OptimalAI
RESEARCH AREAS
Practices for responsibility
A set of tangible practices underpin our responsibility in the research, development and application of AI. They guide its safe and ethical development.
AI as a force for good
We are driven by the belief artificial intelligence (AI) can be a transformative force for good. However, we also recognize the responsibility that comes with this potential. By adhering to a human-centered approach in fundamental AI research and real-world applications, we aim to create AI systems - and eventually Artificial General Intelligence (AGI) systems - that enhance the human experience, empower individuals, and contribute positively to humanity. OptimalAI's Responsibility in AI is a set of tangible practices that underpin our research, development and application of AI.
Practices
Ethical Considerations
We recognize the ethical implications of our work. We prioritize human interests and well-being, weaving ethical principles into the very fabric of our AI development. This includes a commitment to transparency and accountability in our research and decision-making processes.
Safety First
Human-centered AI should be designed with robust safety mechanisms to prevent unintended consequences and ensure that humans have control over the technology. This involves addressing issues related to the system's behavior, decision-making, and potential risks.
Collaboration
The impact of AI extends beyond organizational and national boundaries. To address the complex societal and ethical implications of AI, OptimalAI actively collaborates with leading AI research organizations. We engage in open and constructive dialogues to navigate the evolving landscape of AI technology responsibly.
Fairness
Equity
We are committed to building AI systems that treat all individuals fairly. This means working diligently to mitigate bias and discrimination, ensuring that AI decisions do not unfairly disadvantage any group or individual.
Algorithmic Transparency & Explainability
Making AI transparent and explainable to users is essential. Users should understand how and why the AI system reached its conclusion and have confidence in its decision-making processes. We strive to make the decision-making processes of all of our AI systems as transparent and comprehensible as possible and empower users to override or enhance that conclusion.
Interpretability
Explainability
We recognize the importance of AI systems being interpretable. Our goal is to provide users with the ability to understand the inner workings of AI models, allowing for trust and informed decision-making.
Privacy
Data Protection
OptimalAI places a strong emphasis on data protection. We employ robust measures to ensure the secure handling and storage of user data, respecting privacy rights and regulations.
Consent and Control
We empower individuals by providing control over their data. In our AI applications, we seek informed consent for data usage, respecting users' rights and choices.
Safety and Security
Robustness
Ensuring the robustness of AI systems against adversarial attacks and unintended behaviors is a critical aspect of our approach. We actively research and implement safeguards to prevent and mitigate risks.
Deployment Standards
Our responsibility extends to the deployment of AI systems. We incorporate safeguards to minimize risks and potential harm when AI is used in real-world applications.
Uncertainty and Traffic-Aware Active Learning
The traditional method of training a semantic parser - a tool used to understand natural language - involves collecting and annotating large amounts of data, which is expensive, time-consuming and raises privacy concerns as customer data needs to be handled by human annotators. Uncertainty and Traffic-Aware Active Learning is a new method that selects utterances (or spoken phrases) for annotation based on how often they appear in customer interactions and how confident the model is in understanding them. The technique proved significantly better than previous methods when tested on both an internal customer dataset and the Facebook Task Oriented Parsing (TOP) dataset. Remarkably, it achieved the same accuracy as the traditional random sampling method, but used 2,000 fewer annotations, thereby proving more efficient.