尤物视频

• 64% of researchers report AI as beneficial to their research, with over two-thirds (68%) using AI chatbots on the open web for their research, and just under half (49%) using machine translation.
• AI tools are most commonly used for discovering existing research (55%), summarizing existing research (46%), and editing the write-up of research (45%). Many (68%) agree that AI tools can reduce accessibility barriers, too.

“It reduces the time to locate arcane material. What might have taken hours/weeks in a library can be found instantaneously.”

• However, 35% of those in the arts and humanities never use AI tools in their research, as opposed to 15% and 18% for STM and the social sciences, respectively.

• Only 20% trust AI-generated outputs. 37% of researchers who haven’t used AI for their research report concerns about accuracy, bias, and trustworthiness, and only 26% find AI-generated literature summaries to be accurate.

“My trust in AI outputs would increase if the responses consistently included clear and verifiable sources (such as case law, treaties, or academic references), showed transparent reasoning for how conclusions are reached, and accurately distinguished between established legal authority and interpretation or opinion. It would also help if the AI acknowledged uncertainty where appropriate and avoided overgeneralization, especially in complex legal or academic topics.”

• However, 83% of those surveyed have established practices for checking citation sources generated by AI, and 80% always cross-check AI outputs for reliability.

• Based on respondents’ perceptions and behaviours towards AI, survey analysis identified five different profiles of researcher.
• These range from the least trusting of AI鈥攖he ‘persistent sceptics’鈥攚ho are most commonly found in the arts and humanities, to the ‘carefree casuals’, who are less likely to record their own AI use and are more common among later career stage respondents than early career researchers.
• Meanwhile, ‘frustrated trailblazers’ have benefited most from AI but are also the most concerned about the administrative burden of disclosing its use. They are more prevalent in the social sciences and STM disciplines, and among early career researchers.

“I am ethically opposed to AI. AI has been built on the stolen work of artists and writers. I am also concerned about the environmental consequences of AI. I grant that in some fields or in some inquiries, AI may be of some genuine use鈥攏ot in my research (which is about literature). But even when it has genuine use, the ethical and environmental concerns remain. If I were to use AI on my research, it wouldn’t be my research anymore.”

• Nearly three-quarters (74%) report uncertainty around AI disclosure requirements, while only 36% actively keep records of their AI use during research. One respondent acknowledged:

“I tend to make sure my AI use is undetectable as I’m not sure what to disclose.”

• 44% are concerned that disclosing AI use could negatively affect perceptions of their work鈥攎any of whom are in the arts and humanities鈥攚ith 35% finding AI disclosure to place an additional administrative burden on researchers.

• 45% of researchers report their institution’s approach to AI use as unclear, while 43% report a somewhat or very enabling approach.
• Looking at different research outputs over the last year, many researchers acknowledge that they do not fully disclose their use of AI tools:
  • 48% of those writing grant applications had not fully disclosed AI use, particularly in narrative drafting, translating or editing, and plain language summaries.
  • 46% of journal authors and 39% of book authors had not fully disclosed their AI use in preparing their manuscript, particularly in narrative drafting in each case. Only a third (30%) of book authors and 42% of journal authors were asked to attest whether AI was used in their manuscript submission.
  • 25% of peer reviewers had not fully disclosed AI use, particularly in language editing. 60% felt their AI use and disclosure complied with the reviewer policies, but the views were more mixed as to whether the journal policies on AI use were clear.

“Make the rules clear, coherent, and reasonable. Also, not all AI use should be disclosed (e.g., checking language), the same way that you do not disclose using a calculator. The question in the end is whether the author is the originator of the work in a deep sense. AI will be used.”

• Many authors want future AI disclosure policies to change, with some calling for clearer and more consistent guidelines on AI use and disclosure.
• Others suggest policies should support a move towards the normalization of AI as a tool with minimal disclosure.
• Looking at future AI disclosure policies to support researchers as readers of others’ work, many want to see standardized and mandatory AI disclosure introduced.

“The policy should protect normal responsible use of AI as a tool without forcing authors into vague or excessive reporting requirements.”