Augmented Intelligence Quotient
Nebuli’s AIQ (pronounced “IQ”) offers fully referenced and expert-driven large language models to help you deploy responsible AI and smart knowledge discovery systems faster.
AIQ helps you set standards that ensure your AI-powered systems are fair, transparent, explainable and aligned with the growing demands of digital ethics and regulations.
AIQ is Nebuli’s suite of specialist and fully cited large language models (LLMs) that focus specifically on vertical knowledge built on our Deep Vertical Understanding (DeepVU) framework.
With the DeepVU framework, we trained AIQ’s models using Deep Reinforcement Learning from Human Feedback (DRLHF) methodology but with a focus on specific vertical and societal parameters.
These parameters may include industrial operations, players, acronyms, issues and trends of a given domain while simultaneously addressing cultural, demographic and psychographic influences that may dictate outcomes and behaviours within this domain.
We also apply Federated Learning (FL), a distributed machine learning technique that allows multiple clients with their data and computation resources to collaborate and train AIQ’s models. This is particularly useful in cases involving private or sensitive datasets.
Determining What’s Relevant and Trustworthy through Vertical Understanding.
From our team’s collective experience in working with public and private data sources spanning several decades, we produced a global data map that groups the key sources based on their citation score, content quality, and accessibility.
By applying our DRLHF methods, we use deep neural networks to establish the ability to handle high-dimensional and complex datasets and automatically extract meaningful segments to achieve vertical understanding. The same model is applied to the behavioural models indicated within the extracted data segments linked to specific verticals.
The key advantages of using DRLHF as the core component of our DeepVU framework include the following:
It helps data science teams to overcome potential limitations of hand-designed reward functions, which can be challenging to specify and may only periodically reflect the actual goals of the system.
It helps address the safety concerns associated with conventional Deep Reinforcement Learning techniques by allowing humans to intervene and correct an AI agent’s behaviour when necessary. This is a vital aspect of our emphasis on explainability.
It can lead to more human-friendly and personalised policies that better align with an end user’s goals and preferences.
Digital content provenance involves collecting information about the origin of a digital asset, such as an image, video, audio recording, or document. Specifically, we focus on such details as ownership, authorship, history, citation score, and who controls its distribution.
In the context of AIQ’s large language models, digital content provenance is essential as we train these models on large amounts of text data, which can include sensitive, confidential, or proprietary information.
The measures we apply through AIQ ensure that the datasets used to train these models are trustworthy and appropriate, especially to avoid any legal or ethical issues associated with using these datasets.
Our aim with AIQ is to extend our measures beyond text-based large language models to other data models, such as image and video processing, robotics, and autonomous systems. From our point of view, ensuring the integrity and provenance of any data used to train AI/ML systems is critical for all organisations to maximise their reliability, accountability, trustworthiness and safety.
With AIQ, we help teams build intelligent, responsible and safe human-centric large data models by continuously improving the above measures using our Datastack framework to ensure client data security, privacy, and scalability, delivered safely to the end users through Nebuli’s Nanobot robotic coworker framework.
AIQ is at the heart of Nebuli’s human-centric priority, focusing on transformative human empowerment, not pretending to be human.
Challenging Bias, Misinformation and Harmful Content.
AIQ is our response to the increasing use of AI-powered services and chatbots, which pose dangerous wider economic, social, environmental, cultural, and political outcomes through misinformation or bogus human-like interactions.
AIQ offers a dynamic approach to supporting the growing concerns over data privacy, security and the potential negative impact of AI bias on marginalised communities.
We addressed these concerns by researching and developing solutions for our enterprise customers using state-of-the-art techniques, such as explainability and interpretability methods, Human-in-the-loop (HITL) techniques, Fairness, Accountability, and Transparency (FAT) algorithms, and AI governance and oversight frameworks.
With AIQ, we help organisations explore mitigation strategies against challenges that impede their successful development and deployment of responsible (“ethical”) AI and machine learning algorithms.