Starting Over (1)
A roadmap for an AI-allied course
This letter embarks on a first step along the path to rethink the academy from the ground up: re-building a single course.1 Courses are the basic unit from which the university is constructed.2 With the bigger picture in mind, I sketch out a roadmap to prepare teaching in the fall term – and we will implement this over the coming months.
The big picture
Generative AI is is transformative. Just like with any other transformative change, generative AI is also disruptive at first. One could manage this as radical transition – tearing down the old to let the new emerge, I would call this the Phoenix Model of transformation. Or one could rearrange parts, rebuild, reimagine, reconfigure – the result can be just as radically new, but it is achieved on a trajectory that protects continuous operations, and remains committed to consensus and collaboration. I will call this the Chrysalis Model.3 This is the strategy that we pursue here.
The contours of what we are up against are appearing:
The nature of employable skills is changing.
What students learn is changing. This includes changing objectives: many technical tasks can be delegated, but to stay in control, we need renewed emphasis on the context of our teaching. As well, our students need to learn to plan, to critique, to validate, and to assess – significantly more than they did in the past. This also includes new contents, for example new challenges of technical literacy have appeared, such as “effective prompting”. And, of course, old knowledge remains relevant.
How students learn is changing. Personalized tutoring is now possible. Allowing students to progress along individualized timelines is now possible. That means that a move to mastery-based learning4 is now possible. Bloom’s two sigma problem5 may find solutions.
What students do is changing. Writing is no longer a good proxy for thinking – but thinking skills still need to be developed.6 Overall, assignments will become more process-based rather than focus on single, final results.
The nature of assessment is changing. We need more direct measures of how our students succeed with their objectives – proxy measures are no longer reliable. But assessing learning directly, just as assessing the process of creating submitted work, does not scale well and new approaches are required.
Finally, even though we are working within the scope of a single course, we need to keep the big picture in mind. This includes aligning the course with institutional policy (or updating the policy), establishing communities of practice, and engaging with stakeholders at multiple levels to ensure the value of our changes is recognized. Engaging with society on this topic not only acts outside the academy, for example to influence the way workplace requirements are changing, but it also becomes a central proposition for student recruitment.
This is a lot to consider and could fill many books. Moreover we are definitely at a stage that is far from equilibrium – the possibilities and constraints, the paradigms and the fears seem to be changing daily. We need to integrate many voices, many values, while keeping the fundamentals of education in focus. Experience tells us, the solutions will not be technological – they never are. But none of this will work without technology.
And in the end, all this thinking needs to crystallize into two or three pages of a course syllabus.
We need to get specific.
I will re-imagine a course that I plan to teach in the fall term, from the ground up, based on everything we know so far about generative AI, while using our best estimates for its developing impacts, and our best predictions for future changes.
The course that we craft will serve as a model for what can be done, a model that can help others in their own work, either by adopting some of the patterns, or by figuring out what they don’t like and how to go their own way. This needs to happen on a compact timeline, so that it may be useful to colleagues who need to review their own fall-term materials and rebuild their courses.
In fact, we will work through two courses – one from the sciences, and one from the humanities.
Computational Biology Foundations
Computational Biology Foundations introduces first-year university students to a domain of research that they probably have never heard of before. Computational biology is more than a set of methods, it is a conceptual approach in which we apply computational methods to questions that at first do not appear very computationally tractable at all.
Computational biology is especially suitable as a model for our needs, because it is quintessentially interdisciplinary, and in the way I teach it, there is a heavy emphasis on conceptual understanding. Concepts however are our way to understand reality through constructs of language – and such understanding is what Large Language Models have learned to do as well. But the course is not entirely conceptual either, a fair amount of factual knowledge is involved, as well as meta-skills like computer literacy, documentation, and time management.
As a result, Computational Biology Foundations aims to introduce a particular way of thinking that will benefit students in whatever direction their career takes them. This makes it especially suited as a course in the first or second year of university.
Design Philosophies I – Epistemology
However, Artificial Intelligence will not only affect STEM courses, and I am very fortunate to work with Dr. Yi Chen who is redesigning a course on Design Philosophies, to add this course to the roadmap. We have collaborated in the past, and this is an exciting new project: the first part of a series that develops the major domains of philosophy from works of art, not as lenses through which to examine art, thus closing the hermeneutic cycle through cultural practice.
This concept too is quintessentially interdisciplinary, but its “mechanics” were conceived along the lines of the typical humanities course: assigned reading, classroom discussion, and personal reflection in written form. These are also the aspects of teaching that many feel are the most vulnerable to generative AI, and we will need creative and constructive approaches. In fact “Nobody plagiarizes in their diary.” were Yi Chen’s words – so we will take this as a challenge: a course that becomes a genuine expression of self, as much as a diary is.
A project without an endpoint is only an activity. The endpoint for us is the the milestone at the end of the first week of July and by then (a) theory needs to be in place; (b) infrastructure needs to be available; (c) lectures and other content need to be mapped out; (d) final syllabi need to be done; and (e) we have presented and discussed the results.
We structure the roadmap into four sections:
Contents: Early on we need to decide on an educational framework to work with; there are many to choose from but we will start from Knowledge Building (cf. Bereiter 2014) – its constructivist roots appear especially well suited to the kind of co-creation of knowledge that is implied in having the computer think with you, not for you. The next major task is to review learning objectives, in particular to envision how the objectives will change, and how course contents needs to adapt. Course contents, and course activities will likely develop in a backward course design paradigm, working from outcomes to stimuli. Finally this all needs to come together in a mastery-based ordering – perhaps with parallel options. This constitutes our first milestone.
