Responsible
AI

Blend is committed to the responsible use of Artificial Intelligence across our organization. We recognize the remarkable power to drive competitive advantage and improve productivity, while acknowledging the significant ethical and societal implications it presents.
‍
Our clients trust us to manage with a strong ethical foundation across all aspects of our business, and AI is no exception. We adhere to the highest standards of transparency, fairness, privacy, accountability, and social and economic benefit.

Principles

Blend is committed to delivering AI that is based on a strong set of principles that ensure safe and effective use.

Beneficial

Our AI systems support human goals and keep humans at the center of the experience

Consider both economic and ethical factors to benefit people and society at large

Maximize their benefits while addressing any potential harms

Mitigate potential negative impacts, such as job displacement or the amplification of bias and other harmful content

Transparent

Transparency helps our stakeholders to make informed decisions about using our AI systems which ultimately builds trust and conﬁdence

Clear and consistent documentation using data and model cards provides the visibility to understand the proper uses of the system

AI should operate transparently, with clear explanations about how decisions are made, the algorithms used, and the limitations of the system

Fair

Along with transparency, fairness in AI systems mitigates the potential for individuals or groups to be systematically disadvantaged

Minimize biases in our AI systems by rigorously testing our models to uncover and address any unintended biases

Ensure that data used to train these models is complete, accurate, and representative of the diversity of the user base

Contribute to social justice and not exacerbate existing inequalities

Privacy

Blend respects the privacy of our clients, their users, and our stakeholders

We comply with best practices and legal requirements for anonymizing, encrypting, and safeguarding all data we use

We will never use the data provided by our clients for any purpose other than those expressly agreed upon

Safe

An asian girl playing with interactivity

Secure design protects the interests of organizations by protecting it from misuse

Safety principles ensure respect for privacy and the people who interact with our systems

Protect both the AI system and the underlying data

Accountable

We keep humans are at the core to ensure accountability

Deployed systems provide a channel for groups or individual to voice concerns if they feel unfairly impacted by AI systems

Process

Blend builds responsible AI into our development
process to ensure fairness, privacy, security,
and overall accountability

Planning
and Design

Consider and document the intended use
‍

Establish the ethical guidelines to which the model will adhere
Establish intended goals around principles like fairness, privacy, security, and transparency
Clarify limitations/risks associated with the model and the underlying data, including potential bias, application context, potential legal/regulatory impacts

Data Assessment
and Application

Data used for training must be representative of the users who will interact with the system, safeguarding against discrimination and upholding privacy and security measures
‍

Following regulatory environment
Ensure that data sources are reliable, and that data collection methods are ethical and respect privacy, including any required consent
Assess data for potential biases and take steps to mitigate them

System
Design

System designs will leverage best practices in MLOps to ensure that the system is safe and reliable

Document decisions and trade-offs made during system design
Ensure that the system is robust
Procedures are in place to monitor and audit its use

Model
Development

When possible, use Techniques (algorithms, training, and validation) that provide explainable outcomes with a focus on accuracy

Qualitative and Quantitative.
Document decisions and trade-offs made during development and deployment
Key components of the model and results of the development process will be documented

Model
Deployment

Models will be deployed in a proper MLOps framework
‍

Clear documentation of risk and limitations
Regular monitoring to ensure it remains fair and unbiased as data and social norms evolve

Communicating
About Models and AI

Blend is committed to providing clear documentation and transparency for all AI models and systems to ensure well-informed application. We leverage model cards as a template to maintain and communicate information about all AI systems that we create.

Model Details
Basic information regarding the model version, type and other details

Person or organization developing model

Model date/version

Model type

Information about training algorithms, parameters, fair- ness constraints or other applied approaches, and features

Paper or other resource for more information

Citation details / License

Intended Use
Allows users to quickly grasp what the model should and should not be used for, and why it was created

Primary Intended Uses

Factors
Summary of model performance across groups, instrumentation, and environments

Relevant factors: What are foreseeable salient factors for which model performance may vary, and how were these determined?

Evaluation factors: Which factors are being reported, and why were these chosen? If the relevant factors and evaluation factors are different, why?

Metrics
Chosen to reflect potential real-world impacts of the model

Model performance measures

Decision Thresholds

Variation approaches

Training & Evaluation
Data Considerations when training AI

When possible, training 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

Model type

Details on the dataset(s) used for the quantitative analyses in the card. Datasets. Motivation. Preprocessing

Ethical Considerations, Caveats, and Recommendations
How should the model be used (or not used)

Ethical concerns that could arise if the model is not used properly

Indications of potential issues with fairness or privacy

Model type

Proper use of underlying data to maintain security and efficacy of the model

Responsible AI

Principles

Beneficial

Transparent

Fair

Privacy

Safe

Accountable

Process

Planningand Design

Data Assessmentand Application

SystemDesign

ModelDevelopment

ModelDeployment

Communicating About Models and AI

Model Details Basic information regarding the model version, type and other details

Intended Use Allows users to quickly grasp what the model should and should not be used for, and why it was created

Factors Summary of model performance across groups, instrumentation, and environments

Metrics Chosen to reflect potential real-world impacts of the model

Training & Evaluation Data Considerations when training AI

Ethical Considerations, Caveats, and RecommendationsHow should the model be used (or not used)

AI Transformation Challenge

AI Transformation Challenge

Responsible
AI

Planning
and Design

Data Assessment
and Application

System
Design

Model
Development

Model
Deployment

Communicating
About Models and AI

Model Details
Basic information regarding the model version, type and other details

Intended Use
Allows users to quickly grasp what the model should and should not be used for, and why it was created

Factors
Summary of model performance across groups, instrumentation, and environments

Metrics
Chosen to reflect potential real-world impacts of the model

Training & Evaluation
Data Considerations when training AI

Ethical Considerations, Caveats, and Recommendations
How should the model be used (or not used)