The concept of sentient machines seems to frighten many. After all, emotions seem to be the true distinction between man and machine. However, if machines could understand human emotions, they may not end up “taking over the world” as many sci-fi fans may believe. In fact, sentient machines have already been implemented in various aspects of society. For example, speech processing software aims to identify the true, implicit meanings behind a given phrase to determine the best response. Similarly, the use of sentiment analysis is being observed when tackling societal issues like hate speech and mental illnesses.
Sentiment analysis is a machine learning method utilizing concepts of linguistics in combination with computer science to train software to identify a specific range of emotions in a given phrase. Importantly, this falls under the branch of natural language processing which aims to bridge the gap between human communication and machine understanding. Like all machine learning models, sentiment analysis is rooted in mathematics, considering its base structure in neural networks. The basic concept of neural networks somewhat resembles how our brain functions, of course to a more simplified extent. Within neural networks, there are multiple layers of nodes that are responsible for specific jobs within the model. For instance, one layer may analyze the syntactic structure, while another focuses on the pragmatics of the phrase, and another identifies its semantics. These layers work together to identify the emotion conveyed by the message, and if wrong, the model will self-correct using an algorithm called back-propagation and an optimization algorithm known as gradient descent to minimize future errors. These algorithms work in conjunction to teach a model to analyze the explicit and implicit meanings of phrases to identify an associated emotion.
Such “sentient machines” could help identify those developing mental illnesses like depression and anxiety based on their public posts. These algorithms could also minimize the errors of current hate speech detection models. Many models may suffer due to implicit meanings differing from explicit meanings due to sarcasm or special characters. With a trained sentiment analysis model, these biases could be mitigated, thus enhancing the performance of the model.
Despite the fear surrounding sentient AI machines, such algorithms would allow for a more inclusive society with implementations into social media and other industries like speech processing.
