{"id":25843,"date":"2023-09-12T13:55:00","date_gmt":"2023-09-12T18:55:00","guid":{"rendered":"https:\/\/trainingmag.com\/?p=25843"},"modified":"2023-10-11T10:48:05","modified_gmt":"2023-10-11T15:48:05","slug":"ais-real-role-in-ld","status":"publish","type":"post","link":"https:\/\/trainingmag.com\/ais-real-role-in-ld\/","title":{"rendered":"AI\u2019S Real Role In L&D"},"content":{"rendered":"
\n

The headline of our September 2019 cover story asked: \u201cAI or Just Sci-Fi?\u201d as organizations began dipping their toes into the world of artificial intelligence (AI) and what it might mean for the workplace and training. Fast-forward four short years later, and the explosion of AI tools such as ChatGPT is pushing Learning & Development (L&D) professionals into the deep end as they try to figure out how to successfully put the AI in training while still maintaining the human connection employees say they want. (Tempting though it was, I did not ask ChatGPT to write this article for me.)<\/p>\n

Interestingly enough, while nearly one-third of workplaces are actively using artificial intelligence, employees don\u2019t seem to be quite ready for L&D activities to be taken over by AI, according to new data from Wiley, which surveyed 3,000 professionals across various industries in North America. Nearly 6 in 10 respondents (59 percent) in the latest Wiley Workplace Intelligence report, Artificial Intelligence in Learning and Development: Five Surprising Facts You Need to Know,<\/em>\u00a0say they prefer to see an instructor\u2014in-person or virtual\u2014direct their workforce development learning, while only 7 percent prefer AI-directed learning.<\/p>\n

In addition, the vast majority (87 percent) of respondents say they want the L&D content to be developed by a subject matter expert as opposed to AI technology (12 percent). Respondents are not opposed to L&D practitioners using AI technology to improve their efforts, but they still want humans to create, plan, and direct the activities\u2014although that may change down the road.<\/p>\n

\u201cIt\u2019s essential to remember that AI is here to augment human efforts, not replace them,\u201d stresses Pooja Jaisingh, senior director of Digital Learning at Icertis. \u201cTools such as ChatGPT are designed to enhance the learning experience and streamline processes, not automate everything. There\u2019s still a big role for human judgment, particularly when dealing with topics that require deep contextual understanding, empathy, and ethics.\u201d<\/p>\n

HOW CAN AI HELP IN TRAINING?<\/h2>\n

AI has been gradually taking on a larger role in training over the last few years, including chatbot coaching and learning management system (LMS) functionality, such as Nethix\/Amazon-type recommendations for training courses based on past attendance. David Metcalf, director of the Mixed Emerging Technology Integration Lab at the Institute for Simulation and Training at the University of Central Florida, believes many types of training could benefit further from the use of:<\/p>\n

    \n
  1. ChatGPT and other text-based tools<\/li>\n
  2. Generative AI art tools such as Midjourney, Adobe Firefly,\u00a0Leonardo.ai<\/a>, Photoleap, Stable Diffusion, and DeepFloyd IF, and tools that produce animations and motion graphics such as Runway Gen 2 and WordsEye World<\/li>\n
  3. AI-enabled programming tools such as GPT 4 (with code interpreter), Amazon CodeWhisperer, and Microsoft 365 Copilot<\/li>\n<\/ol>\n

    \u201cThese tools are particularly useful for technical topics, as well as very specific topics that combine multiple domains of knowledge that previously required a substantial amount of time to create valid content with subject matter experts,\u201d he says. \u201cNow, with ChatGPT, EntrepreneurGPT, and other tools, learning objectives, outlines, and even full learning content modules can be produced in minutes to hours rather than days to weeks,\u201d then followed up with quality control and subject expert matter review.<\/p>\n

    The programming tools allow trainers to write out objectives and details of the functions of a computer programming application and then automatically write it for them. \u201cThey can even tell you where to post it for testing and final deployment, whether a Web course module, a mobile app, a simulation, or even an augmented\/ virtual reality application,\u201d Metcalf says. \u201cIt\u2019s like having a programming assistant or a teammate who never has a typo in their code.\u201d<\/p>\n

    One valuable but often overlooked aspect of AI is that it has helped to create a generation of text-to-speech voiceover technology that sounds highly realistic, adds Karl Kapp, an author, speaker, and Director for Interactive Technologies at Bloomsburg University. Kapp points to the tool 7taps, which has a functionality to allow AI to create a series of cards based on a text prompt. It provides an AI-generated \u201cactor,\u201d voiceover, and even content. Instructional designers can then edit that information to ensure accuracy and customize it to their organizational needs.<\/p>\n

    In the Vyond Go platform, Kapp says, instructional designers can type in a prompt, choose the vibe of the video (ranging from formal to informative, casual, or playful), and then pick the format (including an anecdote, announcement, diatribe, how to, and more). Then, the designer picks the layout (ranging from conversation to talking head) and chooses a setting, and Vyond creates a two- to four-minute video covering the topic. The designer can edit and add text-to-speech items to it and then publish the video.<\/p>\n

    One of the biggest applications for AI in training is providing coaching and feedback and allowing for practice through role-playing\u2014all at the learner\u2019s pace. \u201cIf you\u2019re learning a new language, for example, AI tools can generate exercises, help you practice conversations, and give you instant feedback,\u201d says Shomron Jacob, spokesperson and head of Applied ML and Platform at Iterate.<\/p>\n

