DFINITY Foundation

ICDevs.org

CryptoIsGood

Anvil

ICPL.app

ysyms

The Accumulators’ Neuron

RawTech Venture

Paul Young

ICPMANUAL

8yeargangDAO

Isaac Valadez

Krzysztof Żelazko

Nicolas.ic

Smaug’s Neuron for Retail Investors

ICLight.io

Taggr Network

Sonic AMM

Rakeoff.io

OpenChat

ELNA AI

GEEKFACTORY

ICP Hub México

ICP Hub Poland

Jerry Banfield

WaterNeuron

ICP Hub Bulgaria

Gold DAO

NeuronPool

LORIMER

B3Pay

ZenithCode

Aviate Labs

Internet Computer Association

CodeGov

Synapse.vote (original)

Synapse.vote (NEW)

Arthur’s Neuron (used to be cycle_dao)

Inactive (Request Removal) - The Fools' Court

$STACK

Always Rejects

John Wiegley

Change the askAI on the forum and website from web2 to web3, KinicVectorDatabase



## Proposal Summary

To strengthen the unique position of ICP, the only company to have achieved AI on Blockchain, we will run askAI on ICP onchain, currently operating on ICP and Dfinity documents and forums. Specifically, we will store RAG, which is a database of Github, documents, forum posts and blog posts, onchain so that it can be referenced when making inferences.Unlike askAI, you can update the database in real time by automating it.
The vector database used is KinicVectorDatabase. This is competitive in terms of cost and communication speed compared to web2, and is promising, so it should be able to secure ICP position in the AI industry and raise awareness.

## Background

Currently, the askAI widget is running on the ICP website and Forum. This was ordered by Dfinity to kappa.ai and is being operated on web2, which is not on-chain for ICP. At the time ask AI was installed (2023/10), it was difficult to achieve on-chain, so I understand that it was necessary to request a web2 company.
Due to technological progress, it is possible for Dfinity to host the RAG of askAI itself.

## Introduction to KinicVectorDatabase

KinicVectorDatabase is introduced in the [forum](https://forum.dfinity.org/t/worlds-largest-web3-vector-database/33309).
* Developed Vamana by Microsoft in Rust
* Ease of testing and debugging
* 70GB upload
* Query call of 350ms, which is not inferior even compared to the cloud
* Low cost
* Utilization of default decentralization

There is no need to worry that on-chainisation will hinder UX. On the contrary, the default decentralization contributes to saving the trouble of hosting and provisioning in multiple regions in the cloud. Also, once you upload it, there are no additional costs, and it is even possible to be cheaper than using the cloud, which requires additional costs. Even if you build askAI using the competitive KinicVectorDatabase, there is a high possibility that it will not worsen the current accuracy or UX.

## Why is this important?

By declaring to the web3 community that ICP will take a firm position in DeAI, and by making it known that it is possible to host full-fledged DeAI, it is also possible to assume that attention will be focused on the replicated AI inference that Dfinity is developing. At the same time, it will also contribute to the growing interest in the DeAI-related ecosystem (Kinic, ELNA, decideAI).

## Benefits

* There is a possibility that AI services that are being developed or are planned to be developed in web2 and web3 will adopt and consume cycles.
* Through the on-chain demonstration of Ask AI, it is possible to discover or re-recognise the hidden benefits and utility of DeAI.
* By building practical services on ICP, it is possible to expect the community to become more active.
* The application of cryptographic technology developed by Dfinity to AI, etc.

## Disadvantages
* There is a possibility of incurring a penalty for changing the contract with kappa.ai, which has already been contracted (this will not be a problem if the migration period is after the contract with kappa.ai ends)


## Conclusion
ICP is planning to expand into DeAI, which is in the same field as web3, but while the characteristics of blockchain ensure redundancy, it also results in a decrease in UX and high costs. This is particularly noticeable in DeAI, which requires more computing power than web3, so there are concerns about its use. For this reason, we should use whatever is available, and fortunately, a competitive VectorDB has emerged from the ICP ecosystem, so there is no reason not to use it.
I propose that we should rebuild askAI using KinicVectorDatabase and use it as a demo to appeal to web2 AI service providers and products that are interested in AI in web3.



On-chain implementation of RAG used by ASK AI