Stay up to date with the HUMAN project through our bi-monthly newsletter.
Vast quantities of targeted data are required to create functional AI products. While there is plenty of pre-labeled data available, the industry has suffered from the major limitations in data-labeling services and in the types of AI training available at scale. To create new, relevant AI products, practitioners cannot simply use historical data, or they risk exacerbating bias; instead, they need fresh, targeted data sets and active feedback provided by data services.
The first job markets and Exchanges on HUMAN Protocol will address this issue, by giving data scientists access to data labelers (including feedback-style learning), and data labelers the tools they need to get the work done.
A HUMAN job market is simply the result of Requesters publishing work requirements with prices, and available and appropriate Workers on the other end of the system. The job market represents a point where buyers and sellers are matched.
Initially, there will be three verticals of job markets: video, image, and text annotation. The Protocol can take a job that spans several of these markets, decompose it, and send the work to the various Exchanges. It can also crosscheck the same data between job markets to check for quality assurance. One dog annotated in an image on CVAT can be passed to a different user with a different toolkit to see if the results are the same.
An Exchange is simply the application which allows Workers to fulfil jobs. Examples include Intel CVAT, and INCEpTION.
The HUMAN Protocol Foundation defines the tools that are used so that the Protocol has end-to-end control on the quality of a job. It also means that there can be more certainty over job output; similar jobs, when fulfilled through the same Exchange, will have a more deterministic outcome. The Foundation has selected what it believes are the best tools on the market for each kind of job; each Exchange provides Workers with everything they need to do the work.
Requesters looking to access the data labeling job markets could include: BPO (business process outsourcing) of large machine learning practitioners who have massive quantities of ML specific data to label, university researchers who want a data set annotated, feedback given to a model, or startups in the AI industry requiring very specific data for a product they are developing.
The Requesters predefine the price for work in the smart bounty. Workers can see, through their HUMAN Dashboard, the available work and the bounties they will be paid, and choose which tasks to fulfil.
It is a market, not a marketplace, as there is no direct negotiation; Requesters and Workers do not need to communicate. However, it is an open market of iteration, of trial, assess, and adjust. Requesters who offer the best price for work will likely have their work completed fastest; the market is designed so that workers who are found to offer the most time, efficiency, and skill to the project can build their reputation, and become ‘front of the queue’ when it comes to job opportunities. More on job prioritisation below.
Equally, if Requesters set prices too low, they would need to adjust their approach if they are to benefit from the Protocol. Such a system promotes and rewards excellence. That is the essence of a HUMAN job market; it is designed to be open, fair, and transparent.
Transparency is at the heart of any fair market. For decentralized parties to work together, there needs to be trust in the system: trust that collusion is not a problem; trust that opportunities are being fairly represented across the Protocol; trust that funds are being fairly distributed.
The system is designed so that by logging in, Requesters and Workers can see the current prices at which work is complete. Because the Protocol is blockchain compatible, with more and more work moving on-chain, anyone is free to see where funds move. Furthermore, the settled amounts are viewable in each on-chain orderbook.
Requesters must prefund the smart bounty with HMT, which is then held in escrow, so there is no chance for them to steal work. Equally, Workers’s responses are subject to evaluation by a decentralized network of oracles who determine whether or not the work has met the required standard. All of this is predefined and automated; in a decentralized network, there is little room for subjectivity. The software is designed to automate, evaluate, and ensure fairness in Protocol interactions.
The Foundation wants anyone to be able to request work, and anyone to be able to complete work through the Protocol.
One of the basic principles of open-market economics is that market efficiency increases with market size. HUMAN job markets are no different. The more Requesters and Workers partaking in the Protocol, the greater the possibilities for price and work optimisation, and meritocratic outcomes.
The HUMAN Protocol Foundation has decided to create a system of reputation that incentivises best practices and priority for job allocation. Reputation Oracles, which evaluate work quality, are responsible for assigning a reputation score for a given Worker on a given job.
A synthesis of wallet balance and total value of past transactions would help determine priority.
Workers: The current HMT balance of the Worker's wallet address helps determine their priority level for tasks served to them from the Exchange Order Book. Higher balances of HMT would have higher priority as the Exchange selects which job tasks to distribute.
Example: If Worker A has 400 HMT in their wallet and worker B has 50 HMT, worker A would get first pick of jobs if A and B are otherwise similar.
Job Requests: On the other side of the transactions, the Foundation proposes to prioritize equal bids from Requesters in the Order Book on the basis of balance of HMT held by the Requester’s address.
Example: If two requests offer three HMT for a job, and Requester A’s wallet has 2000 HMT while Requester B’s wallet holds 10,000 HMT, Requester B’s job would take priority.
While the Foundation has outlined three Exchanges that will operate on the Protocol, the plan is for there to be many more. Metcalfe’s law states that the utility of a network increases exponentially with the linear increase in users.
If one views these Exchanges and job markets as networks where information is traded, the more participants there are, the more possibilities there are for communication and collaboration. The more collaboration, the more increasingly complex tasks that can, hopefully, be brought to the Protocol, thus beginning the process of applying AI to many more issues than the current bottlenecks in data practices permit.
For the latest updates on HUMAN Protocol, follow us on Twitter or join our community Telegram channel. Alternatively, to enquire about integrations, usage, or to learn more about how HUMAN Protocol supports machine-learning technologies, get in contact with the HUMAN team.
The HUMAN Protocol Foundation makes no representation, warranty, or undertaking, express or implied, as to the accuracy, reliability, completeness, or reasonableness of the information contained here. Any assumptions, opinions, and estimations expressed constitute the HUMAN Protocol Foundation’s judgment as of the time of publishing and are subject to change without notice. Any projection contained within the information presented here is based on a number of assumptions, and there can be no guarantee that any projected outcomes will be achieved.