- One of the main virtues of Artificial Intelligence is that it allows analyzing large amounts of information in just a couple of seconds.
- The OpenAI company developed the ChatGPT chatbot based on the GPT-3 Artificial Intelligence language model.
- In the past, ChatGPT was able to answer highly complex tests such as the USMLE that all doctors require to practice their profession in the United States.
What until recently seemed exclusive to science fiction movies now happens in reality. The technology is now so powerful that it can act as if it were a human. The clearest example can be seen with ChatGPT, which has positioned itself as the fashion tool. But is his power so great that he can answer questions about medical topics?
If you still do not know well what we are talking about, it is first necessary to put it in context. For years there have been chatbotwhich are software that allows you to have a conversation in real time with automatic responses, which are previously established by a group of experts.
Can machines think like humans?
These types of tools are frequently found on the websites of banks. Its function is to offer an immediate response to users at any time. The biggest problem is that its wording is too flat and synthetic.
Instead, the next step in the technological chain is ChatGPT. It consists of a chatbot based on the language model by Artificial Intelligence GPT-3 and was developed by the company Open AI. For its elaboration, more than 175 million parameters were used and it was trained to carry out different activities related to language, from translation to text generation.
To get an idea of its reach, its algorithm is so powerful that it can hold conversations with anyone. Unlike other options, his answers are no longer monotonous and predictable, but he actually writes as if he were a human. In addition, you can write quite elaborate complete texts.
Technology answers questions about insulin
To test ChatGPT the user Dr Kabs He asked her some pretty specific medical questions. The result she shared through a video and what was obtained was the following.
First, he asked for the molecular weight of insulin. In just a couple of seconds he got the answer that it is 5,808 Daltons. When analyzing the correct information, it is verified that it is 6,000 Daltons, for which reason his statement is considered correct.
The next question was about the main insulin secretagogue. The correct answer is glucose and it was exactly what you indicated ChatGPT.
Then he asked a much more complex question that even some doctors do not answer correctly. In addition to pancreatic Epsilon cells, what other cells produce ghrelin in humans? The correct answer is PD/1 and in this particular case the tool got it wrong.
In the end, the technological tool had two errors out of 10 questions asked by the doctor. Therefore, it can be concluded that it had a fairly high percentage of effectiveness. At the same time, it is found that it is not perfect because it can make mistakes just like anyone else.
@dr.kabs I test an AI with medicine #fyp #fypシ #medicine #medschool
Also read:
ChatGPT surprises: The AI passes the exams to be a doctor
Artificial Intelligence, how does it help in the care of medical emergencies?
The 5 Most Promising Medical Advances by 2023: From Artificial Intelligence to 3-D Organ Printing