1. Promoting Science-Policy Interfaces
How can Artificial Intelligence Support Crisis
Response and Policy Making?
NPS Egypt Seminar: Food Prices, Poverty, and Household Diets
May 22, 2023
Institute of National Planning (INP)
Presenter*: Clemens Breisinger
Based on a forthcoming blog by Breisinger, Keenan, Koo et al.
Science informs polices and crisis response
• Science can help designing
policies and responding to crises
• Science-based policy advice must
• Credible & Relevant
• Artificial intelligence can quickly
produce easy to understand policy
information and advice.
• Recommendations are broadly consistent with existing literature
• AI-powered chatbots have been trained on large text datasets scraped from
• The quality of the model-generated responses is only as good as the quality
of the data they are trained on.
• Recent analysis shows that open access publications from IFPRI and CGIAR
knowledge repositories are among those sources.
• AI tools draw on multiple sources and have the potential to reduce bias that can
arise from individual researchers.
• AI based policy recommendations can be generated within seconds.
• Information generated by chatbots does not (yet) disclose the sources
• Legal concerns about using content without consent, especially when it comes to the
use of copy-righted information from private companies.
• Current versions of AI powered chatbots are biased towards international sources,
implying less reliance on knowledge repositories from countries in the global south.
Possible solutions and next steps
• Customize AI:
• This requires targeting AI applications to custom-built repositories and moving beyond standard
AI applications (e.g. ChatGPT) and programming custom applications.
• This also provides an opportunity to include more country-specific data sources, for example
Egyptian university libraries, publications of Egyptian research institutions
• Applications such as “Consensus” can identify the sources used in their responses and identify scientists
who can provide verbal or written input to policymakers designing new policies.
• Customized AI can be used to produce “first-cut” policy recommendations and then advisors or scientific
advisory committees can validate them before passing on to policy makers.
• The CGIAR Initiatives on National Policies and Strategies and Digital Innovations together with national
partners are planning to develop a customized AI-chatbot combined with a scientific advisory mechanism