Artificial intelligence shapes gaming experiences. It allows for computer players, generates unique dialogue, and creates beautiful graphics. As AI grows more powerful, it will be critical for the newest techniques and models to be available for public access, review and alteration. That is why Giza is putting machine learning on the blockchain.
Today’s AI is fragmented by closed development environments. A small group of powerful companies develop their own tools that perform specific tasks. There is no standard method to combine or repurpose different models. This is where the blockchain may prove valuable.
Giza will allow machine learning developers to publish their models onto the blockchain cheaply and quickly. It is built on StarkNet, an ultra-fast and cheap layer two built on Ethereum. Once a developer has constructed a model on some dataset off-chain, they can use Giza to convert the model’s tensors and/or operators into Cairo, StarkNet’s bespoke programming language. From there, the model can be deployed on StarkNet, making it freely available to anyone with internet.
Why is putting artificial intelligence on chain a good idea?
Once a developer trains a model, they need to convert their model to production, which can be slow and expensive. Giza is designed to convert any model, regardless of framework, into a Cairo smart contract. This enables interoperability from across any framework in a straightforward process.
Another benefit is the blockchain becomes the infrastructure. Developers don’t need to worry about server uptimes, load, or architecture, because everything is natively managed by StarkNet. Currently, complex neural net algorithms are too computationally intensive to be supported by StarkNet. Once layer three solutions are built atop StarkNet, it will become possible to create private, ultra-scalable, complex artificial intelligence algorithms that are supported by Ethereum.
Finally, StarkNet can store and record important metadata publicly. The Cairo smart contract can contain information about traces, performance, and quality, meaning developers can hand pick the best models for their needs.
What Will We Do With These Models?
An exciting entry point for on-chain machine learning is allowing developers to permissionlessly improve existing models. Competitions and bounties can be created to incentivize skilled AI engineers to collaborate and iterate on present models. Additionally, models can become monetized, so open source builders can be trained for their contributions.
Giza faces many challenges en route to challenging the titans of AI. Currently, their models are slow to execute, due to the speed of StarkNet (which is an alpha, with network speed being a focus of ongoing development). In order to improve it, optimizations are necessary. But once models can be executed in comparable time to their off-chain counterparts, companies may seriously consider posting them on-chain for public use.
On-chain artificial intelligence is still in its earliest phase. But if it can realize the potential of on-chain, open-source computation, it could open the floodgates for more fun, immersive gaming experiences.
If you interested in machine learning and AI on the blockchain, perhaps you are new to the topic or experienced in the field and looking to explore the possibilities of this frontier tech, join the Matchbox discord. Say hello to franalgaba.eth#1745 of Giza or tarrence#4186 of Cartridge who are both exploring this technology.
You can hear them in conversation about on-chain ML and AI, here.
Authors: