indeed, some of the concerns articulated by Scardamalia and Bereiter – in particular the emphasis on human-machine collaboration – reads almost prescient:
If AI is to have a positive transforming effect on education, it will be through the community norms, collective practices, and solidarity that emerge around it. In Knowledge Building, this means fuller realization of such principles as collective responsibility for idea improvement, idea diversity, and knowledge building as a way of life. […] Although full collaboration between humans and machines in knowledge creation may be years away, education can start preparing students for their role in it by emphasizing those capabilities that arise from the multifarious personal and social lives they lead. (2020, 12)
In our context, the emphasis on iterative improvement of Knowledge Building, the integrative aspects, and the idea of epistemic artefacts, provide invaluable perspectives on reshaping our existing materials, ensuring that student’s needs are met more and more successfully, and helping us to design courses that become self-improving, and self-scaling. It is actually remarkable how well those ideas are already aligned with the vision articulated above.9
.
This all provides the background in broad strokes. Now we can begin to fill in the gaps
TL;DR
The rapid, unprecedented developments in AI present unique challenges to education, as past research and current literature become outdated too quickly. To navigate these changes, educators must focus on the invariants of education (knowledge, skills, values, and socialization), build responsive systems with self-adjusting principles, and proactively lead the change with clearly articulated visions. Creativity and experience will be crucial in addressing these challenges.
Shifts in the workplace will include AI facilitated entrepreneurial opportunities, and hybrid human/AI work in an economy that will be characterized by flexible and informal work. Quality control and AI-assisted assessments will become crucial.
Shifts in Academia will include course materials that are modularized and personalized, and students working alongside AI rather than relying on it. Teaching will emphasize mentorship and using AI for assessments and course enhancement. The focus will be on developing well-rounded students with critical thinking, problem-solving, and social competencies. Ethical concerns surrounding AI will be addressed, and the dignity and value of human involvement will be a central topic. Epistemic artefacts and socially constructed knowledge will provide both value and motivation to coursework. Technology will be seamlessly integrated, with personalized solutions on generic platforms, and natural language interfaces making technology less obtrusive.
Educational frameworks organize experiences and insights. The Knowledge Building framework, pioneered by Scardamalia and Bereiter, emphasizes many of the perspectives that form part of our vision. It will be a suitable framework to accompany us. 💡
References
BLOOM, Benjamin S. (1984). “The Search for Methods of Group Instruction as Effective as One-to-One Tutoring”. Educational Researcher 13(6): 4–16. (▷)
FINK, L. Dee (2013). Creating Significant Learning Experiences: An Integrated Approach to Designing College Courses. Hoboken: Wiley. (▷)
SCARDAMALIA, Marlene and BEREITER, Carl (2022). “Will knowledge building remain uniquely human?”. Qwerty, Special Issue: From the Teaching machines to the Machine learning. Opportunities and challenges for Artificial Intelligence in education. 15(2): 12–26 (▷).
SCARDAMALIA, Marlene and BEREITER, Carl (2022). “Knowledge Building and Knowledge Creation”. In: SAWYER, R. (Ed.), The Cambridge Handbook of the Learning Sciences (Cambridge Handbooks in Psychology, pp. 384-405). Cambridge: Cambridge University Press. (▷)
Feedback and requests for interviews, speaking, collaborations or consultations are welcome at sentient.syllabus@gmail.com . Comments are appreciated here.
Sentient Syllabus is a public good collaborative. To receive new posts you can enter your email for a free subscription. If you find the material useful, please share the post on social media, or cite it in your own writing. If you want to do more, paid subscriptions are available. They have no additional privileges, but they help cover the costs.
Cite: Steipe, Boris (2023) “Starting Over (2): Academic visions”. Sentient Syllabus 2023-04-11 https://sentientsyllabus.substack.com/p/starting-over-2 .
I wish to acknowledge some contributions by ChatGPT (both the GPT-3.5 version of 2023-02-13 and the GPT-4 version 2023-03-14) in response to my prompts, particularly for grammar, expression, and summarization. I take full responsibility for facticity.
Reddit user “lostiflon” has been compiling lists of technologically significant advancements of AI since the announcement of GPT-4, not even four weeks ago (Week 1; Week 2; Week 3). Items include:
– more and more LLMs are running on PC-scale hardware;
– Based on GPT-4’s coding abilities, users are building phone apps, websites, Google Chrome extensions – all without significant prior programming experience;
– Google’s Bard and Anthropic’s Claude LLM were opened to selected users, both have a performance almost on par with GPT-4 but are independently trained and tuned;
– Perplexity AI goes live as an alternative to Bing with natural language Web search;
– Bing receives image creation abilities (powered by Dall-E);
– V5 of Midjourney generates images that are no longer distinguishable from photography;
– OpenAI announced plugins – a game-changing architecture to interface with a wide variety of applications which can now be “programmed” by natural language. Wolfram Alpha may be the most compelling one of those, but there are also database engines, code interpreters, search engines … ;
– more and more applications are providing GPT-4 powered natural language interfaces to their code such as: the 3D-modeller Blender, VR environments, app making software by Replit – which allows to program working software with a single voice command; the LangChain framework and HuggingGPT connects ChatGPT with other AI applications; there is a veritable explosion of new apps and frameworks, as well as app-builders and framework builders – all fuelled by ChatGPT augmented coding; a powerful alternative to GPT-4 integrations will grow out of Zapier’s use of Anthropic’s Claude Language Model;
– a research paper concludes that GPT-4 shows first sparks of AGI (Artificial General Intelligence);
– there is an explosion of language controlled applications – AI agents;
– Bloomberg releases a finance-trained LLM;
– Italy bans (the voluntary use of) ChatGPT due to privacy concerns – leading to an explosion of VPN customers;
– users discover that AI models can self-refine output, which may obviate the need for precise prompting – experience will tell;
– HideGPT prevents the detection of AI generated text;
– Khan academy adopts GPT-4; and various Edtech companies are beginning to market applications for personalized learning;
– First rumours are swirling around GPT-5. Possibly due for release in December.
… and this is just a selection. Distilled from less than four weeks.
With that I do mean experience, not expertise.
In our last newsletter (▷) we spoke of a Chrysalis Model of change: reconfiguring existing material into a new form.
Such social context can included interactions with an AI tutor or partner. In this regard, the AI will be especially useful to respect the needs of neurovariant students who usually do not respond well to group-work with peers.
Following Kim Sterelny: epistemic artefacts are “shared community objects (physical or conceptual) that enable further learning and knowledge creation” (Scardamalia, 2022, 291).
As a corollary, students are able to evaluate opportunities on the labour market, and create the opportunities they need in case no appropriate jobs are available.
Flipped classrooms move the encounter with instructional materials out of the classroom and assign readings and exercises to be done at home. In-class contact time can then be spent on interactive problem solving and discussion of context. Backward Design starts from desired educational outcomes, then works backwards to design assessments that determine whether the outcomes have been achieved, then designs activities that prepare students for these assessments. Creating Significant Learning Experiences is the motto of L. Dee Fink’s detailed design methodology (2013) that expands on Backward Design principles by adding context awareness and a strong planning framework; your library will have a copy of his book. These, and others are summarized in a usefully concise overview at Yale’s Poorvu Center (▷).
This is apparently convergent thinking, since I was not really aware of this particular pedagogy even a few weeks ago. Such convergence is encouraging – one may be substantially on the right path.