Machine learning is one of the most important innovations that our generation is witness to. Many of the most important thinkers of our time have opined that data is the oil of this century - that is where the money is. And rightly so. Business is fueled by understanding the consumer and reacting quickly and efficiently to the insights. And machine learning plays a big role in this. However, the current ML landscape is broken. It is very centralized where we have large corporations owning vast amounts of data, often not amounting to the complete picture, and in most cases accessed surreptitiously from unsuspecting users of their apps. For machine learning to be truly integrated into our lives and to be leveraged to the full effect, this is not a sustainable model. And that is where Decentralized ML comes into the picture, trying to shake up the game a bit.
Most of the machine learning business is captured today by the big cos such as Google, Facebook etc. How do they do this? Well, they can actively use the data they capture in our engagement with them - our use of google maps or facebook for instance. They keep pushing the boundaries a bit by slipping in the additional permissions that we unwittingly accept when installing any app. Some day when you have an hour to spare, go through all the terms and conditions you have accepted. The chances of this exercise not leaving you a little concerned are very low.
Keeping aside the privacy concerns for a second, what these big cos capture is not the full picture as well. There are many behaviors of ours which our online presence captures, but not all is effectively captured by the big cos. The explanatory video by Decentralized ML puts this succinctly into the graphic of an iceberg.
So, in a nutshell, from the end user point of view, the data is leaving his/her phone or laptop, which is not ideal. And from the point of view of making machine learning about consumer behavior potent, the companies do not get the full picture, which is not ideal. So, yes, the system is broken.
What is Decentralized ML doing to fix it?
As the name suggests, Decentralized ML is setting up a blockchain-based platform which enables decentralized machine learning. So, DML acts as a trust-less bridge between companies wanting data and end users. The machine learning is anonymous and no personal data is extracted out of our devices. So how exactly is it done? The company which wants the customer insights fills up the criteria for the machine learning algorithm and sends it to the DML marketplace. The DML marketplace identifies eligible devices which make a good source set for the machine learning algorithm. Idle processing power from these devices is used to process the algorithm. These algorithms are developed by developers who can sell to the companies. The algorithm then collects anonymized data from all the devices and only the aggregated input of all devices goes to the company. And the users who have consented for this data capture get rewarded for the same.
Let us take an example to understand the novelty and implications of this for all players of the ecosystem. The company which wants the data is happy as the end need is served and at a granularity that previous wide-spread data capture could not afford. The end user is happy as he/she is not losing control over personal data while also getting rewarded for anonymized use of their data. The developers are rewarded for use of their algorithms in a very direct manner encouraging innovation and meritocracy. So, in effect, this is a win-win proposition for the entire ecosystem. And possibly one of the finest use cases of blockchain technology to disrupt a currently centralized setup!!
The winning DML formula
Source for all material: DML website and whitepaper