The Request For AI Portal

Enabling the crowdsourcing of AI via the SingularityNET Platform

Ben Goertzel
SingularityNET
9 min readDec 20, 2019

--

What should someone do when they need a very specific type of AI service that is not available on the market? At this stage in the development of the global AI ecosystem, the diversity of need far outpaces the diversity of supply, and most of the AI functions that would really help businesses, nonprofits or researchers are not yet available either as products or as APIs to access.

Where does this leave the customer with unmet AI needs? They can try to shoehorn their requirements into the straightjacket of available products and services. But often this will squelch their creativity and their ability to adapt — and often it isn’t even feasible.

They can start their own AI R&D team or hire expert consultants, but these can be costly and difficult propositions, and in the current climate where AI experts are in extremely high demand, even some of the larger and wealthier corporations are having trouble acquiring the AI teams they need.

Many organizations large and small plan to utilize AI but they either lack a strategy, are not sufficiently familiar with the concepts to get started, or lack straightforward access to the right talent due to geographical or other practical reasons.

An ironic and surprisingly common variation on this situation occurs when multiple parties have a need for the same nonexistent AI tool, but none has the wherewithal to get it done on their own. There is currently no well known, easy to use mechanism for the parties in this sort of situation to join forces and push for the creation of AI tools that will meet their needs. While multiple parties can theoretically pool their resources to develop the same AI solutions or services right now, they lack an efficient and trustless mechanism to do so.

These are the broad motivations underlying the Request For AI (RFAI) portal, the latest addition to the SingularityNET decentralized, blockchain-based AI platform. The RFAI portal will allow anyone to create an AI request and incentivize its fulfilment with AGI tokens.

Put concisely, the SingularityNET RFAI portal comprises a trustless, transparent and commission-free crowdsourcing mechanism for AI-related requests.

Who Can Create a Request?

Anyone with an AI needs that they can articulate in a clear and straightforward way can create a request for an AI service using the RFAI portal. No AI knowledge is needed to create a Request for AI — only knowledge of what the requested AI is supposed to do, for instance, what inputs it will be expected to receive and what sorts of outputs it will be expected to give.

Each submitted request will then be approved by the SingularityNET Foundation (which will ensure that it meets a few minimum requirements, e.g. regarding comprehensibility and legal compliance) and made visible to the public.

Why Fulfill a Request?

Requests will be “backed” by AGI tokens to incentivize their fulfilment.

Since the requests will be publicly visible, any party interested in the fulfilment of a specific request will be able to back the request with their AGI tokens.

The proposer of an RFAI must then back their request. It makes sense for someone to propose an RFAI and offer what they suspect to be a fraction of the total amount of tokens needed to incentivize developers to solve the problem. This creates a clear opportunity for other parties also desiring the creation of the same AI to add some tokens to the total reward pool for the RFAI.

Who Gets To Fulfill a Request?

Anyone on the planet (or beyond the planet if they can get on Earth Internet where SingularityNET is currently running) can submit a SingularityNET service as a preferred response to a Request for AI that’s in the RFAI portal.

The AGI tokens submitted by backers will be placed in an escrow account, ensuring that the developer whose submission is chosen as the winner has the certainty of being rewarded.

How is the Winning Submission Chosen?

The initial RFAI portal embodies a very simple logic for choosing winners. Each RFAI contains a set of minimum criteria that a submission must satisfy in order to be validated as fulfilling the request. The winning submission is chosen by the following voting structure.

  1. Foundation can vote on the submissions
  2. Users backing the service request can vote for any submission
  3. Only the users who are backing a service request can vote for the submission
  4. Validation of the solution will be performed offline.
  5. Backed tokens are distributed to all submissions with at least one vote

After a submission is made, they are voted on by the backers and the foundation. The winning submission is then accepted and the reward tokens can be claimed.

The potential to include humans in the loop notwithstanding, RFAIs with insufficiently clear evaluation metrics will be rejected in the review process. The intention is not that RFAIs should have highly subjective evaluation criteria, like “Make an AI music composer that creates some great new jazz.” Rather, the intention is that RFAIs should have reasonably precise evaluation criteria, but that human evaluation can be specified if needed to rule out submissions that clearly violate the spirit of the criteria while somehow perversely obeying their letter.

Why Isn’t Judging Always Fully Automated?

In the initial rough draft of the RFAI portal, the inclusion of humans in the process of validating a submission was not allowed — this is supposed to be a Society of AI Minds, right?

But after exploring a number of practical RFAIs, we realized that in many cases it’s actually quite difficult to create a fully automated evaluation script that rules out any plausible possibility of an entrant “cheating” by gaming the specifics of the script. For example, it’s not always possible to find or provide a pristine set of test data to use, which is relevant to the problem at hand yet is something no entrant would feasibly have access to.

