-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
Showing
3 changed files
with
172 additions
and
0 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,48 @@ | ||
--- | ||
title: "AI Ethics & Transparency Roadmap Template" | ||
date: 2019-09-03T11:02:05+06:00 | ||
type: "post" | ||
weight : 2 | ||
--- | ||
## **Team Name** AI Transparency Roadmap | ||
|
||
**Executive Summary** | ||
|
||
This is your at a glance version of the document. Details and resources attached | ||
|
||
* Milestone 1 - **Understanding the Data Flow** | ||
* Data Ecosystem Map - Open Ticket | ||
* Information Sharing Protocol - Open Ticket | ||
* Milestone 2 - **Understanding the Algorithm** | ||
* Dataset Structure - Open Ticket | ||
* Common Traps Mitigations - Open Ticket | ||
* Milestone 3 - **Sharing the Model** | ||
* Model Card Created - Open Ticket | ||
|
||
Understanding Data Flow | ||
|
||
* Primary Goal: Create documentation that ensures you know what data you're using and how you plan on sharing it. | ||
|
||
This first milestone is about initially understanding the underlying data powering your model. This is generally important for understanding the plumbing and getting an idea about privacy and implications that may arise from sharing data. | ||
|
||
**Outcomes:** | ||
|
||
* [**Data Ecosystem Map**](https://docs.google.com/document/d/18Zg2JwUDJajVDX5VU0vMijL-c9yfumeAUYDc7rgC4iQ/edit#) **:** This is to document where their data is coming from, what's using it, stakeholders, etc. The benefit here is not just helping organize how data collection will work in production but also projecting partnerships which may need to be formed in the future. | ||
* [**Information Sharing Protocol**](https://docs.google.com/document/d/1MISHbWU7KGo4Z4AR-b222f6uXrtpQ-GJiJemGYoL--E/edit#) **:** This document is all about conducting an initial assessment of the sensitivity of information you are collecting, who you want to share it with, and how. It's fairly well thought out but is worth a review and potentially some edits. If you are in the EU, it might just be better to do a[Data Protection Impact Assessment](https://ico.org.uk/for-organisations/guide-to-data-protection/guide-to-the-general-data-protection-regulation-gdpr/data-protection-impact-assessments-dpias/what-is-a-dpia/) | ||
|
||
Understanding the Algorithm | ||
|
||
* Primary Goal: Create documentation to ensure your whole team understand exactly how your model will be working. A large part of this will be determining the method of managing your datasets as they evolve over the development of the program. | ||
|
||
**Outcomes:** | ||
|
||
* [**Determine Dataset Structure**](https://humanitarian.atlassian.net/wiki/spaces/imtoolbox/pages/61734950/File+and+Dataset+Management): As you create new data sets, update them, and remove old datasets, you should follow an easy to follow protocol for keeping track of your various data sets as they're updated and deprecated. | ||
* [**Common Traps When Building Models**](https://drive.google.com/open?id=10Bun6ucDUbTu7ALvoL_Ul0VhTSwJqkaq1zSJ7dvGRls): Review the following traps and document potential risks of your application falling into any of these traps. | ||
|
||
Documenting the Algorithm | ||
|
||
* Primary Goal: Produce documentation that should be shared with the model. | ||
|
||
**Outcomes:** | ||
|
||
* [**Document Machine Learning Model Card**](https://drive.google.com/open?id=1n9GhLGqhNCKuQei2RJfgTcUJroZVW8A3eYsS4eY_xFw) **:** Before you release or begin production use of your machine learning model, you can use the linked template to comprehensively document the effects, background, and purpose of your model. This will serve as a core document showing how the model makes the decision it does. |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,78 @@ | ||
--- | ||
title: "Machine Learning Model Card" | ||
date: 2019-09-03T11:02:05+06:00 | ||
type: "post" | ||
weight : 2 | ||
--- | ||
## **Model Title** | ||
|
||
_The following is simply an editable version of the model card proposed in_ [_https://arxiv.org/abs/1810.03993_](https://arxiv.org/abs/1810.03993) | ||
|
||
_Model Details._ | ||
|
||
Basic information about the model. | ||
|
||
- Person or organization developing model | ||
- Date of last update | ||
- Model version | ||
- Model type (Information about training algorithms, parameters, fairness constraints or other applied approaches, and features) | ||
- Paper or other resource for more information | ||
- Citation details | ||
- License | ||
- Maintainer contact details | ||
|
||
_Intended Use_ | ||
|
||
Use cases that were envisioned during development. | ||
|
||
