Abstract:

We present a series of AI in learning projects, explaining both success and failures. The point is to learn from these to improve the success of future projects. These include projects on building AI-based learning platforms, performance support systems, student support product, a total of 12 projects. We also discuss the importance of top down support, also presented on the presenter's latest book 'AI and Productivity', along with various techniques for overcoming different types of behavioural, organisational, technical and economic paradoxes. We present these paradoxes, along with solutions for their resolution.

One early realisation was that AI adoption, arguably the fastest in the history of technological diffusion, differed from previous waves of innovation in one important respect: it was largely bottom-up. The script had, in effect, been flipped. Learners found AI useful before many organisations were ready to welcome it. Senior management in universities, schools, the public sector and corporations often appeared cautious, sceptical or openly resistant. This produced a bottom-up, often hidden, pattern of use: students, faculty and even administrators were using AI widely, but often discreetly. That in itself seemed paradoxical. There are other pradoxes.

Behavioural paradoxes:
The first set are behavioural. Human beings are creatures of habit and are subject to persistent cognitive biases. Confirmation bias narrows behaviour and learning, encouraging us to repeat what already works while overlooking better methods. Status quo bias makes the costs of switching appear greater than they really are. Negativity bias leads us to overestimate risks. Added to this is anthropomorphic bias, whereby AI is misread as if it were a person, sometimes trusted too much, sometimes feared excessively.

Technological paradoxes:
There are also technological paradoxes. The Solow paradox famously captured the problem that productivity gains from new technologies often do not show up quickly in the statistics. A more pressing issue is legacy-systems lock-in. Older IT architectures often inhibit the adoption of more productive technologies, including AI.

Economic paradoxes:
A further set of paradoxes has long been recognised in economics. Jevons paradox suggests that efficiency gains often increase usage rather than reduce it; saved time becomes more demand, more scope and more work, rather than fewer hours. The Easterlin paradox reminds us that higher output does not necessarily produce greater wellbeing. Finally, there is what might be called the Pollyanna paradox, where techno-utopian optimism breeds complacency.

We will also present some of the challenges were technical, in particular issues around data it access, quality and ethical obstacles, our AI technical expert, who built these projects will explain the problems that need to be addressed during AI for learning projects. We recommend a particular approach to the management of such data for successful ingestion. This includes the handling of subject matter experts and methods for eliminating gaps and contradictions in content. We used AI techniques to transcribe interviews, as well as do analysis on workflows in healthcare projects for nurses, which showed how this can be optimised.

This is a real world presentation, with real examples of real projects delivered to real learners, covering design, management, build and delivery.

Keywords:

AI, date, projects, paradoxes success, failure.