Responsible AI Policy

How we use AI & Machine Learning.

Responsible AI Policy

At Submit, we are committed to using Artificial Intelligence (AI) and Machine Learning (ML) responsibly to create a safe and secure online environment for our users. Our AI systems play a crucial role in content moderation, anti-spam efforts, and bolstering our platform’s security. Through advanced ML algorithms, we efficiently identify and mitigate potentially harmful or abusive content, detect and block spam, and prevent unauthorized access and other security threats. This proactive approach enables us to maintain the integrity of our platform and ensure a positive experience for our community.

In line with our commitment to user privacy and data protection, we strictly adhere to ethical AI practices. This includes a firm stance on not using any personally identifying information (PII) from our users to train our AI models. We respect user privacy to the utmost degree; thus, we do not utilize user-uploaded or generated media for model training purposes. Our AI and ML systems are designed to operate using non-identifiable aggregate data, ensuring that individual user data remains private and secure. Furthermore, we uphold a strict policy against sharing or selling our trained models and the data used to train them. While we may explore opportunities to offer our AI technology itself to others, we ensure that this does not include any user-generated data or the intelligence derived from it, thus maintaining the confidentiality and integrity of our users’ information.

To make it extremely clear: Submit never has and never will sell any user data or any trained model data, period.

Model Training

In training our AI models, we are committed to ethical data use and transparency. Our image-based models are developed using open-source and publicly available datasets, ensuring that we do not utilize private, personal, or user-specific data for training purposes. This approach allows us to leverage the power of AI while upholding our commitment to user privacy and data protection.

For our text-based models, we employ stringent data handling practices to ensure that all data used is anonymized and never associated with any individual user at any point in time. This ensures the integrity and confidentiality of user data while allowing us to effectively train our models to detect and manage inappropriate or harmful content.

We also collaborate with public and government organizations to enhance the efficacy of our content moderation systems. These organizations provide specific types of data, such as image hashes, which help us identify and take action against content that violates our platform rules or the law. It is important to note that we do not share any information with these organizations, with the sole exception being when a user uploads content featuring an individual under the age of 18. In such cases, our priority is to protect the safety and privacy of minors, and we take the necessary steps to alert the relevant authorities in compliance with legal and ethical standards.

Dataset Integrity and Diversity

Ensuring the integrity, diversity, and accuracy of the datasets used to train our AI models is fundamental to our commitment to responsible AI practices. Our datasets are curated from diverse sources to promote inclusivity and reduce bias, and are rigorously evaluated to reflect a broad spectrum of demographics and scenarios. This diversity helps our AI systems to perform fairly and effectively across varied contexts and user groups.

To guarantee the accuracy and reliability of our datasets, we employ a multi-tiered approach to their selection and use:

  • Sourcing Transparency: We select datasets that are not only publicly available but also well-documented in terms of their creation and scoring methodologies. This transparency helps us assess the quality and appropriateness of the data for training purposes.
  • Review and Validation: The scoring of our datasets, initially performed by the dataset providers, is subjected to additional scrutiny by our team. We assess the original scoring for accuracy and bias, making adjustments where necessary to align with our ethical standards and the specific needs of our platform.
  • Continuous Monitoring and Updates: Post-deployment, our AI models and their underlying datasets undergo continuous monitoring. This allows us to identify any performance issues or emerging biases and to update the datasets and models accordingly to maintain their effectiveness and fairness.

We believe that these measures are essential for ensuring that our AI-driven services remain effective and ethical.

Transparency & Explainability

In our commitment to transparency and explainability within our AI-driven processes, we prioritize human judgment in all content moderation decisions. While our AI systems are adept at flagging content that potentially violates our platform rules or legal standards, the final decision to remove content or enforce penalties against a user is always made by a human moderator. This approach ensures that the nuances and context of each situation are thoroughly considered, reducing the risk of errors that might occur with automated decision-making.

Our AI models serve as tools to support and streamline the moderation process, identifying and categorizing content at scale to assist our human review teams. When AI flags content, it is subsequently queued for review by our trained moderators who evaluate the material against our community guidelines and legal requirements. This human-in-the-loop model ensures that decisions are fair, balanced, and respectful of the complexities inherent in content moderation.

We also maintain a system of accountability and auditability for decisions made, with records of both AI flagging and human moderation actions. This process allows for ongoing training and refinement of our AI systems, ensuring they are aligned with our ethical standards and societal expectations. Users affected by moderation decisions have access to a clear explanation of the reason for the action taken, underscoring our commitment to transparency and fostering trust in our platform.