Can artificial intelligence teach a computer to see? Tointuit a learner’s mood? Not quite—but emerging AI technologies are enablingcomputers and automated processes to use images and language in eerilyhuman-like ways. Four evolving technologies—computer vision, image recognition,sentiment analysis, and emotion recognition—could transform eLearning. Some ofwhat L&D professionals could do with these tools is already evident in theways various industries are using AI to automate processes, detectpatterns, crunch enormous amounts of data, and improve efficiency. This articleoffers an overview of the technologies and how they could impact eLearning.
Computer vision
Computer vision refers to enabling computers to identify andprocess images scanned from the environment. It allows robots to “look around”and identify obstacles and items in the vicinity so they can safely movethrough an environment. It’s a vital part of the technology powering autonomousvehicles. The technology also enables an algorithm to process a 2-D (flat)image, understand what the image depicts—and create a 3-D image.
Computer vision is in play any time a computer makes senseof visual input. That means that bar code scanners and facial recognition appsuse it to interpret and categorize scanned images, whether from a JPEG file ora camera recording images from the environment.
In the L&D paradigm, computer vision is already havingan impact:
- Tools that tracklearners’ eye gaze offer real-time feedback on what learners do (anddo not) pay attention to. L&D teams can use this data to improve eLearning.
- Computer vision is a foundation of technologiesthat recognize learners or employees and provide individualized access toinformation or training.
- Safety tools use computer vision to guidevehicles and machinery and to scan codes to trackmovement of equipment or goods through a production facility orcount items and monitor volume.
Image recognition
Apps based on image or facial recognition are cropping upeverywhere—including in eLearning and performance support. These technologiesuse computer vision to scan an object, person, or image. They then use trainedalgorithms and machinelearning skills to identify what’s in the image. The uses for thistechnology are myriad, ranging from categorizing objects to using facialrecognition to grant or deny access to locations or information. Imagerecognition is also a key element of OCR, optical character recognition.
Performance support tools that pair computer vision withimage recognition to identify parts might assist employees in ordering orrepairing them; an app could match up a scanned image from the environment witha catalog, instructions for repair, or explanatory text and diagrams. Socialplatforms, including Pinterestand Houzz,use this technology to help people find products they’ve seen and liked. It’seasy to imagine a virtual assistant that automates these processes in a factorysetting.
A search and sorting function, which relies on image andtext recognition, is essential to effective curationof online content or organization of a library of training and performancesupport tools. Together, these technologies can automate analysis, categorization,and delivery of personalizedor targetedinformation to learners as they need it.
Sentiment analysis and emotion recognition
Sentiment analysis of text is a special application of naturallanguage processing and classification or categorization of items. Thealgorithm looks at specific text and calls on its training with data thatinclude terms and phrases categorized as positive, negative, or neutral. Amachine learning element can enable a sentiment analysis app to become moreaccurate over time, providedit receives new data and feedback on its decisions. Emotionrecognition algorithms perform similar tasks using images, rather than text, asinputs. The images might be static photos or they could be scans of people’sfaces taken in real-time.
Sentiment and emotion detection and analysis technologies arealready being used to monitor customer satisfaction in marketing and salescampaigns. But they have L&D applications as well: A sentiment andemotional analysis tool could scan learner feedback on eLearning andface-to-face training, quickly helping L&D teams identify problems andsuccesses. It can also be used during training to provide insight into the moodand motivation level of learners while they are doing eLearning.
Correlating the sentiment and emotion recognition data withtraining events and with performance data could provide valuable information toinstructional designers and developers on the effectiveness of their trainingand on how the learners are responding to it. And matching up positive andnegative reactions with demographics like employee age or job role could aidL&D in improving engagement by targeting training in preferred formats oron more relevant topics to the right employees.
Caveats
AI-based technologies are still evolving, and that meansthey have their share of potential pitfalls. A big one is the way algorithmsare trained and deployed. If the data used to train an algorithm isn’tsufficiently varied and comprehensive, the accuracy of the algorithm willsuffer. Even after initial training, the algorithms depend on a constant streamof new data to work with—and learn from—to keep improving; algorithms used insome apps do not have that machine learning element to provide continuousimprovement.
Technologies that use facial recognition or are guessing atlearners’ moods or motivations are also likely to bump up against serious privacyconcerns. Learners might object to having their eye movements andfacial expressions tracked and analyzed; eye tracking, motion tracking, andcomputer analysis of images and words could feel intrusive to many learners. Developersin some countries must adhere to strictlaws governing the collection and use of personal data.
As L&D teams learn about and adopt new technologies,including these emerging AI technologies that could transform eLearning, theyneed to remain mindful of these issues and mitigatepotential harms wherever possible.
