POLICY

Artificial intelligence at work

Dos and don'ts

DALL-E, "A classical painting showing the eschatological battle between a good robot and an evil robot", 5 May 2023.

• mention all use of an AI in documents you share with colleagues, explaining which AI you used, how you used it (including any prompt), and (if relevant) why you used it
• use AIs to self-edit and improve your grammar, vocabulary, style, and structure, while always carefully editing the AIs’ output before sharing with anyone
• use AIs to summarize the main argument of your text or section, to check if your point is coming through clearly 
• use AIs to reduce your word count or delete redundancies  
• use AIs to seek inspiration on how to build an outline or a template, while never settling for what the AI suggests
• ask AIs to explain their suggestions and edits, to learn from the process
• ask AIs to vary their response to a prompt, to explore several options and trigger more thoughts of your own

• use an AI to generate a first draft for you to edit, as trying to add in what makes your work distinctive (novel fieldwork, original analysis, and your voice as a writer) will only produce a garbled text that will have to be entirely rewritten anyway
• neglect to add your own value to any output generated by an AI, short of which your own work becomes valueless
• cite an AI as a source
• trust an AI to preserve the meaning of the text you ask it to edit; it is likely to change the meaning, including in critical areas such as direct quotes

• use AIs to generate an indicative overview of a topic
• ask AIs to define and unpack technical concepts
• ask AIs to list and summarize potential sources on a topic, to help you form an opinion on what primary sources to consult
• ask AIs to list institutions most relevant to a topic
• ask AIs to suggest possible interview questions
• ask AIs to suggest a list of research questions you should ask in relation to your topic
• use AIs to role-play conversations 
• keep in mind that AIs are likely to store your user history in ways that may both violate your privacy and lend themselves to leaks

• trust any factual information (to include names, dates, events, numbers, and bibliographic information or URLs) delivered by an AI without cross-checking with at least one credible primary source
• use an AI to fact-check anything conclusively
• treat an AI’s output as a primary source, rather than a synthesis of other sources
• assume that an AI’s synthesis is satisfactorily unbiased or complete
• assume that an AI’s summary adequately represents the work being summarized

• use AIs to produce illustrations, while bearing in mind that original human work is the source of AI-generated images, and therefore should be promoted whenever possible
• duly credit any AI that helped you produce an image you share with others, explaining which AI you used and how you used it (including any prompt)

• assume that any image is authentic unless it is sourced professionally by a credible publisher and author
• share any image as proof of a factual claim, unless you can establish its authenticity
• use an AI to produce any deceptive image, even privately and for fun
• use an AI to produce images that emulate (and thus devalue) the style attached to a real creator, especially a living one
• assume that you can determine if an image is authentic, as tell-tale signs differ from one AI to another, and are likely to disappear entirely over time

• ask AIs for suggestions on how to perform a certain task (like writing a cover letter or making a punchy presentation) or how to approach a certain problem
• ask AIs to outline policies and explain best practices in areas where you feel your organization could improve
• use AIs to test your knowledge and rehearse for job interviews

• use AIs in areas such as recruitment, without having developed proven measures to guard against AIs' built-in biases 
• use AIs to make decisions that you couldn’t explicitly justify yourself
• use AIs to draft resumes and cover letters without taking the time to edit, elaborate, and inject your own skills, personality, and values

• try out any verified AI application or extension to familiarize yourself with these fast-evolving tools
• share your lessons learned as you go, with colleagues and even the public, to contribute to a collective learning journey

• download any unverified AI-powered application or extension on any device you use for work (check with your manager when in doubt)

• use an AI to outline, review, document, and possibly refactor your code
• use an AI to explain systematically code you don’t fully understand
• make the most of any time you gain from using an AI to focus your efforts on cleaning, testing, and documenting your code; tackling higher-level problems, such as product design and user experience; and keeping ahead of new technologies

• assume that libraries pulled into your code by an AI are up-to-date or even exist 
• assume that AIs follow coding best practices, which remain yours to understand and attain 
• neglect to carefully reread any code drafted or refactored by an AI, because bugs, missing features, or poorly designed features are entirely your responsibility 
• use an AI to document the trickiest parts of your code, because documenting it yourself gives you an opportunity to truly and fully understand it

• Understand the capabilities and limitations of AI before using it in research.
• Ensure that the data used to train the AI is of high quality and representative of the problem being studied.
• Regularly monitor and evaluate the AI's performance to ensure that it remains accurate and reliable.
• Consider ethical implications of using AI in research, such as bias and privacy concerns.
• Collaborate with experts in AI and related fields to ensure that the research is of high quality.
• Clearly document the AI algorithms and methodologies used in the research.
• Ensure that the AI's results are appropriately validated and tested.

• Rely solely on AI to make important decisions in research.
• Use AI without first understanding how it works and what it is capable of.
• Use AI to make decisions that are illegal or unethical.
• Use biased or incomplete data to train the AI.
• Overstate the capabilities of the AI in research.
• Use AI without proper validation and testing.
• Ignore ethical implications of using AI in research.

Illustration credits: Prompt driven image via DALL-E / currently unlicensed; Robot EmojiOne clipart / licensed by Creative Commons.