Common AI Techniques in L&D

Pros

Cons

• Can provide quick answers to commonly asked questions
• Simple to use (i.e. simple user interface that people are already familiar with) and always available
• Can pull information from a variety of sources (e.g. Dropbox, Google Drive, intranet, etc.) and deliver that to learners within the chatbot message box
• Ability to provide personalized/adaptive feedback to learners
• Low bandwidth required for text-based chatbots (or no bandwidth if chatbot messages are delivered via SMS)
• Potential to use voice interface for voicebots

• Natural Language Processing (NLP) still has some gaps
• Many bots on the market but quality varies
• Bias in content selection or incomplete content selection
• Context is still missing, thus difficult to have deeper or more meaningful conversations
• Potentially causes mistrust of the system and the learning content when chatbots push out wrong information/not able to help (due to lack of context or gaps in content selection)
• There is a lack of research in evaluating the effectiveness of chatbots used in learning especially in terms of supporting self-regulation and self-efficacy in learners

Pros

Cons

• Good for content curation
• Recommender system – source a variety of content based on learner profiles, knowledge levels, and preferences
• Can adapt content and activities to the right level of knowledge/skill
• If it is well-designed, it can create a learner-centred experience that can be sensitive to the learner’s needs as well as context
• Potential application of sound pedagogical principles to structure learning that will meet the learning objectives and competencies
• Can possibly provide personalized experience in large online courses (i.e. MOOC)

• Could be too prescriptive
• These platforms are designed by tech people – they decide on which data is relevant in order to categorize learners in certain ways, which are often indicators that don’t necessarily help you arrive at a conclusion or to provide learning interventions
• If the adaptive pathways are being pre-defined by someone (either tech people or instructional designers or both) – what assumptions and bias do they have? There aren’t a lot of good practices and guidance on this front
• Context is still missing, thus difficult to have deeper or more meaningful conversations
• You don’t know what you don’t know – how do you know what you missed out on if you go down a particular adaptive pathway
• Difficult to evaluate as most algorithms are opaque and not open for review (or too complex for educators to review)

Pros

Cons

• Save time and effort on routine-based tasks
• Auto-tagging, indexing, and metadata
• Potential to scale content
• Localization and globalization functions

• AI technologies such as Deep Fakes have a bad reputation and promote a lack of trust
• There is a lack of quality control/standards in place
• If the adaptive pathways are being pre-defined by someone (either tech people or instructional designers or both) – what assumptions and bias do they have? There aren’t a lot of good practices and guidance on this front
• Tweaking and preparing the information upfront still requires fairly labour-intensive work

Pros

Cons

• Can be used to understand learners, the learning experience, and learning programs overall
• Ability to turn data into insights and actions for instructional designers and instructors to use to improve learning
• Can be used to augment human decision-making
• Potential use of formative assessment data

• Too much data can be a bad thing especially when data collected are not meaningful or actionable
• Surveillance and control issue – who own the data? Can learners choose to opt out?
• Mistrust of the system when predictions are inaccurate
• Inaccurate predictions could potentially demotivate learners/impact their confidence
  

Pros

Cons

• Save time and reduce cost
• Can potentially complement human instructors in assessing quality of content
• Can scale up for large learner groups
• Technologies have improved significantly since the early days

• The whole is greater than the sum of its parts – easy to miss the whole picture by focusing on the specifics
• Pattern matching could be misleading
• Still missing nuances/context
• Can easily trick the system and there are many skepticism around this*

Pros

Cons

• Makes simulations more realistic (i.e. More likely to enable transfer of learning from simulated environment to actual environment)
• Enables experiential learning
• Able to support creative, non-linear learning*

• Hardware is still a limitation
• Expensive to produce, update, and maintain
• Slow or inconsistent framerate impacts performance
• VR interface makes some tasks more difficult to do
• User privacy and data controls on data tracking features have not been addressed adequately