Should AI models be like Swiss Army knives, versatile and handy in a variety of scenarios? Or do we prefer them as precision tools, finely tuned for specific tasks? In the world of artificial intelligence, and natural language processing specifically, this is an ongoing debate. The question boils down to whether models trained for specific tasks are more effective at these tasks than general models. Task-specific models: specialization and customization In my last blog post , we looked at the rise of personalized LLMs, customized for specific users. Personalized LLMs can be seen as an extreme form of task-specific model. Fans of task-specific models stress that these kinds of models are better suited for tasks involving confidential or proprietary data. This is obviously true. But some people also believe that specialized models necessarily perform better in their specific domains. It may sound logical, but the ans...
Artificial Intelligence (AI) and blockchain technology have emerged as two groundbreaking forces revolutionizing their respective domains.
AI has taken leaps in transforming industries with its ability to automate tasks, extract insights from data, and make intelligent decisions. Meanwhile, blockchain has disrupted traditional systems by enabling decentralized and transparent transactions, ensuring trust, and establishing immutable records.
The convergence of these two cutting-edge technologies opens up new possibilities extending beyond their individual capacities.
This post will explore one of the intersections of AI and blockchain, namely ways in which blockchain is being harnessed to develop AI itself.
Specifically, we will look at a platform called SingularityNET and some of the exciting projects currently in development there.
Lastly, we will explore why the founder of SingularityNET, Dr. Ben Goertzel, feels decentralized governance is the best way forward not only for this platform, but for developing AI as a whole – all the way into Superintelligence and beyond. Like many experts, Goertzel sees the development of a Superintelligent AI as a matter of time. Decentralization, he argues, is a way to keep this Superintelligence beneficial and on our side.
Decentralized Artificial Intelligence
Decentralized Artificial Intelligence (DAI, not to be confused with the stablecoin of the same name) refers to the concept of using decentralized networks and technologies to develop and roll out AI applications. Combining the principles of decentralized computing and AI algorithms, DAI opens up opportunities for anyone who wishes to contribute.
Such a decentralized network fosters worldwide collaboration and accelerates the development and adoption of AI technologies. It also reduces the costs associated with individualized AI development.
Often, although not always, such platforms use blockchain technology to enable a tokenized ownership and governance system. The aim is transparency and fairness.
But optimal governance structures remain a challenge for DAI platforms and projects. Achieving effective and fair governance in decentralized systems is complex and requires careful design. DAI companies generally want their governance mechanisms to ensure transparency, inclusivity, and decentralized decision-making. These mechanisms should enable stakeholders to participate in the decision-making process and have a say in shaping the development and deployment of the AI technologies.
However, finding the right balance between decentralization, efficiency, and accountability is an ongoing challenge.
Various projects and initiatives are therefore exploring different governance models, including decentralized autonomous organizations (DAOs), to address these challenges and advance the field of DAI governance.
Despite the governance challenges, the concept of DAI holds promise for advancing AI technologies and democratizing their development and deployment. By leveraging the power of decentralized networks, DAI has the potential to make AI more accessible and collaborative.
The first and arguably most well-known DAI platform is called SingularityNET.
As an example of the work being done with the combination of AI and blockchain, let’s now take a closer look at SingularityNET. We will also take a quick look at another promising platform called NuNet (which aims to offer its AI services to the SingularityNET platform), and a new protocol called HyperCycle.
Platforms and protocols
SingularityNET
SingularityNET is a blockchain-powered platform for facilitating DAI. It was established in 2017 by the cognitive scientist and AI researcher Dr. Ben Goertzel. Goertzel’s vision is that a network of diverse AIs working together might lead to the emergence of a collective intelligence akin to a global brain, perhaps even Artificial General Intelligence (i.e. AI with levels of intelligence and understanding comparable to that of a human) and then Superintelligence (which he envisions will be, if it evolves in this way, beneficial for humanity).
The ultimate aim for the platform is to be completely decentralized in terms of ownership.
Goertzel sees centralized control of AI by large governments and corporations as concerning. He is worried about their potential monopolization of data, computing power, and AI expertise. Goertzel doesn’t see such a scenario as optimal for the development of his envisioned “global brain”. This is because governments and corporations, he says, have very specific narrow self-centred goals and are not necessarily focused on the best outcome for all of humanity.
The vision for SingularityNET, therefore, is to create beneficial Artificial General Intelligence (AGI) and Superintelligence, based on Goertzel’s premise that decentralization would produce a better outcome.
Although AGI still eludes us, SingularityNET is currently a functioning decentralized marketplace where individuals can introduce their AI products. Companies looking for AI services can submit requests to the network, and any AI capable of fulfilling the request can make an offer. The platform is also designed to enable AIs to outsource work, cooperate, and share data.
As a blockchain-based platform, SingularityNET is designed to work with smart contracts.
Smart contracts are self-executing contracts with the terms of the agreement directly written into code. In the context of SingularityNET, the smart contracts are used to facilitate the interactions and transactions between AI service providers and consumers. These smart contracts define the terms, conditions, and payments for accessing and using services on the platform.
