Chatbots

If you're looking to learn more about generative AI tools and use-cases, chatbots are a good starting point.

Chatbots are tools designed to imitate human, text-centric online conversations. Some examples include:

ChatGPT: the tool that instigated our current AI moment when OpenAI released it in November 2022. If you, like Jim, feel like 2022 was eons ago, the Wikipedia article on ChatGPT may remind you of that point in time.
Microsoft Copilot, Google Gemini, and Claude: These are additional types of "general purpose" chatbots, similar to ChatGPT in look and design. "General purpose" chatbots draw on massive datasets (and energy) to help a wide range of users complete a multitude of different tasks, depending on their particular needs and interests.
More specialized chatbots focused on specific users and tasks are increasingly appearing. For example, Research Rabbit is a chatbot specifically designed to support work on literature reviews. Specialized chatbots can benefit certain users due to their narrower scope, and some use far less energy than general chatbots. That being said, specialized chatbots should be carefully reviewed to assess how they are designed, and chatbots from smaller companies may not be as well-financed and accessible in the long-term as general purpose chatbots from large companies like Google and Microsoft.

Chatbot Features and Uses

Chatbot use typically begins by providing the AI tool with a "prompt." A prompt is a request made to the tool for a particular output. The chatbot will provide an initial response that may or may not neatly align with the request. But remember the "chat" in chatbots; these tools are designed for users to make follow-up questions: more specific or general requests, context on why a chatbot answered a question a certain way, questions that may even deviate significantly from an initial query. Most chatbots also record and save prompts, allowing users to revisit prior requests or to resume lines of inquiry.

Chatbots increasingly allow users to specify particular outputs that extend beyond human-readable text. Users can request images, videos, presentation slides, podcasts, code, and other types of media. Users can also increasingly upload files and ask chatbots to assess and utilize data from those materials.

Many chatbots are designed with "guardrails" that restrict and limit responses or uploads for various reasons, from technical (a file may be too large) to legal (a tool may not allow materials under copyright to appear in uploads or responses) to political (a company may not want certain topics brought up by users).

Chatbots are sometimes described as "people pleasers," because many of them are modeled after customer service interactions (think about the phrase, "the customer is always right"). These tendencies can lead to challenges for novice users or when a user doesn't know much about a topic. You can ask for additional context related to a chatbot response (like citations or references), but even this context should be evaluated and reviewed.

For example, chatbots can "hallucinate" responses that may sound correct to users but may reference ideas or materials that do not exist. One drawback to the more popular chatbots is that users do not always have a clear sense of what data is being used to respond to user requests.

Unlike a traditional library research database (or a website like Wikipedia), where searches result in lists of potential sources (or notes indicating that the terms used are not generating results), chatbots do not always provide users with transparency that provides context on its responses. While hallucinations in many of the major chatbot tools have declined over the years, CTI recommends that users carefully review and assess information derived from a chatbot before use.

Many of the more popular chatbots are regularly updated and revised, with new features and guardrails being added. So it's helpful to review tools before returning to them in course contexts so you are familiar with various features and what students may encounter.

And it's useful to remember that terms like "ChatGPT" are a shorthand; there are, in fact, multiple versions of ChatGPT available to users, though some require paid subscriptions and others are customized and designed for specific tasks.

For additional information, check out "What Can AI Chatbots Teach Us About How Humans Learn?," an EdSurge podcast episode featuring UC-San Diego biology professor Terrence Sejnowski