Skip to Main Content

Exploring Text Analysis for Digital Scholarship: Tools, AI, and Responsible Practices

Limitations of AI in text analysis

  • Bias in training data and results [Note 1]
  • Lack of transparency in algorithms “black boxes” 
    - difficult to understand how specific text analysis results are produced [Note 2]
  • Accountability and authorship
  • Copyright and Intellectual Property  [Note 3]
  • Ethical concerns (e.g., plagiarism, misuse of AI-generated content)
  • Decline in linguistic diversity [Note 4]

  • Over-reliance on AI vs. critical thinking [Note 5]

  • Generation of False or Misleading Content: AI models can produce plausible-sounding but inaccurate or fabricated information (known as “hallucinations”), which can mislead users or propagate misinformation if not properly checked. [Note 6] (https://doi.org/10.48550/arXiv.2402.08323)

  • Difficulty in Data Correction or Deletion: Once personal data is incorporated into large language models, it can be challenging or impossible for individuals to access, correct, or delete their information, contravening data protection principles. (https://www.pcpd.org.hk/english/news_events/newspaper/newspaper_202309a.html)

[Note 1]

[Note 2

[Note 3]

[Note 4]

Source : Empirical evidence of Large Language Model’s influence on human spoken communication

  • ChatGPT influences language with words like "delve", "realm" and "meticulous".
  • Analysis of 360K+ videos and 771K podcasts shows increased usage 18 months post-launch.

[Note 5]

[Note 6]