Multiple Coupled Network Effects and Inter-AI Synergy in the SingularityNET
If SingularityNET is to experience the radical exponential growth that its name suggests, multiple network effects will need to occur and to work together.
One is simply the “standard” network effect native to any first-mover “app store” type framework. If enough AI Agents are created in the SingularityNET, and enough users start leveraging the SingularityNET for services, then neither AI developers nor customers will have a rational reason to shift away from SingularityNET to some different platform. From the Agent author and owner’s point of view, most of the customers will be at SingularityNET. From the customer’s point of view, SingularityNET is where most of the Agents providing valuable services are.
Then there is an additional network effect that is somewhat special to the SingularityNET. There is, in many cases, additional synergetic value obtained by putting two AI Agents together, in a way that allows them to closely interoperate in carrying out functions and serving customer needs. Very roughly, the more Agents are in the network, the more such synergetic value one can expect. So that, from an Agent developer’s point of view, removing an Agent from a mature SingularityNET and placing it in some other network would not only take the Agent away from a large market, it would take it away from its collaborating Agents that allow it to perform with maximal intelligence, or that allow it to participate in federations carrying out different sorts of tasks than it can carry out on its own.
Based only on the standard network effects, SingularityNET would be purely a market and platform. It would enable customers to find lower-cost AI services, and would foster more advance AI R&D via providing a better way for AI developers to monetize their inventions. But it would not intrinsically constitute an advancement of AI.
The additional inter-Agent-synergy network effect, however, brings SingularityNET to a different level, and turns it into a medium for the evolution and emergence of new forms of AI with unprecedented levels of intelligence.
Some instances of inter-AI-agent synergy will be quite simple and practical, others will be more profound.
As a relatively simple example, consider the relationship between an Agent carrying out text summarization, and an Agent carrying out relationship extraction from text (e.g. recognizing that the sentence “Bob ate a huge steak” involves an actor Bob, carrying out the action Eating, relative to the object Steak). The text summarization agent needs to prioritize and decide which sentences in each document it’s looking at are the most important ones (to include in the summary, in whole or in part). In order to do this, one approach is to extract relationships from the sentences in the document, and then prioritize the sentences that (among other factors) contain the most central relationships in the network of relationships expressed in the document. On the other hand, if a relationship extraction Agent is given a huge number of documents to extract relationships from, and the customer wants it to do this for a low price, then its best bet may be to ask a text summarization Agent to summarize each document first, and then extract the relationships from the summarized document. This will likely provide OK results, and will be much faster and cheaper than extracting relationships from every sentence in every document. In this case, one sees, there are two AI Agents that can benefit from each others’ services, each enabling the other to do its work better.
If the text summarization Agent wanted to operate outside the SingularityNET, it would not do its job as well unless it either A) convinced the relationship extraction Agent to operate outside the SingularityNET as well, or B) ran partially inside and partially outside the SingularityNET, which will be possible but more difficult to script and likely also slower. To the extent that this sort of cooperation exists (and it’s precisely this kind of cooperation that SingularityNET is designed to facilitate), there is a strong reason for Agents to remain in the SingularityNET once they’re there, and to recruit additional supporting Agents to join the network as well.
As a subtler example, consider the interaction between an Agent that does logical reasoning using some powerful logic framework such as predicate logic, and an Agent that does some reasonably general sort of “frequent or surprising pattern mining” in trees or graphs. Now, the logic-engine Agent may require the services of some pattern-mining Agent, in order to mine patterns in the database of its prior inference, and thus learn patterns regarding which kind of inference works best in which contexts. On the other hand, the pattern-mining Agent may require the services of some logic-engine Agent to generalize the simplistic patterns it mines in data via statistical means, and transform these into more abstract patterns that can generalize to different contexts better. If a certain logic-engine Agent and a certain pattern-mining Agent find they work especially well together, then they will ongoingly outsource work to each other, and they will essentially form a two-node “federation,” with a combined intelligence (at reasoning and/or pattern mining) greater than the sum of the individual intelligences of the parts.
This particular example happens to be drawn from the OpenCog AI software framework — OpenCog’s PLN (Probabilistic Logic Networks) logic engine and OpenCog’s Pattern Miner interact in exactly this way. OpenCog is designed to work toward general intelligence, and is based on a principle of “cognitive synergy,” according to which multiple AI components work closely together in a way that enables them to achieve “whole is greater than the parts” type emergent intelligence. Placing OpenCog AI tools in separate AI Agents in the SingularityNET, in essence allows a (possibly quite large) federation of Opencog AI Agents in the SingularityNET to act as a decentralized, distributed OpenCog system (or a set of overlapping “OpenCog systems”). However, this sort of system also goes well beyond the OpenCog design as such, because it enables these OpenCog AI Agents to also interact with multiple different AI Agents that are not part of OpenCog. OpenCog can be considered a sort of prototype for the deeper sorts of cognitive-synergetic interactions that will arise in a mature SingularityNET.
These subtler sorts of network effects, acting together with the more standard sort of network effects involving coupled increase in the customer and user basis, are what will allow SingularityNET to live up to its ambitious name and grow exponentially in intelligence, user base and financial scale.
