From Dean to Czar

I am preparing for a change in my career, see the story. I have been a dean at Sac State for 7 years, and for seven years before that, at two different institutions. For those of you hesitant to switch to the dark side and become an administrator, my advice is this: Do not listen to people complaining. These are actually great jobs, very stimulating, engaging multiple facets of one's personality, and can be very fulfilling. Most of it is about matching people with other people and resources, helping them to do something that is beneficial and good for the institution and for students. That, basically, is the job description. So when things work out, when you see a simple idea turning into a full-fledged program or a major or a project - wow, this feels good. Of course, when things do not work out, it is all due to unfortunate circumstances.

I had a great time in the Dean's office at Sac State. I worked with a very capable crew of people, most of whom are very independent and run things well if left alone. I like that, I appreciate those who can take charge and see a project from the beginning to an end. Together, we achieved quite a few things. We got out of poverty, created a bunch of new programs. We eliminated graduation gaps between URM and non-URM students. We were able to hire 50-75% faculty of color over the last 6 years. We straightened up a bunch of policies, built a consulting business, strengthened our partnerships, and improved our reputation. Of course, there is a list of things that did not work out; I am not going to dwell on it.

It is nice to leave on a high note. I think 7-9 years is the average shelf life of a dean, after which it is time to let someone else take charge. So I am very grateful to my colleagues, my leadership team, and staff of our college for their support, their incredible ethos, and the strive to always do the right thing. Thanks also for pushing me to improve, for preventing me from doing stupid things. I am still around, so this is not really a goodbye.

So, the vision for the national institute for AI in education actually belongs to President Wood and Provost Nevarez. They learned about my interest and sought to use this interest to advance the institution. This is a great turn of events for me, for I can both stay at a university I learned to love, and also get the change that I need. I have a whole bunch of ideas and am grateful for the opportunity to refresh without a loss.

.  

AI and neurodiversity

If AI were human, what would it be diagnosed with? Perhaps it would be Autism Spectrum Disorder (ASD). AI, akin to individuals with ASD, often struggles with social interactions and grasping emotional nuances. While they excel in specific tasks, abstract thinking or unpredictable social contexts pose challenges. Then there's Attention Deficit Hyperactivity Disorder (ADHD). AI can display ADHD-like traits: losing context in lengthy conversations or abruptly shifting focus. This metaphorical attention deficit mirrors the challenges individuals with ADHD face in maintaining long-term conversational coherence. Lastly, consider Executive Function Disorder. AI often falters when adapting to new, unstructured tasks, akin to the challenges faced by individuals with executive function disorder in organizing and executing tasks. AI's dependence on structured data and clear objectives limits its ability to handle open-ended scenarios.

Of course, treating every limitation as a diagnosis is ridiculous.  When building a relationship with AI, we should not pigeonhole it with human diagnoses. Instead, adopting a neurodiversity framework allows us to appreciate AI's unique cognitive makeup. This approach emphasizes focusing on strengths and working around limitations, acknowledging that AI represents a different kind of intelligence.

Neurodiversity is a concept and social movement that advocates for understanding and appreciating neurological differences as natural human variations, rather than disorders or deficits. Originating from the autism community, the term has expanded to include a range of neurological conditions like ADHD, dyslexia, and others. This perspective emphasizes that neurological differences should be recognized and respected just like any other human variation, such as ethnicity or sexual orientation. The neurodiversity framework promotes the idea that individuals with these differences have unique strengths and perspectives, advocating for accommodations and support systems that allow them to thrive in society. This approach shifts the focus from trying to "cure" or "fix" these individuals to celebrating and utilizing their distinct abilities, fostering a more inclusive and understanding society.

Understanding AI through the lens of neurodiversity offers an alternative perspective. We should not try to make AI closely mimic human intelligence; that would be counterproductive. Instead, we must consider embracing AI as a distinct 'other.' This approach allows us to benefit from each other's strengths and compensate for weaknesses. This approach will also reduce the anxiety about AI eventually replacing us. If we remain different, we will need each other.

In constructing our relations with AI, we can benefit from reflection on our species' internal diversity. This recognition paves the way for a more harmonious coexistence, where the strengths of one can offset the limitations of the other, creating a synergistic relationship between human and artificial intelligence. If we apply a strictly normative framework, trying to make AI exactly like the neurotypical human mind, we’re inviting trouble; the same kind of trouble human societies experience when trying to be more homogenous than they are.

Understanding AI through the neurodiversity lens offers a chance for growth and collaboration. It is not just about programming and algorithms; it is about building a relationship with a fundamentally different form of intelligence. This approach will enable us to fully harness AI's potential while respecting its unique cognitive characteristics. As we continue to evolve alongside AI, this perspective will be crucial in guiding our interactions and expectations, fostering a future where diversity in all its forms is not just accepted but celebrated. 

A case against prompt engineering in education

Do we give students examples of great prompts, or do we allow them to struggle with developing their own prompting skills? This dilemma is common amongst educators integrating AI into their pedagogical strategies.

Refining prompts is as a pivotal vehicle for cognitive advancement. It fosters growth by nudging students to navigate beyond their current capabilities. A meticulously crafted ready-made prompt, while yielding impressive results, might overshoot a student's zone of proximal development. The essence of learning lies in recognizing and rectifying flaws of the output. In other word, giving students a great prompt to begin with may produce the result that is painfully obviously flawed to the instructor, but the flaws are completely invisible to students. When students are handed sophisticated prompts, there's a risk of them becoming passive users, merely applying these tools without understanding or growth. Here is some empirical evidence of this provided by Jack Dougal. One of my colleagues, hopefully will soon present similar results.

