ConsciAI: Embracing Responsible AI Practices

ConsciAI is an intelligent Responsible AI(RAI) audit solution that can comprehensively assess AI applications based on defined laws/guidelines and help agencies certify them for ethical AI development

Comment 3

Inspiration

AI solutions have been around for more than a decade. However with the advent of Generative AI and readily available high compute capabilities, GenAI powered AI applications exploded in almost all business areas. A recent estimate by the McKinsey Global Institute suggests that there are over 1.5 billion AI-enabled devices in use today. This includes everything from smartphones and smart speakers to self-driving cars and industrial robots.

In addition to these specific devices, there are also many AI systems that are embedded in software applications. For example, many online retailers use AI to recommend products to customers, and many banks use AI to detect fraud.

As AI technology continues to develop, it is likely that the number of AI systems in the world will continue to grow. This is because AI is becoming increasingly affordable and accessible, and it is being used in a wider range of applications.

Ethical Concerns in AI Development

Although this showed a lot of promise for businesses to leverage AI quickly to expand and grow, it came with a significant question on the Ethical, responsible and secure way of collecting data, generating loads of information and using it to build AI apps.

To solve this challenge, for past several years, governments and companies have been trying to setup standards to ensure AI applications are developed ethically. However these were more of guidelines rather than actual rules and laws. Recently the EU came up with their first document called the EU AI Act, which documents and details out, the categorization of AI applications. This Act will be reinforced very soon and all AI applications built in EU or to be used in the EU will need to be compliant with this Act.

When the EU Act will be enforced, the AI world will face a herculean task of auditing each AI application to check its compliance with the EU Ai Act. Imagine auditing 2 billion AI applications manually?

ConsciAI: Enabling Ethical Compliance

Introducing ConsciAI - An intelligent auditor which will help categorize AI applications and help certify them for compliance to Responsible AI standards defined by the EU AI act.

What it does

ConsciAI helps governments perform audits to analyze AI applications and provide certificate of compliance. The steps are as follows:

Submission and Analysis Process

  1. User submits Project Details like Project description, Technology stack, Models used and architecture diagram through the ConsciAI website
  2. Based on submitted project details and EU AI Act knowledge base(embeddings), an intelligent assistant dynamically generates relevant questions to ensure compliance of the application to specific risk category. This is done through an intuitive chat interface powered by Gemini pro.
  3. At the end of the conversation, the solution provides a detailed report on risk categorization as well as next steps and recommendations
  4. The solution also provides a downloadable compliance certificate

How we built it

We followed a systematic approach to construct ConsciAI:

Development Process

  1. Worked on the architecture and a wireframe of the proposed solution.
  2. Identified the core components of the solution. These included the platform, data storage, embedding algorithms, base LLMs to be used, prompts to be built, fine tuning requirements and interaction through an intuitive User Interface
  3. All components were independently built through required APIs/platforms.
  4. Integrated all the components to complete the flow.
  5. Performed thorough testing using various datasets to ensure correctness and accuracy

Challenges we ran into

We faced several challenges during the development process:

Obstacles Encountered

  1. Defining problem statement and scoping of the solution
  2. Data Acquisition - Since EU Act is quite recent, there are very few resources available on the same.
  3. Learning various platforms, algorithms and methodologies in a short period of time

Accomplishments that we're proud of

Despite the challenges, we achieved several milestones:

Achievements

  1. Gained expertise on GenAI platforms and components
  2. Gained good understanding of responsible AI terms and intricacies.
  3. We started looking at data and application development through a different lens, taking ethical and responsible AI impact into consideration

What we learned

Building ConsciAI provided us with invaluable insights:

Key Learnings

  1. Google Vertex AI as a platform for fine tuning and hosting LLMs, embedding data for RAG (Vertex AI Studio)
  2. Google Cloud Platform for storage (CloudSql, Buckets), Cloud Functions, Cloud Run
  3. Building the Retrieval Augmented Architecture
  4. Writing good quality prompts for optimum usage of LLMs
  5. Data Acquisition and ingestion using correct embeddings
  6. Decent understanding of Responsible AI Practices and the EU AI Act!

What's next for ConsciAI: Embracing Responsible AI Practices

We have ambitious plans for the future:

Future Plans

  1. Adding support for multiple geographical regions in addition to the EU
  2. Integrating with MLOps/LLMOps platforms to reduce RAI debt

Built With

Share this project:

Updates