How artificial intelligence might transform learning and development

I attended an interesting webinar this week with Donald H Taylor from Learning Technologies Exhibitions & Conferences and Ulrik Juul Christensen from Area9 Lyceum , titled Beyond generative AI – how Artificial Intelligence will transform L&D.

Views on AI's impact on workplace L&D

The webinar began with an invitation to participants to share their views on AI's impact on workplace L&D.

Some people used words such as 'exciting', 'intriguing' and 'essential'. Commonly used words and phrases included:

• potential, opportunity, transformative, change, game-changer, revolutionary, disruptor, powerful, huge impact
• content creation, multimedia content
• time saving, automate, efficiency, outsourcing
• personalisation
• more agile, support for developers, concentrate on strategic thinking, increase value

Other people used words such as 'unknown', 'dangerous', 'cautious' and 'scary'. Concerns and potential pitfalls included:

• replacement / loss of jobs, unemployment
• lower quality content, bland content, loss of uniqueness, formulaic
• dehumanisation of experience
• costly, expensive, gimmicks
• skill gap, need for training / upskilling, coping with the pace of change, depends on quality of prompts
• bias, incorrect content, outdated information
• trust issues, assessment
• data security, data protection
• copyright
• ethical implications, issues with regulation
• accessibility, equity

People generally felt that human interactions will still be important and AI will not replace human ingenuity. "AI will focus on the what; humans on the why." Some people made interesting points about having had awful experiences with humans as well, and that we hold AI to higher standards and have less empathy towards error than we do with our human counterparts.

What is generative AI?

I have shared previously some of my learning from Google Cloud courses which give an introduction to how generative AI works:

• Introduction to Generative AI
• Introduction to Large Language Models

I found ChatGPT to be a helpful resource to clarify my understanding of some of the concepts, as I could ask it questions and ask it to explain an idea more simply or to give examples. But I did check this information by reading content from blogs and from other learning resources.

Large language models are statistical models - they identify the word that has the highest probability of fitting next, based on the huge amounts of data on which they have been trained. This means that they can also generate content that is inaccurate or misleading ('hallucinations'). They can reflect biases and stereotypes present in the training data.

Large language models are language processing tools, not search engines. They rely on their training data and cannot access the internet, which means that they operate from outdated data. For example, ChatGPT's most recent training data is from September 2021.

It was interesting to read about the AI principles of major generative AI providers such as Google and Microsoft . Google Cloud's course recommended that every organisation should work to define its own similar principles, outlining how they will (and will not) use AI.

Adaptive learning

Donald Taylor compared traditional e-learning to a guided tour of a city, in which everyone follows the same, sub-optimal route, visiting some new places of interest, but also some that they have seen many times before or that they are not interested in.

AI enables non-linear adaptive learning, where delivery can be automatically adjusted to the needs of each learner, recreating the teaching approach of a one-on-one personal tutor. Adpative learning enables inclusion, allowing everyone to take as long as they need to reach proficiency. It can understand what people are doing and react in real-time.

Personalised learning is already common in education platforms designed for use in school or for home-learning. Two examples that I used as a teacher were Sumdog and IXL . They use the data from children's responses to each question to provide further explanation where needed and to adapt the next questions presented, providing easier questions if the child is struggling or harder questions if the child is giving correct responses.

Another analogy shared was that traditional e-learning is like having a printed list of directions, whereas adaptive learning is more like Google Maps, where you are able to customise the route and course correct.

Sal Khan 's TED Talk demonstrates how AI can be used to act as a personal AI tutor for every student and an AI teaching assistant for every teacher.

"We can use Khanmigo to give every student a guidance counselor, academic coach, career coach, life coach..."

The role of knowledge acquisition

Ulrik Juul Christensen predicts that we will need to learn more subject matter in the future, not less.

He talked in the webinar about the continued value of conversation and social learning, but that adaptive learning gets the knowledge acquisition out of the way so that you are able to use it in different ways.

Humans are good at making associations, but we first need to master the content, before we are able to make connections.

"You can't join the dots if you don't have the dots."

Area9 Lyceum build on the model of four-dimensional education developed by The Center for Curriculum Redesign . This includes knowledge, skills, character and meta-learning.

Ulrik Juul Christensen also talked about the need to move away from the assumption that attending or completing a course means that learning has taken place.

Designing courses

When thinking about a new book or course, the first step has always been to get the lay of the land and to identify what is already out there. This is made much easier by ChatGPT, which can survey all of the information on which it has been trained.

Ulrik Juul Christensen shared MindFlow , a tool that Area9 Lyceum are currently working on. This tool has an iterative conversation with GPT-4 based on the inputs about the course objective and audience. It then generates a course outline, assessment questions and course goals, and can generate text content for the course, narrated by an AI text-to-speech tool.

He stressed that everything generated by the tool will need to be reviewed and tweaked by humans, but that they have found so far that they only need to change about 2% of what is generated.

The tool does the trivial work, freeing up learning engineers to do the higher order thinking and to build more sophisticated components to add to this.

Ulrik Juul Christensen discusses this further in a recent article for Forbes:

"The next generation of AI opens up more potential for humans—not against them. In particular, AI has the potential to revolutionize how humans learn through greater personalization."

