Artificial intelligence and machine learning for L&D
Artificial intelligence (AI) is now a widely used term for an extremely broad set of technologies. AI, at its core, is about designing software to help make decisions. Human intelligence is more than just the ability to make decisions that are based on past experience. Creative thinking is about generating new insights.
In the early stages of AI the focus was on building decision trees using if, then, when statements. The programming of the types of decision tree is extremely complex and highly likely to fail. The field of AI has moved towards an approach called machine learning, where, instead of all the rules and options being programmed, the system is ‘trained’ on data, and the actual machine learning system builds rules around this data. The machine learning system is finding patterns in the data it’s been given.
Image recognition systems are a good way to explain how this works. A machine system might be given a series of images of a cup. It begins to build a model of what a cup looks like – a cylinder with a half-circle at the side – but of course there are variations on this. Then, when the machine sees another image it can check to see if it’s a pattern that it’s been trained on, and the machine learner system determines how likely it is that the image is a cup. This last stage is called a prediction.
AI people often talk about being able to replace routine and predictable work. Robotics can replace the physical actions that humans have to take, but all the challenges of robotics are now in how to make it smarter by using AI.
In a workplace this means AI can be used with tasks that involve finding patterns. There is a lot of hype about AI being able to automate work and remove jobs. In reality it’s more likely that the parts of the jobs that are more routine and predictable will be automated and humans will still do the more complex parts of the role, working alongside automated systems. This has two effects on L&D:
1) The nature of the capabilities L&D are developing will change
If the work that humans do is complex and focused on skills that are harder to automate, for example interacting with people, creative and design thinking, then L&D needs to be more focused on developing skills in:
- interpersonal communication
- self-directed learning and learning how to learn
- strategic decision making
- design thinking and creativity.
The exact skills your people will need will depend on your industry.
2) There will be changes to the way L&D can work
Many people have tried to predict the effect of AI on the workplace and jobs. In the Digital disruption – Short fuse, big bang report by Deloitte, education is predicted to be a growth industry in the future because the work of educators is highly complex and often involves complex human interactions. In most workplace learning the focus is on the complete opposite, e.g. content and removing human interactions.
In all industries there are opportunities for automation, especially in administration tasks; many admin tasks in L&D could be automated without AI. The key to automation is the integration of systems and workflows. Some L&D teams don’t even have sophisticated systems for booking and managing face-to-face programs.
Augmentation of L&D work
There are many opportunities for the augmentation of L&D work, for instance augmenting the design of learning strategies.
Imagine a tool that helps an L&D consultant plan learning strategies. If the tool had data about past programs and how successful they were for different learners for a range of learning problems then it could guide which solutions might be best in each context. Then if the project had a need for external partners the system could make suggestions about who might be the right fit and even automate the process of contacting partners.
Automation of the actual learning experience
We are already beginning to see AI being used to automate learning experiences. Some examples are:
- chatbots
- recommendation engines
- predictive machine learning
- intelligent adaptive learning systems.
Chatbots
The use of chatbots is exploding. If you haven’t experienced one already, it’s where you communicate with a ‘chat robot’ powered by an AI system instead of a real person. Chatbots are powerful tools for performance support, where employees can ask the bot questions and it returns possible answers.Chatbots could also be used for building semi-automated coaching systems, and are particularly common in help desks.
Recommendation engines and learning
Most people have experience with recommendation engines but might not have realised that they are being driven by machine learning – for example, suggestions by Amazon and Netflix. These systems are looking at your past actions – what you have watched or bought before – then comparing that with other groups of people. They’re looking for patterns, predicting how likely you are to be part of a group, and then making recommendations based on your being part of that group. Recommendation features are starting to be seen in learning management systems too. They are based on data from job roles and an employee’s past completions. What has not yet begun to happen is recommendations being based on employees’ activity in learning experiences or actual on-the-job performance.
Predictive machine learning
As businesses become more data driven, learning needs to keep pace. Many L&D people are becoming more data driven but the biggest challenge is that they are often only measuring the amount of learning, e.g. time in a course or activity inside of learning experiences, not the impact of learning and whether or not people are actually changing. Machine learning can be used to sort and find patterns in relationships that a human might not normally see, but this is still only providing information on how people are learning, not whether they are actually changing.
The next step with data driven learning that many L&D people have begun to take is bringing in performance and recruitment data. Recruitment is one area of HR that machine learning is rapidly being applied to. It's possible to survey applicants and existing employees to build a profile and then bring in simple KPIs on, for instance, how long the employee has been employed or their on-the-job performance. The machine learning system can then build models based on the data. This means that during the application process the same questions that were used to build the profiles could be asked again, then the machine learning system can predict if the person fits the profile of being ‘a successful employee’.
xAPI (experience API) is one on the key technologies to have embedded this new approach. It means that data beyond what can be captured in an LMS can be added. xAPI data doesn’t have to be just learning data – it can be performance data as well, e.g sales or billable hours.
Higher education has been the fastest learning sector to adapt to machine learning. Many of the LMSs that are designed for use in high education include machine learning systems that can predict if a learner is struggling and not going to complete a course.
Intelligent adaptive learning systems
Recommendation engines are the first step in the personalisation of learning. Suggestions are normally based on course completions. The next stage of personalisation is adaptive learning, where the actual learning experience changes depending on an employee’s past interaction. The idea of adaptive learning is not new, it's a perfect example of what ‘thinking in rules’ means in learning. The challenge for this is not the technology. The challenge is being able to produce all the content for the different possibilities. If you have five focuses, then the content development task is similar to developing five courses. And if you are doing this for different levels – e.g. for people who are starting and then for people who are struggling – then content development becomes even more complex.
Technologies for working with machine learning in L&D
The technologies and platforms for working with machine learning are developing rapidly. Many LMSs are adding recommendation engines, and there is a wide number of existing machine learning systems already in use by organisations like Amazon and Google, which need software engineers to develop integrations. The process is getting the data into the right form, choosing the right algorithm for looking at the data, and then understanding the results. One of the more user friendly systems is BigML.
Questions and ideas to spark new thinking about AI
- What L&D process could be automated?
- How could you be using chatbots?
- Could you be using recommendation and adaptive learning systems?
- How could you be become more data driven?
- What other data sources should be part of your L&D data ecosystem?
