Draft:Artificial Intelligence-assisted Generative Search

AI-assisted Generative Search is an emerging online search technology that uses generative artificial intelligence to deliver users more personalised, creative and relevant answers, by combining the natural language processing and content generation capabilities of AI with the prior functionality of search engines.

The Principle
AI-assisted generative search relies on generative AI models such as generative adversarial networks (GANs) and generative pre-trained transformer (GPT) architectures. These models are capable of learning the patterns and structure of input data and then generating new data with similar characteristics.

Examples

 * Komo Search: this AI-powered search engine offers a fast, private, ad-free search experience, and features a virtual assistant (Komo Explore Assistant).
 * ChatGPT: this generative AI tool (developed by OpenAI) was integrated into the Bing search engine by Microsoft in 2023.
 * Google Search Generative AI Experience (SGE): an AI-powered search experience Google that was still being developed in 2023 — for English only — in the USA, Japan and India via the Search Labs program. In addition, SGE is anticipated to benefit from the arrival of Google Gemini (expected by the end of 2023).

Expected advantages and challenges
AI-assisted generative search offers new possibilities and "research experiences", relating to speed, fluidity and relevance of responses. But it still presents challenges and limitations, particularly in relation to the properties of generative AI models (quality and reliability, — possible misinterpretations, hallucinations, biases, or formulation of undesirable opinions — ethical issues or data security) which, according to Google, should gradually be addressed by improving model updates and taking user feedback into account, via postmortem documentation.

Tone, quality and ethics of the search engine response
In 2023, Google expressed their desire for the generative AI in the SGE user experience to have a neutral persona, i.e. that its answers avoid bias or personality. For Google, SGE should always provide objective, neutral answers, confirmed by reliable web results. Unlike ChatGPT, it is neither familiar nor empathetic, and never uses the first person ("I") in its answers. To this end, the AI has been trained to not exhibit a personality — a common phenomenon with Large Language Models (LLMs) — that by virtue of their everyday human language, can easily give the impression of expressing opinions or emotions. Google's Principles of Responsible AI (see the Montreal Declaration for the Responsible Development of Artificial Intelligence) also insist on avoiding bias, inaccuracy, contradiction or undesirable content in AI-generated responses.

Connected Articles

 * Search Engine
 * Artificial Intelligence
 * Generative Artificial Intelligence
 * Chatbot
 * Reinforcement learning
 * Supervised learning
 * Bard (chatbot)
 * Large language model (LLM)
 * Trustworthy AI
 * Natural Language Processing