A lot. Of course. Ask the various start ups that are betting futures upon it. Ask the students and teachers that have already adopted so many applications of AI to make their life easier. Ask the skeptics who have to concede that it does make things work faster even if it bypasses process , brain or human engagement. It is the future and it is here. There is no choice but to embrace it. With caution and not fear.
First: The fears. There have been many…and they are valid. The fear of bias is real and proven. The fear of hallucinations, of the AI – being generative – giving responses that are not true. The fear of ethical and moral bounds being crossed in ways that are totally unacceptable. (Witness the grok incident just this week with so many being unclothed by the AI). We may speak of bounds, guardrails and controls but as with every other technology, the gatekeeping will lag the road being built.
The key concern for me is when the AI bypasses all human engagement and yet succeeds in checking the boxes that were designed to be proof of process that helps humans upskill and learn. We have enough and more instances of AI generated assignments being done by AI assistants for students and then being assessed by AI on behalf of tutors or teaching assistants. In this process what was bypassed was the actual learning. This makes nonsense of our older systems and completely defeats its very purpose. Worse, we are dumbing down while thinking we are being clever. This is where we need to engage wisely with AI both at the individual level and as a system. “If we seek to help our students, we should not be making them weaker in the long run. An AI deployment that builds their intellectual muscle is the goal. We are currently delivering lazy sop that undermines the mental muscle, that cedes agency to the algorithm, disassociates work from process, and disintermediates the human from their own possibilities.”(1)
Where does it work?
In administrative efficiency. It is wiser to enable AI to perform repetitive tasks for administrative efficiency. For example, basic response letters for routine situations. Or simple analytics that flag – for example – student attendance in class, predict trouble ahead and generate appropriate communications and intervention measures. Simple sensors and robotics combined with AI can monitor for safety, well being and regular maintenance. Basic ERP analytics can be made predictive using AI in order to flag problems and opportunities early enough to do something about it. This is where our systemic efforts must be focused. Efficiency is both about costs, delivery and about reliability in this case – and success here must be measured on all three factors.
AI also does well in specific study modes. For example, multiple ways to explain a concept to a student who does not get it. Humans do not have the patience or possibly the range of ways to communicate the way an AI might – since it is a collective of prior human expression. Customised learning for problem areas identified is a prime target for AI intervention before escalating it to the next level. Another area where AI does well is adaptive test based question answer driven pedagogies that train students to take exams. This is where AI can not only design question banks, but also rank them and deliver them to enable adaptive learning. This involves analysing the responses, identifying gaps and setting up tiny tutorials right there to resolve the problem. These, and a few other basic pedagogies are ready for AI adoption at scale.
The true wonder of AI in education comes from the decoupling of systemic from scaled.
Pre AI, any response for tens of thousands, or even lakhs of students would be designed into a system where each student would roughly have the same input in terms of content, structure, standards and even assessment. The only saviour to customise learning would be the teacher who was expected to be responsive but often did not have the time or resources to deliver, or a genius head teacher who would enable a rare culture that actually nurtured each student to their individual potential. This was as rare as genius is rare – possibly, but an unlikely daily encounter. AI breaks through this barrier.
The AI opportunity for education and skills lies in diagnosis, delivery and destinations.
Diagnosis: Or rather – prognosis. AI has the capacity to trawl through both large data and small individual data to be able to seek and match patterns that predict success in specific ways. These have already been sold to schools and parents in order to help them ‘stream’ their children…and since they already exist, we know the technology part of it is feasible. Now we need to refine it for success, for appropriate guidance (not the nursery school results to NASA astronaut predictions we see today!) and for planning ahead to support the potential. The first goal of an education system has always been to spot potential – and now with AI we can. This will be prone to errors for a long time, and we should have decades of discussion ahead about A/B types of errors and the systemic cost of specific AI based alogrithms for diagnosing potential. This is the path ahead but as we improve upon this path, we have the wonderful chance to identify the best for each student.