Polly Watt shared in the chat her layered process for using AI to develop courses, inspired by a course by Dr Philippa Hardman :

• Discovery - understanding demographics, psychographics, and learner motivation.
• Creating and comparing course outlines on multiple AI (e.g. GPT-4 and Claude Anthropic ).
• Creating learning objectives - ensuring that they are action-oriented, benefit-oriented, concrete, fit within the learners' zone of proximal development, and are engaging.
• Creating content, including multimedia content created using GPT-4 (for example, charts generated using the Data Interpreter plugin).
• Generating assessments - including quizzes, answers, and feedback.
• Writing case studies to add narrative.

She stressed the importance of refining prompts at each step, asking for feedback and checking for coherence, as well as carrying out proper research to check the accuracy of information generated by AI.

Ross Stevenson recently shared his report titled The State of AI in L&D 2023 , based on conversations with over 130 L&D professionals. He discusses the need for individuals to learn about AI in order to make the best use of it and shares some useful resources for exploring further.

"L&D pros understand that AI has the potential to significantly impact their initiatives and, consequently, the future success of their organisations. We now need to turn this into strategic value and leverage in performance."

Learner engagement

There is a risk that AI will be used to generate content that is not engaging.

Ulrik Juul Christensen talked about how providing opportunities to demonstrate proficiency can be one way to engage learners.

Donald Taylor's post about 'learner engagement' interprets the term at three levels: asset, cultural and environmental. He recommends spending time unpicking the individual or business need before creating content, working on the culture of learning within an organisation, and working with leaders to make space for learning within the flow of work. In a comment on the post, Nick Shackleton-Jones talked about people's interests being shaped by challenges that require us to seek out information. It is important that learning, however it is developed, is focused on what people want and need.

How we learn

A comment about personalisation led to the usual discussion in the chat about the myth of learning styles being debunked. But some helpful resources were shared about learning from cognitive science.

Another debunked myth shared was that attention spans are becoming shorter. The evidence for this is vague, and many psychologists say the idea of an "average attention span" is pretty meaningless.

Copyright

You cannot copyright content produced with ChatGPT.

Ulrik Juul Christensen talked about the need to rethink the value of intellectual property.

Beyond ChatGPT

Other AI tools shared in the chat included:

• Otter and Fathom - for live captioning, transcription and summarising
• Microsoft Copilot - integrated into Microsoft 365 to improve productivity
• 7 Taps - for microlearning content creation
• Pi, your personal AI - simulating emotional intelligence

Assessment

A concern discussed in the context of education was assessment, for example generative AI being used by students to write essays. One tool shared for detecting the use of AI was GPTZero .

Others discussed the need to develop better methods of assessment that promote critical thinking rather than information retrieval. In a video about four-dimensional learning, Ulrik Juul Christensen and Donald Taylor discuss how the ability to perform well in a timed test is not a good measure of future performance. He shares the work of Tony Wagner , who is exploring the changes needed in schools to prepare children for the 'innovation era'.

What does this mean for employees?

When asked 'What L&D jobs are most at threat because of the rise of AI?', ChatGPT responded , "While it is not accurate to say that AI will directly "threaten" jobs, it is more appropriate to say it could transform or change the nature of these roles... Professionals in these roles can remain relevant by continually updating their skills, adapting to new technologies, and focusing on the areas where human input is still critical."

Ulrik Juul Christensen talked about the opportunity for L&D employees to shift from content creation to helping people learn.

In this article , Benedict Evans puts AI into the context of other waves of automation over the last 200 years.

"Every time we go through a wave of automation, whole classes of jobs go away, but new classes of jobs get created. There is frictional pain and dislocation in that process, and sometimes the new jobs go to different people in different places, but over time the total number of jobs doesn’t go down, and we have all become more prosperous."

In 2018, The Institute for the Future predicted that 85% of jobs that will exist in 2030 hadn't yet been invented. Nicolas Hazard , in his book, The Future of Work, explores some of the new jobs that may be created by the advances in robotics and AI through the experiences of 21 pioneers who are at the forefront of their fields.

The International Labour Organization (ILO) have assessed the potential effects of generative AI on job quantity and quality.

"The most important impact of the technology is likely to be of augmenting work – automating some tasks within an occupation while leaving time for other duties – as opposed to fully automating occupations."

In his 2016 book, Klaus Schwab introduced the term 'Fourth Industrial Revolution ', or Industry 4.0, to describe the rapid technological advancement in the 21st century. He looks at how we can take collective responsibility to shape this emerging future positively.

One person commented in the webinar chat that the end goal for humanity might be to have automation replace most jobs so that humans get to do the things they enjoy, so the fears about job loss may be overblown.

Donald Taylor closed the webinar by sharing his recollections of watching 'Now The Chips are Down ', a 1978 Horizon documentary about the applications and implications of microprocessors to employment, which raised very similar concerns about the potential loss of jobs. "There is going to be tremendous social upheaval." The Third Industrial Revolution (or Digital Revolution) changed the world, but did not result in the scale of problems that some feared.

"It is time to think about the future."

Webinar replay

The replay is available here: LSG Webinars ( learningtechnologies.co.uk ). You can find out more about future webinars here: Digital Hub ( learningtechnologies.co.uk ).