Learning: Support for different learning approaches should be generic enough so that solutions can map to other courses and to different environments. This begins with a recognition of diverse learning styles, and the need to support materials in different modes. We have discussed a proof of concept in Assignment and Agency, a few weeks ago (▷), and this will contribute to the requirements for the kind of personal tutoring, and individualized progress that we think is promising. This will lead to our second milestone. In this phase we also need to start defining required infrastructure, which will be built, tested and deployed in the following phases. Requirements will need careful thought but will likely include: (i) that resources can be installed in a variety of environments – intramural, or in the cloud, to keep a nimble profile; (ii) that non-specialists can use and customize the resources; (iii) that cost is minimized through use of appropriate APIs; (iv) that there is support for content creation in courses; (v) that there is support for the documentation of process; (vi) that features are added in a rapid life-cycle; (vii) that robust fall-back strategies exist; (viii) that developments are open-source and proceed through shared effort; etc. etc.
Assignments: In principle, assignments should be evident from the designed contents, but there are recursive interactions between objectives and activities to consider. Since major tangible progress that can be talked about should be apparent after the second milestone, the third phase may be a good time to start building local community of practice networks. The list of assignments will form the third milestone.
Assessment: The morning fog has not quite lifted on that area. We have discussed some of the requirements above, but how assessment in the new course paradigms can scale is still one of the most important problems to solve. My thoughts on that don’t quite fit into the margin … yet. But defining rubrics and assessments will comprise the fourth milestone and that must also include the prototyped infrastructure.
Final touches: This short phase will be occupied with final touches, and documentation. Sanity checks need to be performed, perhaps feedback from a focus group can be organized – expectation management is a crucial aspect of syllabi. The fifth milestone is The Syllabus.
Communication: The form of events in the communication phase is to be determined, but options include lectures, workshops, writing, and more. Completing this places our final milestone at the end of the first week of July.
All engines ahead full.
The academy is experiencing a transformative change due to generative AI, and this letter presents a project plan – part of a Chrysalis Model of transformation – to reconstruct courses while preserving continuity. Employable skills, learning objectives, and assessment methods are rapidly evolving. In response two courses will be openly redesigned: one in the sciences (Computational Biology Foundations) and one in the humanities (Design Philosophies I - Epistemology). These courses will serve as models for adapting education for the new era of widespread adoption of AI assistants. Our roadmap comprises four sections: contents, learning, assignments, and assessment, and it concludes with lectures and/or workshops to share insights and experience. By embracing technology, integrating diverse perspectives, and fostering widespread collaboration, we aim to start building the academy of the future from the ground up.💡
BEREITER, Carl and SCARDAMALIA, Marlene (2014). “Knowledge building and knowledge creation: One concept, two hills to climb.” In S. C. Tan, H. J. So, J. Yeo (Eds.) Knowledge creation in education (pp. 35-52). Singapore: Springer. (▷)
BLOOM, Benjamin S. (1984). “The Search for Methods of Group Instruction as Effective as One-to-One Tutoring”. Educational Researcher 13(6): 4–16. (▷)
Feedback and requests for interviews, speaking, collaborations or consultations are welcome at firstname.lastname@example.org . 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 (1): A roadmap for an AI-allied course”. Sentient Syllabus 2023-03-28 https://sentientsyllabus.substack.com/p/starting-over-1 .
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, for which I take full responsibility.
Future models of the academy may look very different – one could conceive a return to an apprenticeship model, or pursuing a program based model that is not subdivided into course units. But since our current universities are organized around courses – for a number of reasons – implementing gradual, evolutionary change is the most realistic strategy.
In the life cycle of holometabolic insects, such as butterflies and moths, the chrysalis stage is the phase of radical transformation during which the larva becomes an adult. The pupated insect rearranges its entire body structure to prepare for its adult life. This dramatic change involves breaking down most of the larval tissues and reorganizing them into new adult tissues and body parts. This process results in a complete metamorphosis; the adults emerge as an entirely new creature, with different physical characteristics, including wings, articulated legs, reproductive organs and stingers, the ability to fly, and, in some species, extraordinarily complex social behaviour. Intriguingly, this is – at least to a degree – compatible with a retained identity, and memories of larval experience shaping adult behaviour.
Mastery-based learning emphasizes the mastery of a particular skill or concept before progressing to the next level or topic. Typically, students progress at their own pace, thus learning is more effective. In addition, this strategy prevents students from skipping over important parts and there is less risk of them becoming completely lost along the way.
Bloom's Two Sigma Problem refers to Benjamin Bloom's observation that students who received one-to-one tutoring in a mastery-based curriculum performed two standard deviations, or "two sigma," better than students who were taught through conventional classroom education. The problem is to find ways to replicate the benefits of one-to-one tutoring and individualized learning in a way that scales to the educational realities and remains affordable (cf. Bloom 1984).
Some writers have proclaimed a risk of atrophying thinking skills, once knowledge is outsourced to AI. Analogies are drawn with technologies that include pocket calculators and their effect on simple arithmetic, GPS and its effect on orientation skills, the way that autocorrect aids degrade spelling, and the loss of cursive handwriting.