nestauk / narrowing_ai_research Goto Github PK

• Might be missing AI papers further away from the applications?
• Analysis of diversity taking a sample of papers every year
• Use distributions to decide thresholds instead of hard values

• Look for papers that are highly cited by AI papers but not in our categories

• Do non-dl topics publish in arXiv?
• Does this matter?

• Take samples by year and consider evolution
• Compare over multi-year periods

• Highlight specific features of DL that make it risky
• Focus the lit review on the role of the private sector in AI research
• Clarify mathematical expression re: technology performance in all contexts
• Clarify the link between context and technological diversity
• Change the header to avoid the perception all literatures about the direction of technical change agree
• Link the discussion to the table which connects threads in the literature to trends in industry
• Discuss further the role of technological diversity in reducing the risks created by homogenisation
• Improve section titles eg. literature review
• Clarify level of analysis - individual artefacts versus fields (internet?)
• Stronger focus on the product lifecycle / dominant design literature
• More explicit discussion of the reasons why corporates publish

• More careful with the use of 'how' in the research questions
• Refer to publications instead of 'research in arXiv'
• We are not testing hypotheses formally so change the way we talk about research questions, perhaps linking them to the 'process table'?
• Advantages of using arXiv over traditional sources
• Clarify footnote 8
• Explain potential biases produced by matching error
• Explain how company names are disambiguated (eg IBM vs International Business Machines)
• Remove mathematical notation to avoid confusion
• Figure 3.11 add indices
• Interpretation of the evolution of activity in different categories - remove?
• Explain how what is done in 3.3.4 contributes to the rest of the analysis
• Add diagram explaining analytical pipeline?
• Explain how topics are allocated to categories
• Start with descriptive analysis of levels of activity before moving to topics
• Add various references

• Include references to potential drivers of differences between academia and industry
• Focus a bit more on the drivers of specialisation considering differences between academic institutions and industry
• Highlight who is driving changes over time (academia rather than industry)
• Change description to refer to 'organisations responsible for diversity' rather than drivers
• Remove 4.2 or foreground its introduction further?

• Unnecessary nesta requirements
• Fasttext parameter
• Missing metadata

• Explain the methodology better
• Some robustness testing using an alternative topic model?
• Provide a list of topics

Includes:

• Removing low impact contributions
• Comparing the impact of corporate interventions