The Future of Artificial Intelligence – Sarath Somana
There hasn’t been a better time in human history than the 21st Century. As we transition into The Age of Artificial Intelligence, it’s worth noticing the direction we are headed and the challenges we face today.
Machine Learning systems often require huge computational power and large datasets to train. They are costly to build and maintain. Additionally, they are prone to the risk of becoming outdated, if not retrained regularly.
Engineers at Uber are trying to solve the computational power problem by distributing the data and model across a cluster of computers.¹ However, this approach is still expensive as GPU is still the costliest computer component.
It’s time to take a step back and listen to one of the greatest thinkers and philosophers Aristotle,
In every systematic inquiry (methodos) where there are first principles, or causes, or elements, knowledge and science result from acquiring knowledge of these.²
We lack power because we are not able maximise our usage.
Nowadays, a lot of phones, computers, IoT devices, carry great computational capacity. We should efficiently leverage these devices to distribute our model and data to the masses. This decentralized way of computing will improve collaboration and democratize the creation of world class AI.
The future Internet will be decentralized. Solve a real problem. If the blockchain is the best tool, then use it.
Applying one of the most promising decentralized technology, blockchain to incentivize collaborators could help build an ecosystem for distributed AI development.
Microsoft researchers Bo Waggoner and Justin D Harris has presented a paper³ at IEEE International Conference on Blockchain which details this approach. Through their framework Decentralized & Collaborative AI on Blockchain³, collaborators can continously train models, build datasets and create a reliable AI system.
The paper demonstrates how to build a secure, trustworthy incentive based system to encourage collaborators to contribute data and power using Ethereum smart contracts. Models are distributed using smart contracts and the code is shared on a blockchain.
Smart contracts are immutable and hence ensure reliability. These permanant records also help delivering rewards for good contributions. Models like decision trees and perceptrons which train effeciently will be good choices to be distributed publicly on smart contracts. Complicated models can be proxied to various machine learning services.
The paper also proposes a prediction market model to reward contributors based on how much their contribution helped model to improve. Currently, this proposal only conforms to small models. However, this can be a starting point to build more collaboration AI and blockchain enthusiasts who can lead us into the future.
Note : All the resources i’ve used to collate this article are mentioned below. Feel free to explore more and be part of the age of AI.
References
[1] : Horovod: Uber’s Open Source Distributed Deep Learning Framework for TensorFlow
[2] : Aristotle’s contribution to First Principle Thinking, Wikipedia.
[3] : Decentralized collaborative AI on Blockchain, Microsoft Research
[4] : Ethereum, An open source platform for decentralized applications
Published at Sun, 25 Aug 2019 19:53:05 +0000
Bitcoin Pic Of The Moment
✅ This image from Marco Verch (trendingtopics) is available under Creative Commons 2.0. Please link to the original photo and the license. 📝 License for use outside of the Creative Commons is available by request.
By trendingtopics on 2019-03-19 08:47:07