The general principle should be to calibrate potential outputs to a level where students can discern imperfections. It is also to ENCOURAGE them to look for imperfections, guiding them to be critical to the output. Just because it sounds good and grammar is perfect does not mean the text is good. This approach encourages active engagement with the learning material, prompting them to question, adapt, and evolve their understanding. It's akin to guiding someone through a labyrinth; the instructor's role is to provide just enough light to help them find their way, without illuminating the entire path.

In the educational sphere, the prompt industry's role is contentious. While it offers a plethora of ready-made prompts, enhancing efficiency, this convenience comes at a cost to cognitive development. In academia, the journey of crafting and refining prompts is crucial for fostering critical thinking and problem-solving skills.

On the research front, the prompt industry does contribute valuable insights, empirically testing and refining prompts to optimize AI interactions. I love to find out about the chain-of-thought approach, for example. However, a significant portion of the prompts available in the market are of dubious quality. These prompts, lacking empirical validation, are frequently oversold in their capabilities. The indiscriminate use of these untested prompts can result in suboptimal outcomes, reinforcing the necessity for a discerning approach to their adoption and application.

The overarching promise of AI lies in its potential to democratize content creation, designed to comprehend natural, imperfect language and provide equitable access to all, regardless of their mastery of writing mechanics, their disability, or fluency in the dominant language. This vision is threatened by attempts to monopolize and professionalize access to AI, a trend that runs counter to the very ethos of this technology. The notion that one must know 'magic words' to effectively communicate with AI is a form of self-interested deception. It undermines the inclusive and accessible nature of AI, turning it into a gated community where knowledge is unfairly hoarded rather than shared. Vigilance against such practices is essential to preserve the integrity and egalitarian promise of AI, ensuring it remains a tool for empowerment and collective advancement, rather than a vehicle for exclusion and profiteering.

Is AI doing too much for students?

Educators’ worry about AI boils down the concept of 'Goldilocks zone.' A learning task should neither be too challenging nor too simplistic, but just right, fitting within the learner's zone of proximal development. It is something that the learner can first solve only with help, but eventually internalized and can solve on their own. The concern is that AI, in its current form, might be overstepping this boundary, solving problems on behalf of learners instead of challenging and guiding them. It is like that rookie teacher that keeps solving problems for students and rewriting their papers, and then wonders why they have not learned anything. I just want to acknowledge that this concern is very insightful and is grounded in both theory and everyday practice of teachers. However, the response to it isn't that simple. AI cannot be dismissed or banned based on this critique. 

First, there's the question of what skills are truly worth learning. This is the most profound, fundamental question of all curriculum design. For instance, we know that certain basic procedural skills go out of use, and learners leapfrog them to free time to concentrate on more advanced skills. For example, dividing long numbers by hand used to be a critical procedural skill, and it is not worth the time, given the ubiquity of calculators. There is a legitimate, and sometimes passionate debate whether the mechanics of writing is such a basic procedural skill that can or cannot be delegated to the machines. I don’t want to prejudge the outcome of this debate, although I am personally leaning towards a “yes” answer, assuming that people will never go back to fully manual writing. However, the real answer will probably be more complicated. It is likely that SOME kinds of procedural knowledge will remain fundamental, and others will not. We simply do not have enough empirical data to make that call yet. A similar debate is whether the ability to manually search and summarize research databases is still a foundational skill, or we can trust AI to do that work for us. (I am old enough to remember professors insisting students go to the physical library and look through physical journals). This debate is complicated by the fact that AI engineers are struggling to solve the hallucinations problem. There is also a whole different debate on authorship that is not quite specific to education, but affects us as well. The first approach is then to rethink what is worth teaching and learning, and perhaps focus on skills that humans are really good at, and AI is not. IN other words, we reconstruct the “Goldie locks zone” for a different skill set.

The second approach centers on the calibration of AI responses. Currently, this is not widely implemented, but the potential exists. Imagine an AI that acts not as a ready solution provider but as a coach, presenting tasks calibrated to the learner's individual skill level. It is sort of like an AI engine with training wheels, both limiting it and enabling the user to grow. This approach would require creating educational AI modules programmed to adjust to the specific needs of each user’s level. We have the Item Response Theory in psychometrics that can guide us in building such models, but I am not aware of any robust working model yet. Once the Custom GPT feature starts working better, it is only a matter of time for creative teachers to build many such models.

Both approaches underscore the importance of not dismissing AI's role in education but rather fine-tuning it to enhance learning. AI is here to stay, and rather than fearing its overreach, we should harness its capabilities to foster more advanced thinking skills.

These are conversation we cannot shy away from. It is important to apply some sort of a theoretical framework to this debate, so it does not deteriorate into a shouting match of opinions. Either Vygotskian or Brunerian, or any other framework will do. Vygotsky has been especially interested in the use of tools in learning, and AI is just a new kind of tool. Tools are not note all created equal, and some are better than others for education. The ultimate question is what kind of a learning tool AI is, and whether we could adjust learning, adjust the tool, or do both.