Generative AI chatbots are good at many things and can be useful. It's important to be aware of ethical, equity, and privacy issues related to chatbots, and be informed about how chatbot companies handle personal data and chatbot conversation text.
The resources below may help with navigating this new and complex way of finding and understanding information.
The main popular, standalone chatbots are free and, ChatGPT has a paid premium tier (ChatGPT Plus). If these tools become increasingly fee-based, or restrict premium features behind paywalls, there is potential for inequity of access. Most of the more specific AI-powered research tools have a freemium model, and some are exclusively for paying subscribers .
Chatbots base their responses on patterns in their training data and may unintentionally respond in ways that mimic portions of that data. The precise content of Large Language Model training data is not transparent, but it is derived from past open internet sources and includes incorrect and biased information.
There’s a concentration of power in a few companies that can afford to develop and train Large Language Models. LLM training also has a large carbon footprint, and relies on human data annotators and feedback.
Currently, the popular, standalone chatbots require you to create an account and log in to use them. The companies that make them store and have access to your chatbot conversations. The conversations are used to improve the product and could be used to train future models.
Never share confidential, sensitive, or proprietary information in a chatbot conversation.
It is possible to customize some account, data, and privacy settings in many chatbots and tools.
Anthropic's Claude does not have such settings because of their safety-first approach. Anthropic was founded by former OpenAI employees who disagreed with OpenAI's departure from its initial non-profit plans. Read more about Anthropic's core views on AI safety.
Director of AI Ethics Lab Cansu Canca explains how AI takes in information and uses it to make informed predictions, but when opaque algorithms make decisions on your behalf, ethical problems can arise.
Duration: 1:28 minutes.