    Similarly, he says that AI can create simulated customer interactions in customer service training, allowing you to practice your skills in a safe and controlled setting. \u201cThey also can analyze your performance, identify areas for improvement, and offer personalized recommendations to help you enhance your skills,\u201d Jacob says.<\/p>\n

    AI tools likewise can be an asset in the onboarding process. \u201cIt\u2019s like having a virtual assistant who guides you through the initial stages of your employment, making sure you feel confident and well-prepared,\u201d Jacob says. \u201cThroughout your learning journey, AI tools provide personalized reminders, reinforcement exercises, and micro-learning modules to help you retain information and continuously improve your skills.\u201d<\/p>\n

    Jacob also cites AI tools\u2019 ability to curate relevant learning materials. \u201cThey can analyze a vast amount of content and handpick resources that align with your specific learning goals and progress while adjusting the pace and difficulty of training based on your performance.\u201d<\/p>\n

    CAVEAT EMPTOR<\/h2>\n

    While it\u2019s easy to jump on the AI bandwagon, these tools can present some significant dangers. \u201cThis is new, uncharted territory, and these models were built to make money\u2014 not to serve you,\u201d warns Stefanie Boyer, Marketing professor and director of the Northeast Intercollegiate Sales Competition at Bryant University, and cofounder of the RNMKRS app, which offers opportunities for skill development and enhanced performance. \u201cIt\u2019s a buyer beware situation. It\u2019s best to use AI tools in walled-off applications where they can do the least amount of damage if they go haywire and start hallucinating (when AI generates false information based on the information within its language model).\u201d<\/p>\n

    A hallucination can be misstating something, creating a research article for a citation, or even writing a sentence that contradicts a previous sentence, Kapp explains. \u201cSince AI embeds the hallucination within the body of other information that is often true, it becomes critically important for the designer of the instruction to vet the content to ensure accuracy.\u201d<\/p>\n

    Jaisingh has come across occasional pronunciation issues in AI-generated audio and inaccurate responses from ChatGPT. \u201cThese hiccups highlight the importance of using an iterative approach to AI integration, consistently reviewing outputs, and not solely depending on AI for quality control,\u201d she says.<\/p>\n

    Regular review and updating of the data and models used by AI systems are crucial to maintaining accuracy, Jacob agrees. He recommends having a process in place for users to report errors and provide corrections to help ensure the integrity of the content.<\/p>\n

    Engines like Prometheus from Microsoft will greatly improve the quality of outputs, Metcalf believes, as will only using verified references from credible sources. \u201cBut for now, it is a best practice to always review every response for accuracy,\u201d he stresses.<\/p>\n

    Boyer also reminds users to assume that any proprietary content\u2014including sensitive training materials\u2014they put into a Large Language Model (LLM) such as ChatGPT is unsafe. L&D professionals must ensure that any AI tools they use comply with relevant data privacy laws and adhere to best practices.<\/p>\n

    Last but not least, Jacob cautions it\u2019s crucial to avoid over-reliance on AI. \u201cComplex topics often require the expertise and guidance of human trainers to provide effective instruction. Human trainers excel in areas such as emotional intelligence, cultural sensitivity, and nuanced decision-making, which are areas where AI may fall short.\u201d<\/p>\n

    TIPS FOR CREATING PROMPTS<\/h2>\n

    The key to successfully using AI tools is knowing how to tell the AI model what you want. This is called prompting. \u201cThe more specific you are, the better,\u201d Boyer says. \u201cRemember, the model doesn\u2019t \u2018think,\u2019 even though it seems that way. Rather, it hunts, combines data sources, and returns information. And, for the most part, models are limited to delivering accurate responses based only on data that existed at the time they were trained.\u201d Think about the parameters in your prompts, Boyer suggests, and don\u2019t assume the model can read your mind. For a text search, give the model as much specific content as possible. For example:<\/p>\n

    \u201cWrite\u00a0(this is the mission or action)<\/em>\u00a0a \u201cfunny\u201d\u00a0(tone)<\/em>\u00a0200-word\u00a0(a scope parameter)<\/em>\u00a0description\u00a0(the desired result)<\/em>\u00a0in the style of a \u201cJohn Deere sales brochure\u201d\u00a0(a guiding reference with double quotes denoting emphasis)<\/em>\u00a0of a car trailer hitch with wheels\u00a0(the subject)<\/em>\u00a0available for sale\u00a0(a parameter)<\/em>\u00a0to consumers\u00a0(another parameter)<\/em>\u00a0in 2023\u00a0(context)<\/em>\u00a0in the United States and Canada\u00a0(more context).<\/em>\u00a0Do not give me advertising copy, only sales copy\u00a0(parameter).<\/em><\/p>\n

    Coming to grips with AI language is like learning to dance\u2014it\u2019s all about rhythm and cues, Jaisingh believes. \u201cWe\u2019ve found success by treating the AI interaction as a conversation. It\u2019s a back-and-forth process\u2014ask questions, review the AI\u2019s response, then guide the AI to the desired output.\u201d<\/p>\n

    Jaisingh says crafting prompts is like preparing a roadmap for AI. Here\u2019s how she navigates the process:<\/p>\n