- Primary intended uses | ||
- Primary intended users | ||
- Out-of-scope use cases | ||
|
||
_Factors_ | ||
|
||
Factors could include demographic or phenotypic groups, environmental conditions, technical attributes, or others. | ||
|
||
- Relevant factors | ||
- Evaluation factors | ||
|
||
_Metrics_ | ||
|
||
Metrics should be chosen to reflect potential real-world impacts of the model. | ||
|
||
- Model performance measures | ||
- Decision thresholds | ||
- Variation approaches | ||
|
||
_Evaluation Data_ | ||
|
||
Details on the dataset(s) used for the quantitative analyses in the card. | ||
|
||
- Datasets | ||
- Motivation | ||
- Preprocessing | ||
|
||
_Training Data._ | ||
|
||
May not be possible to provide in practice. When possible, this section should mirror Evaluation Data. If such detail is not possible, minimal allowable information should be provided here, such as details of the distribution over various factors in the training datasets. | ||
|
||
_Quantitative Analyses_ | ||
|
||
Quantitative analyses should be broken down by the chosen factors. | ||
|
||
- Unitary results | ||
- Intersectional results | ||
|
||
_Ethical Considerations_ | ||
|
||
In this section, you should explore and document the following: | ||
|
||
- **Sensitive Data** _: Does the model use any sensitive data (e.g., protected classes)?_ | ||
- **Human life** _: Is the model intended to inform decisions about matters central to human life or flourishing – e.g., health or safety? Or could it be used in such a way?_ | ||
- **Mitigations** _: What risk mitigation strategies were used during model development?_ | ||
- **Risks and harms** _: What risks may be present in model usage? Try to identify the potential recipients, likelihood, and magnitude of harms. If these cannot be determined, note that they were considered but remain unknown._ | ||
- **Use cases** _: Are there any known model use cases that are especially fraught?_ | ||
|
||
_Caveats and Recommendations_ | ||
|
||
This section should list additional concerns that were not covered in the previous sections. |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,46 @@ | ||
--- | ||
title: "Common Traps When Building AI Systems" | ||
date: 2019-09-03T11:02:05+06:00 | ||
type: "post" | ||
weight : 2 | ||
--- | ||
## **Avoiding Common ML Traps** | ||
|
||
_Assessing common mistakes that lead to unintended side effects._ | ||
|
||
**The Framing Trap** | ||
|
||
_Failure to model the entire system over which a social criterion, such as fairness, will be enforced._ | ||
|
||
For this trap, we will want to look at our outcome variables. Are these variables a proxy of the actual outcome you wish to achieve? What evidence of existing negative bias currently exists with regards to these variables? | ||
|
||
**The Portability Trap** | ||
|
||
_Failure to understand how repurposing algorithmic solutions designed for one social context_ | ||
|
||
_may be misleading, inaccurate, or otherwise do harm when applied to a different context_ | ||
|
||
Here, you will want to be fully understanding of the context in which this model is being built for and the context in which you will be using it. There should be clear documentation distributed across the entire team focusing on this. What stakeholders do you expect to have an impact on where this technology will be used. Are they informed? | ||
|
||
**The Formalism Trap** | ||
|
||
_Failure to account for the full meaning of social concepts such as fairness, which can be_ | ||
|
||
_procedural, contextual, and contestable, and cannot be resolved through mathematical formalisms_ | ||
|
||
How does your chosen outcome continue to implement and solve existing procedural "catches" in order to ensure the decision making process is fair? What method of recourse are available for those who are unfairly judged? | ||
|
||
**The Ripple Effect Trap** | ||
|
||
_Failure to understand how the insertion of technology into an existing social system_ | ||
|
||
_changes the behaviors and embedded values of the pre-existing system_ | ||
|
||
How do you think the introduction of this recommendation system will affect your users? What changes in behavior do you intend to change? Can you think of any possible changes in behavior that you don't intend as a result of your software? | ||
|
||
**The Solutionism Trap** | ||
|
||
_Failure to recognize the possibility that the best solution to a problem may not involve technology_ | ||
|
||
Will this technology elevate social values which can be quantified? Will it devalue those which cannot? What values might take a back seat if this technology is implemented? | ||
|