Crafting safe Generative AI systems

This Post is based on the article "Crafting safe Generative AI systems" Which was Published in the The Hindu on 21st August 2023. 

Introduction

The emergence of Generative AI holds the potential to bring about revolutionary changes across various sectors, potentially contributing trillions of dollars annually to the global economy through Large Language Models (LLMs). As an illustration of this transformative potential, consider the case of the Jugalbandi Chatbot deployed in rural India, driven by ChatGPT.  This innovative chatbot has the capacity to democratise information access, thereby enhancing the economic prospects of numerous individuals. 

However, this promising trajectory is accompanied by legitimate concerns regarding the risks associated with Generative AI. Malicious actors can exploit AI-powered tools to craft highly convincing human-like entities, paving the way for misinformation, disinformation, fraud, hate speech, and other detrimental outcomes.

Risks and Concerns:

• The AI revolution presents challenges, including the creation of artificial entities that are indistinguishable from humans, enabling bad actors to propagate harmful content. 

• This includes misinformation, disinformation, security breaches, fraud, hate speech, and more. 

• Several incidents highlight the risks, such as AI-generated images causing market disruptions and AI-generated content influencing polarised politics and elections. 

• Concerns about safety, misinformation, and election interference are on policymakers' agendas.

Challenges in Regulation:

• Proposed regulatory measures, such as requiring AI bots to self-identify and criminalising fake media, are discussed but are not deemed sufficient to address the challenges. 

• While established companies may comply, bad actors can bypass such rules, undermining their effectiveness. 

• A more conservative assurance paradigm is suggested, where all digital entities are assumed to be AI bots or fraudulent unless proven otherwise.

Identity Assurance Framework:

• To enhance internet safety and integrity, an identity assurance framework is proposed. 

• This framework ensures trust between interacting parties by verifying the authenticity of involved entities, applicable to humans, bots, and businesses. 

• It emphasises openness to various emerging credential types globally, providing privacy protections. 

• Digital wallets play a crucial role in this framework, enabling selective disclosure and safeguarding against surveillance. 

• Multiple countries are working on digital identity initiatives, and India's Aadhaar and the EU's new identity standard are highlighted as potential foundations.

Information Integrity:

• Information integrity, which ensures the authenticity of accessed content, is crucial. 

• Three pillars support credibility: 
• source validation, 
• content integrity, and 
• information validity. 

• Achieving information integrity and identity assurance is challenging, raising debates about privacy, surveillance, censorship, and defining truth. 

• Balancing diverse national values and risk appetites within a larger framework is deemed manageable.

Global Responsibility and Conclusion:

• Leadership is necessary to ensure secure Generative AI deployment. 

• Beyond regulation, a trust framework for global identity assurance and information integrity needs to be engineered. 

• Safeguarding the positive potential of Generative AI while addressing its risks requires a collaborative effort to reimagine safety assurance paradigms.

Important Terms and Facts related to them in the Article :

Jugalbandi Chat bot:

Jugalbandi is a free and open platform that combines the power of Large Language Models such as ChatGPT and Indian language translation models such as those under the Government of India's Bhashini mission to power conversational AI solutions in any domain.

Launched by : Microsoft

Large Language Models : 

• Large language models (LLMs) are deep learning algorithms that can recognise, summarise, translate, predict, and generate content using very large datasets.

• In other words, Large Language Models (LLMs) are a type of artificial intelligence model designed to understand, generate, and manipulate human language. 

• They are trained on vast amounts of text data to learn the underlying patterns and structures of language, enabling them to perform a wide range of natural language processing (NLP) tasks. 

• These models have gained significant attention and popularity due to their ability to generate coherent and contextually relevant text.

Generative AI :

• Generative AI, short for Generative Artificial Intelligence, refers to a subset of artificial intelligence that involves the creation of new content, such as text, images, music, and more. 

• Unlike traditional AI systems that rely on predefined rules or patterns, generative AI systems have the capability to generate novel and creative outputs.

• At the heart of generative AI are neural networks, specifically a type known as generative models. 

• These models are trained on large datasets to learn the underlying patterns and structures of the data. 

• Once trained, they can generate new content that is similar to the examples they were exposed to during training.

Examples: Chat GPT, Google bard 

Bhashini Mission:

• Digital India BHASHINI represents India's initiative in the field of AI-powered language translation. 

• It serves as a platform that utilizes Artificial Intelligence (AI) and Natural Language Processing (NLP) technologies to facilitate language translation. 

• The primary focus of this platform is to provide MSMEs (Medium, Small and Micro Enterprises), startups, and individual innovators with access to AI and NLP resources in the public domain.

• Operated under the National Language Translation Mission, the Bhashini Platform contributes to the mission's goal of enabling more Indians to connect to the internet while ensuring they can access global content in their native languages.