You may have heard of ChatGPT, the powerful language model developed by OpenAI, that is changing the way we interact with technology, providing accurate and human-like responses to a wide range of questions and tasks. The model is trained on a massive dataset of text, allowing it to learn the patterns and structure of human language. This enables ChatGPT to understand the context and intent behind a given text and generate a response that is not only grammatically correct, but also semantically meaningful. With its cutting-edge capabilities, ChatGPT is revolutionizing the field of natural language processing. It can be possible to develop a chatbot for mental health and help individuals with depression.

ChatGPT is the smartest chatbot in the world so far. Chatbots that were not trained on the extensive training datasets fail to mimic the performance that ChatGPT has. First, to understand why ChatGPT outperforms any other models in the world, it is important to note that its 175 billion parameters, which are trained on a massive dataset of text, is a key factor. In any field of machine learning, the size of the dataset is crucial, as the larger it is, the better the model will be. ChatGPT is trained on a huge amount of text data which allows the model to learn from divergent inputs and have more accurate responses for more given inputs. Additionally, it is important to note that training your model with a massive dataset of text requires extensive computing power. According to Arxiv, “Computational power and computing architectures shape the speed of training and inference in machine learning, and therefore influence the rate of progress in the technology.” As computers keep developing to display their astounding processing power, this enables us to train a model with a more extensive dataset. When we have our hands on larger dataset, it is possible to completely mimic a highly accurate human conversation with real people. This means we can potentially build a model that could simulate human conversation to such a degree that it could serve as someone’s companion.

As we know machine learning models require datasets, and the dataset comes from the internet. Every day we use the internet, this could be browsing through social media commenting on posts, texting our friends, or using a search engine to study. We can gather datasets from the internet quite easily through what is called “Web Scrapping.” Web Scrapping is the process of visiting and gathering information from websites. It is used by search engines, price comparison websites, and other organizations to gather large amounts of data and index it for searching or other purposes. This is a great way to get datasets from the internet, we use this dataset for training a machine learning model. According to Analytics Vidhya, “Web Scraping deals with collecting web data and information in an automated manner. Basically, it is the extraction of web data. Web Scraping deals with information retrieval, newsgathering, web monitoring, competitive marketing and more.” One thing to mention is that there are many open-source websites on the internet that let you use their datasets for free. The most prominent one is CommonCrawl.org, which serves petabytes of data for free. We can also gather datasets from Reddit or other social media platforms. To help with depression, we would want a chatbot that is friendly and give advice to others. This implies that people can train their model with datasets that are related to providing advice to others. However, we are not limited to training our model only for advice. People can train the model on casual conversations between individuals, and this dataset would be garnered from social media such as Instagram and Facebook.

It is important to note that specific datasets will be ideal for a chatbot that can replicate friendly human conversations. According to Freshman Elyse Yung, “I have never been formally diagnosed with depression, but I probably have mild depression because of school stress and lack of social life.” Depression is caused by a variety of reasons. The most prominent causes for students are school stress and isolation. Our dataset can include advice on how to get past stressful situations and contain a chatlog of two people having a friendly conversation. The model will perform how we want, it will give advice to a person when they need it, such as telling them “Take care of yourself” or “Hard times do not last forever.” According to NHS, “When these stressful events occur, your risk of becoming depressed is increased if you stop seeing your friends and family and try to deal with your problems on your own.” In other cases, someone may just need a friend and for this case, our model will casually talk to the person just like a real friend would. This is all possible because of the datasets we trained on our model. According to NHS, “You may be more vulnerable to depression if you have certain personality traits, such as low self-esteem or being overly self-critical.” An individual may be afraid of judgment from others when seeking out help, such as expressing your feelings with someone or disclosing your personal information with someone. According to Senior Zachary Bi, “As long as my information is not shared with anyone else, I would be comfortable talking to an AI about personal topics.” Individuals may have an easier time talking to their AI companion because they would not be getting judged for being vulnerable during hard times or not being tough. It can be hard for someone to express how they feel due to rejection and judgment. According to Junior Bryson Hu, “AI technology can be used to make treatment more personalized and help with symptom management, which can improve the overall quality of life for people with depression.” Advanced chatbots can learn to adapt to the user’s needs and change its persona which makes it versatile. A good example of this is BlenderBot 3.0 which is a chatbot that can learn and memorize the user’s linguistic style and change its persona. According to Sophomore Jennie Long, “AI can be used to help people feel better by giving them personalized help.” As mentioned before, BlenderBot3 has already done a great job of being versatile and changing its persona to personalize the user’s needs. According to Meta, “The BlenderBot series has made progress in combining conversational skills — like personality, empathy, and knowledge — incorporating long-term memory, and searching the internet to carry out meaningful conversations.” It is noteworthy that the BlenderBot is also trained on 175 billion parameters. We can appreciate that what BlenderBot is capable of is possible due to its 175 billion datasets.

As technology continues to evolve and advance, it is allowing for the development of stronger and more advanced forms of artificial intelligence. These advancements have the potential to greatly benefit humanity, particularly in areas such as mental health. For example, using AI to assist in the treatment of depression, which is a leading cause of disability in the United States. This also brings up ethical concerns on the table, however, it is up to us how we manage to leverage these tools upon us.