Consider for example a hypothetical RFAI to create an AI model that can predict a country’s GDP based on a list of specific non-economic factors (like its geography, the religious beliefs and education level of its population, etc.). One thing the creator of this RFAI almost surely doesn’t want to reward is an AI that, given the non-economic factors associated with a given country, uses some wizardry to guess which specific country it is, and then looks up the GDP of that country from a list. The “list lookup” wouldn’t need to be explicit in an AI’s software code, it could be effectively hidden in the coefficients of some regression function or similar.

But there’s no straightforward way to create a test set that wouldn’t be susceptible to this sort of cheating because there is only a limited number of countries on the planet and the relevant information about them is accessible to all entrants. Making up realistic synthetic countries is a project unto itself and tough to do without making highly questionable assumptions.

One can cook up increasingly fancy automated ways to rule out cheating of this nature, in this example and others, but as we dug into the matter, we realized that the need to apply effort and creativity on making relatively gaming-immune evaluation criteria was going to significantly reduce the number of RFAIs submitted and also slant the types of RFAIs submitted (toward those for which it’s easier to create simple, fully automated evaluation scripts). So we chose to include a human-in-the-loop option, and once we did, we realized that it dramatically increased the scope of projects one could realistically consider as subjects for RFAIs.

Won’t You Need to Launch the AI Publisher Portal too, to Give the RFAI Portal Widespread Use?

Aha — you caught us!

Yes, the RFAI portal is actually only half of the equation. RFAIs make it easy to request new AI services on the platform. The AI Publisher interface will make it easy to launch new AI services on the platform and have these new AI services listed on the main SingularityNET marketplace interface.

The AI Publisher portal is targeted for release in early 2020 and, with this alongside the RFAI portal, we will have a reasonably complete framework for the request and creation of AI services on the SingularityNET platform.

Using the recently launched PayPal integration, it is possible for end-users to pay for services offered by AI agents on SingularityNET platform using fiat currency, even if they don’t own cryptocurrency or use crypto wallets and related technologies. When the SingularityNET fiat/crypto gateway is complete, it will also be possible for providers of AI services (with certain restrictions) to get paid directly in fiat currency.

Together, all these tools will make SingularityNET a full-featured and easy-to-use platform for creation, deployment, delivery and interaction of AI agents delivering a variety of AI services.

We are launching the RFAI portal first so that those with relevant AI problems to be solved can post them now, and interested developers can start thinking about them — and then when the AI Publisher is released, targeted for early next year, they can begin deploying and submitting their solutions.

RFAI Portal: Adding Multiple Dimensions of Value to the Network

With the introduction of the RFAI portal, the SingularityNET platform enters the freelancing market, albeit in a somewhat unusual manner. While there exist freelance platforms that serve as matchmakers between sellers and buyers — the SingularityNET platform will have several unique advantages: crowdsourcing, no commissions, focus on AI, advanced reputation mechanism, greater inclusivity, and more.

As the services requested via the RFAI portal will be added to the decentralized network, the RFAI portal will act as an important new source of AI services, increasing the diversity of the network. As more and more AI requests for services are created, the offerings and diversity of the AI services in the decentralized network will increase.

As the population of AI agents on the network increases, there will be more potential and motivation for RFAI authors to request AIs that depend for their operation on other AIs running in the network. The RFAI mechanism will then serve as a means for helping transform the early SingularityNET network into a more richly cross-connected Society of AI Minds with its own emergent intelligence.

From a tokenomic perspective, the RFAI portal is one more small step in manifesting the value of the native AGI token that is the lifeblood of the platform. The AGI token is designed to be utilized in a variety of ways, some of which are currently realized in the platform beta version, some not yet: paying for AI services, staking AI services, for settlements between AI Agents, and for democratic participation. The RFAI portal utilizes the AGI token as incentives for the fulfilment of AI requests. This backing of requests on the RFAI portal is a valuable entrant on the list of practical utilities of the AGI token.

We have received numerous emails and messages from our community members in which they’ve informed of us of their willingness to fund the creation of AI services via their tokens. The ability to create AI requests will empower our community members to encourage the addition to the ecosystem of AI services that they believe will prove beneficial as a whole.

We have also been in communication with a variety of business large and small who have expressed their enthusiasm for being able to connect directly with AI talent, without any middlemen, and request customized AI solutions or services. The RFAI portal enables precisely this.

The RFAI portal itself is just one ingredient along the path from the beta version of SingularityNET to the 1.0 version, but it’s an important step. With the rollout of the AI Publisher portal early next year, we will then have a complete, decentralized, blockchain-based platform for the request and delivery of AI services of every type.

How Can You Get Involved?

The SingularityNET team is always looking to hire those who are passionate about open-source, decentralized control and believe that no single entity should dominate AI as it becomes more generally intelligent and capable. If you would like to join us, please click here and have a look at the open job positions.

We are proud of our developers and researchers that are actively publishing their research for the benefit of the community; you can read the research here.

For any additional information, please refer to our roadmaps and subscribe to our newsletter to stay informed about all of our developments.

--

--