Delivery: AI continues to provide patient repetition in multiple ways in order to arrive at the satisfaction of the student. This when combined with better knowledge of the student’s abilities, competencies and priors (including diagnosis of preferred learning enablers) is an opportunity for mass customisation of learning like never before. Combined with robotics, we have a chance for each student to build their own devices, create solutions for everyday problems and apply their learning in ways that work for them. Some applications of this technology break through the cost barrier, becoming cheaper per student as the scale increases. This bodes well for the average student who can now both learn in their own ways and also be tested in ways that showcase their best efforts. The delivery of learning that supports teachers can now go deeper, do better and facilitate excellence for all in ways that were not possible even a few years ago.
The delivery challenge becomes even better when we look past only the average student. We discover that if the mass customisation is programmed not just for the average student, but for the extraordinary at either end of the curve, even the average student receives a better deal. This is where the true value of AI in education is unlocked. If designed for the visually less able, or say someone with 60% eyesight, even the average student benefits from access to better quality visuals and better hearing led content. If designed for the neuro divergent savant, the learning algorithm learns to pace itself while identifying triggers that are not constructive. In designing for the slow student, the blinkers and blind spots of each smart student are served.
Each of these remain a design challenge for quality delivery at accessible cost, but the clear goal of this must be excellence for all. The design for all of this will play out in the next few years and we may succeed and fail in turn. As a system we have but one check – is the delivery of learning enabling the student to do more or is it bypassing effort, engagement and enablement. This of course finally means that at a system and policy level, we should be testing the system and not the student.
Destination: The ultimate test of AI enabled upskilling will be where it leads the student. I have recently heard of students preparing for a post work world, where they will have to find meaning when AI has taken over all their jobs. This is certainly not a successful deployment of AI either to skilling or to the world of business. Indeed, it would count as system failure in my book. A well designed AI enabled skilling system would also have diagnosed the needs and gaps in its predictive diagnosis and delivered to that gap when creating learning experiences. The proof of that would be in meaningful destinations for its cohorts. This then becomes a circular, iterative system of aligning the demand and supply of trained talent in an AI world (can explain more later, if required).
A well designed AI enabled learning and skilling system would also be able to help in discovering the specific niches that each individual in each cohort can pursue in order to lead a purpose driven life of satisfaction and progress. This must not trap the human in choices selected by the machine but must enable them to find opportunities without the restrictions of location, prior biases or even precedent. This is one of the ways that the AI enabled world can break through to demonstrate innovative and unprecedented placements where one had not even anticipated a potential career.
Policy ahead
A successful AI policy at this stage must be designed to enable the discovery of the future (generative AI should do at least this much) and to prepare content for delivery and identify people diagnosed to have the potential to do well in such future opportunities. While this may seem to be three levels removed from current reality, we know all the components for building such discovery already exist even as we admit that we may not be able to create a smooth and accurate system for this just yet. This is why it has to be policy led, publicly funded, community governed and built for self correction and smarter alignment with human needs and aspirations.
CAVEAT
There is much to fear, but fear does not stop us from building forth. As ever, each choice we make comes with a paradox that we hope to reduce to a trade off when we implement or design. In the case of AI enabled learning this is “Discovery” vs ‘Agency”. When the machine charts your path faster than you, it may be easy to conclude that this is a better path. There will be social nudges and parental/systemic pressure to follow the path cast by the machine. But we must remember, these are shadows cast by the light of data, these do not have to be the choice. The light remains with the person, so should choice. It is only those who will break through the efficiency set up by the algorithms who will discover the new, the stupidly intrepid who will break through. Make sure there is room for them. A purely AI enabled world will certainly be more efficient than today, it may look more slick and smarter, but it is only an algorithmic generation based on the past. If we get stuck in its loop, we will continue to repeat the past, just smoother, smarter and faster. It will feel like progress for a while. The AI may even hallucinate futures that our past selves may have predicated in their blinkered responses to what they could see then…that is not the future we may want. The caveat to all AI enabled learning is certainly this – leave room for humanity to break through the loop. Mere AI is not a noose worth hanging ourselves. So design to support each individual, not control.
————————–
(1) https://www.linkedin.com/posts/meetasengupta_december-saw-me-speaking-at-and-joining-various-activity-7412803423085101057-JgOQ in brief.
Meeta W Sengupta 