Text based AI will answer whatever question you will give it (provided that it is ethical) based on the information that it has been trained on from its database. Whilst AI can be a useful tool for studying at university you will need to apply your own critical thinking skills to any information provided by AI just as you would do when finding information for your assignments.
For simple concepts AI is mostly correct, however with university level concepts it can be confidently incorrect. ChatGPT for example states that "ChatGPT may produce inaccurate information about people, places, or facts"
It is important therefore that you check information gained through AI with academic sources, and that you do not rely on information gained through AI in your assessments.
A good way to test an AI's knowledge is to ask questions based on a topic that you know well with some complexity to it. You may notice that whilst it is mostly correct there are parts which may be incorrect, or missing key information. Here are three text based AI that you can use for this activity ChatGPT, Google Bard and Bing Chat.
Example: Asking ChatGPT for board game rules.
For commonly known board games such as Monopoly and Uno, ChatGPT was able to identify the core rules of each game, however it omitted a number of less frequent rules.
For newer games such as the word guessing game Codenames, ChatGPT explained the core game but missed out a few key mechanics and did not explain rules for creating clues (a vital part of the game). If this was for an academic topic used as the sole basis of research for an assignment there would have been missing gaps in knowledge that would have resulted in a lower mark.
When tested on a lesser known game created in 2022 (Feed the Kraken), ChatGPT was completely incorrect and essentially created it's own game on the theme. Therefore it is important to be aware that with newer or niche topics it is more likely that the information would be incorrect.
With any information gained through AI, it is important that you ask questions of its validity. Check the information elsewhere before believing it to be true. If you need support in finding and evaluating information please see our critical thinking guides.
It is important to understand how generative AI tools work, in order to be aware of some of their weaknesses.
Effectively, an AI tool like Chat GPT provides you with what an answer to your question looks like. So if you ask it to write an essay on a particular topic, it will predict what it thinks an essay on that topic should look like, based on the data it is trained on and the way it is coded, rather than actually being an exact and precise answer to your question. Essays refer to facts and figures, so it will include facts and figures - but these may not be accurate, it may have made them up! Essays include academic references, so it will include academic references - but again, these may have been made up!
There have been many stories in the news recently about information provided by AI tools that turned out to be false:
A lawyer citing fake cases in court
The Guardian discovering AI is referring to fake Guardian articles
An American professor being named in a non-existent sexual harassment scandal
Fake quotes from political figures
It is vitally important that you are aware that AI tools do this and know how to independently verify the information provided.
AI tools are also only as good as the data used to train them. If that data is biased, one-sided or incomplete, the outputs from that AI tool will be too. For example, many current AI tools are very US-centric, because most of them are produced by US companies using data that comes from and contains mostly US sources. This can distort the answers these AI tools provide.
For example, if an AI tool is designed to help you identify flowers, but the data it uses only covers North American flowers, it is not going to be very helpful in identifying flowers from outside North America! It may also tell you incorrectly that the flower you are trying to identify is a North American one, because this may be the closest match in its databank. If you are not aware of this issue, and accept the answer you are given, you will go away thinking you have correctly identified your flower, when you haven't at all!