Ethical questions, risks, issues, framework and tools to help create responsible AI products and services.
This is a framework and non-exhaustive list of issues, questions and risks when trying to create an ethical / responsible AI service / product. This is a work in progress–get involved by creating issues and pull requests for this repo or contacting us [email protected]. This was created in part by Ben Byford of ethicalby.design.
Framework docs:
The Issues file is a list split into broad categories that relate to the type of people the question relates to, or the general area of discussion. E.g. Data, Design, Research etc. This feeds into the general guidance of the Framework which relates to this list and the roles therein.
This framework is used by Ethical By Design as a method for helping businesses create responsible AI services. The framework is used along with Design, Business and Brand strategy workshops, leadership and educational talks on AI and Ethics, governance frameworks and context specific research and cultural change initiatives.
As such, this framework shouldn't be viewed as a final product or a complete method rather a tool to help an organisation discover issues they may have overlooked and put better practises in place to mitigate inherent issues with implementing AI technologies in the real world (e.g. other than in simulation). For a short snapshot of AI project considerations, please see the questions doc. For a consultation please see Ethical By Design.
The Governance doc explores ways to structure ethical consideration into project life-cycles and ways to give teeth to ethical principle stipulations.
The Workshops file is a guidance document for doing specific workshops with your team at specific times talked about in the Framework. You can also check out the Workshop doc for links to other peoples workshops and for inspiration for your own.
Final caveat: This work is not relevant to AGI research, singularity or theory of mind but refers to a person/business interested in or currently using Machine Learning, genetic algorithms, expert systems etc, in a larger system context for their service / product.
There is an assumption going into this framework that you and/or your team/company have a base level of understanding, competency, and willingness to create the best services you possibly can for your customers, citizens and the environment. If this assumption is incorrect then this may not help you (and I would recommend doing some self reflection). If this is true, you may find that some of the content here you may know, are currently implementing or indeed have never thought about before. This is fine, use what you need and feedback where things don't make sense.
My hope is that this framework can enable more people to self-serve better technological services or prime themselves before getting dedicated consultancies like Ethical By Design to help with such activates.
TODO: general framework advice
The issues doc is split into categories that are board approximations for issues that relate to a type of person within an organisation best suited to make them. Each category will stipulate these people before each list. This doesn't mean that if that person isn't what relates to you or your experience then not to read that section, but it is a proxy for whom these decisions or choices often fall. For example it may be advised that all decision making parties within the organisation read the entire list and framework, some of it may go over individual's heads but it will mean you are all on-boarded to issues that arise in different areas of the organisation and can discuss on a similar level of understanding.
Many topics could fit into various places, but to avoid duplication categories have been placed with the greatest resonance with that category and decision maker/experience. Each category is a reflection of where approximately in the pipeline these decisions get made or where issues generally arise. Most issues indeed arise because of human decisions.
An even better idea would be to use this framework as an advocacy opportunity to take some of these issues to the decision making parties within your organisation and pertain to whether they have already look at / discussed the issues in question. Start opening up dialogues with your colleagues on ethical issues and foster a culture of ethical reflection.
As with the technology itself, this framework will evolve and update to reflect new research, experiences and case studies. Again this is only a helpful tool when looking to use AI/ML algorithms to better understand their affect on society, NOT a definitive HOW TO GUIDE.
The framework is largely written in the negative e.g. What negative environmental impact is your design going to have? However, it may also be useful to use the framework to concentrate on the positive by flipping some of the meanings to it's binary positive where appropriate e.g. What positive environmental impacts will this design have?
Much of the framing of this document is around the use of AI technologies, however as you're see in the issues section, much of the thinking found here can be used for technology projects generally. Building digital services will have many of the same issues here whether using AI technology or not. I've highlighted what I consider AI specific issues within the docs.
Format key:
Category
-> persons to who in an organisation this may relate
[general or specific technology consideration]
- issue / problem / technology area [tags, tag...]
- General issue articulation e.g. problem space and/or question
- reasons for consideration
- possible solutions / directions of travel
- references
- possible solutions / directions of travel
- reasons for consideration
- General issue articulation e.g. problem space and/or question
This framework is NOT a collection of general AI Ethics principles, please refer to the European Commission's Ethics guidelines for trustworthy AI or the OECD Principles on AI which have been widely accepted. Also here is a list of many more put out by various companies and organisations machine-ethics.net/machine-ethics-compendium. This Framework is to be used more practically at a lower level view from the more general Ethics guidelines or principles.
This project and Ethical By Design headed by Ben Byford welcome issue and pull requests to this repository. To find out more about us and what we're doing please find us here https://ethicalby.design or on twitter here @ethicalbydesign.
Why are we even considering AI ethics? Hopefully though this endeavour we're able to reduce harm and negative impacts to humans, animals and the environment while maximising positive outcomes for society. A healthy dose of reflection, some design tools, governance structures, research and technology knowledge can help us achieve this aim.
References to research can be found in the References document.
