The Human Protocol redefines the market for data marking and crowdsourcing to achieve better accuracy

artificial intelligence (AI) can only fulfill its purpose if it is trained on high quality data. The success of a AI algorithm depends largely on the quality and quantity of the training data used. Accordingly, it should not surprise The creation of a AI project is devoted to the optimization of training data

Most AI projects are faced with the difficult task of collecting or purchasing high-quality data. There are several cases in which projects often end with low quality data or marked data. While several data identification services have been created in recent years that face the challenge to a certain extent, they have their own problems. The main reasons for marked data of low quality are, for example, the people, processes or technologies used for labeling.

but what exactly are labeled data?

data marking: the fuel for AI models

In connection with AI, marked data relate to data that are "marked or commented" so that a machine learning model can predict the desired result. In general, the entire data identification process usually includes several steps, such as data comment, classification, tagging, moderation and processing.

There are several approaches to data labeling that can either be used independently of one another or in combination. This includes internal data marking, outsourcing, crowdsourcing and the use of machines (whereby data is labeled using machine learning algorithms).

Depending on the complexity of the problem, AI projects often use extensive labeling processes in order to convert not labeled data into the training data that you need to teach your AI models, which can be identified to create the desired edition.

of the many available methods is crowdsourcing, in which a third-party platform is used to access large quantities of human workers at the same time, one of the most frequently used tactics of projects to identify data. In recent years, among other things, several platforms such as Amazon Mturk, Appen Meeta Dash, Label Box and Tagog have emerged as some of the most promising platforms for crowdsourcing human workers for data marking.

However, several projects have expressed concerns about the data quality of crowdsourcing platforms. For example, take the data quality problem with Amazon Mechanical Turk (Mturk), which goes back to 2018. Many data researchers suspect that data using bots in addition to half and fully automatic code or scripts were identified in order to support people in the reaction quickly to certain data sets.

Part of the problem was attributed to users from different locations that used VPNS to participate in surveys and questionnaires that were not suitable for their area scheme. Since crowdsourcing platforms pay human workers appropriately for the completion of tasks, users often participate in double activities to achieve more income. For example, a number of users from different countries can use VPN to participate in a data identification program that requires certain answers from American residents. This leads to inferior and nonsensical answers, which in turn lowers data quality.

If data of low quality are submitted, this raises serious questions about the existing quality assurance process. Since most of the existing crowdsourcing platforms for data identification are strongly centralized, it is almost impossible to assess the quality and workflow. All of these problems, paired with the comet-like growth of blockchain technology, have paved the way for decentralized and approval-free crowdsourcing solutions.

Here the Human Protocol presents a new new approach to data labeling by creating an infrastructure that supports the permissionless labor markets that also provide human workers with work and give organizations access to workers-all without central intermediaries.

facilitating of approval-free job markets

human protocol