Unlocking the Potential of AI-Driven Chatbots in Medicine and Science

Pablo Riascos Caldas, Santiago Guzman, Bernardo Trevino, Ryan Dervay.

---

AI in Medicine: Opportunities and Challenges

The use of artificial intelligence (AI) in medical research brings exciting possibilities but also comes with risks. As AI-generated content becomes more common, there’s a danger of spreading false information if it’s not properly verified. If this happens repeatedly, inaccurate information could become part of future AI datasets, making lies seem like the truth.

Balancing the Pros and Cons of AI

AI has many benefits, but it must be used carefully and responsibly. Instead of avoiding AI, it’s better to focus on reducing risks by tailoring its outputs to meet specific needs. Our team recently experimented with creating a safe, research-based chatbot. This process involved trial and error, as we’re not experts in computer science, but we learned a lot along the way.

We would like to share our initial experiences with this straightforward list showing a few methods we decided to test:

1. Custom OpenAI GPT: Helpful But Imperfect

With a ChatGPT subscription, you can create custom AI models by setting specific instructions and functions.

Pros:

• Detailed Answers: Provides in-depth responses to complex questions.

• Easy to Use: Simple to set up and integrate into workflows.

• Team Sharing: Custom models can be shared for group projects.

  
  

Cons:

• Biases: Relies on pre-existing data that may include outdated or wrong information.

• Errors: Sometimes generates responses that seem right but are incorrect.

2. Private Chatbot/LLM - Llama3: A Good Start

We used an open-source model called Llama3 for our chatbot. These models analyze large amounts of text to create human-like responses.

Pros:

• Accurate Outputs: Handles detailed and complex queries.

• Data Privacy: Running the model locally ensures better security.

• Free to Use: A cost-effective option.

Cons:

• Resource Heavy: Needs a powerful GPU to work, which can limit access.

• Mistakes: Still prone to creating inaccurate or made-up answers.

3. Retrieval-Augmented Generation (RAG): Looking Ahead

RAG improves accuracy by combining text generation with real-time information retrieval. While we haven’t implemented this yet, it’s a promising next step to enhance our chatbot.

Future Steps:

• Pairing RAG with fine-tuning could create a tool more precise and reliable.

4. Coding Your Own Chatbot: Two Methods

   
   

Standalone Python (.py)

This involves creating a chatbot in Python that interacts directly with local files.

Pros:

• Secure: Works with local files, keeping data private.

• Stable: Operates as a standalone script.

Cons:

• Manual: Requires input through the terminal, which can be tedious.

• Limited Features: Doesn’t handle multiple files or complex tasks well.

Python with HTML Interface

Adding an HTML interface makes the chatbot easier to use by enabling file uploads and retrieval.

Pros:

• User-Friendly: Simplifies interaction with the chatbot.

• Cost-Effective: Free to develop.

  
  

Cons:

• Basic Functions: Only works for simple text searches.

• Limited Responses: Can’t handle deeper, more complex queries.

5. API Platforms: HeyGTP and Chatbase.co

   
   

API platforms like HeyGTP and Chatbase.co make it easier to use AI in projects.

Pros:

• Simple to Use: User-friendly platforms.

• Reliable Responses: Delivers solid answers when given good input.

Cons:

• Token Use: Requires careful tracking of usage.

• Dependency: Relies on external APIs.

• Errors: Can still produce inaccurate information.

Conclusion: AI as a Partner, Not a Replacement

AI can improve research speed and depth, but it’s not a replacement for human knowledge. Medical professionals play a key role in analyzing data and drawing meaningful conclusions. By combining AI with human expertise, we can make progress while staying safe and accurate.

Keep innovating and stay curious!