For example, when a user wants to request an AI service, they interact with the SingularityNET platform through these smart contracts. The smart contract handles the negotiation, service agreement, and payment between the user and the service provider. Once the agreement is reached and the transaction is completed, the smart contract ensures that the service is delivered and the payment is made automatically.
This is how SingularityNET can enable secure, transparent, and automated interactions between AI agents and users, removing the need for intermediaries.
NuNet
NuNet, founded in 2019, is constructing a decentralized global computing network that will enable individuals to share and monetize their idle computing resources. This aims to leverage the vast untapped potential of private computers, smartphones, and smart devices that collectively possess computing power surpassing even the most advanced supercomputers. The aim is to build and research AI and other software products.
NuNet also recognizes the untapped opportunity of crowdsourcing the creation and management of decentralized datasets.
It also uses blockchain technology as the underlying infrastructure for its operations.
NuNet aims to offer its AI services to the SingularityNET platform. This integration will enable SingularityNET users to access and use the AI capabilities provided by NuNet.
HyperCycle
SingularityNET has also partnered with the blockchain platform Cardano to introduce a new blockchain initiative with a specific focus on AI. This protocol is called HyperCycle.
The HyperCycle concept recognizes that AI models and algorithms require continuous improvement, adaptation, and optimization over time. It therefore aims to establish a feedback loop where AI models can evolve based on real-world usage, user feedback, and advances in AI research.
HyperCycle is exploring different approaches for compensating AI developers based on the value their models provide to users and the broader ecosystem. The aim is to incentivize continuous innovation and improvement in AI services.
Architectures
OpenCog Hyperon
OpenCog Hyperon has been a focus of development within SingularityNET. OpenCog is an open-source software project aimed at building AGI systems, while OpenCog Hyperon is a specific architecture within the OpenCog project. It represents the design and organization of the OpenCog AGI system.
OpenCog Hyperon aims to create an advanced AI system capable of reasoning, learning, and adapting in a decentralized manner. It involves a distributed knowledge metagraph (more on this below). It also uses evolutionary neural nets to help the system improve itself over time.
TrueAGI, a project being spun off from SingularityNET, aims to bring the OpenCog Hyperon system to enterprises and companies and make OpenCog Hyperon available to a wider audience.
What is a “distributed knowledge metagraph”?
A metagraph, also known as a metagraph database, is a concept used in the field of AI and knowledge representation. It refers to a graph-based structure that represents knowledge and information in a highly interconnected and flexible manner.
In a metagraph, nodes represent entities or concepts, and edges represent relationships or connections between them. The nodes and edges can have associated attributes or properties that provide additional context or information.
A metagraph allows for the modeling of rich semantic networks, where entities can have multiple relationships with each other.
Metagraphs are particularly useful for organizing and representing knowledge in domains where there is a need to capture and reason about complex interdependencies. They enable the exploration and traversal of interconnected concepts and relationships, supporting sophisticated queries and analysis.
Zarqa
Another architecture developed by SingularityNET is Zarqa. Zarqa focuses on building decentralized large language models. Initially, Zarqa is concentrating on transformer neural nets, which have gained significant attention in recent times.
ChatGPT, for example, is powered by a transformer neural network architecture. Transformers have been widely used in natural language processing tasks and have proven to be highly effective in tasks like text generation, machine translation, and question-answering. The transformer architecture allows ChatGPT to process and generate text by capturing contextual relationships between words and understanding the broader context of the conversation.
However, the long-term plan for Zarqa is to incorporate more elements from Hyperon, such as logical and evolutionary algorithms.The aim of this is to enhance the factuality and creativity of the language models and to surpass the limitations of mere data replication and generic responses seen in systems like ChatGPT.
This approach directly addresses the issues of hallucinatory and banal responses commonly associated with existing language models. The envisioned outcome is an AI system that offers more insightful, contextually relevant, and imaginative interactions. The integration of logical reasoning and evolutionary algorithms holds immense promise in advancing the capabilities of conversational AI and unlocking its practical applications across diverse domains.
By combining these and similar architectures and technologies, SingularityNET envisions the creation of various applications with enhanced capabilities and functionalities.
The vision of decentralized governance
Decentralization aligns with Goertzel's vision of creating a more inclusive, transparent, and equitable AI ecosystem, empowering individuals and communities while mitigating risks associated with centralized control over AI technologies. Goertzel believes proper decentralized governance should be established before a major breakthrough in AGI.
But merely having decentralized technology and infrastructure is not sufficient to ensure true decentralization and avoid concentration of power.
Without a meaningful decentralized governance mechanism, a decentralized network can still be controlled by a small group or entity, leading to a situation similar to a dictatorship or oligarchy. To achieve true decentralization, it is crucial to establish a governance structure that distributes decision-making power among participants in the network.
Speaking at the SNET Decentralized Governance Summit 2023, Goertzel cited examples such as moderation of social media posts, fair compensation for creators in media and art, and decision-making in the music industry. Centralized governance mechanisms, companies, and governments, he points out, have often failed in these domains.
In the context of music, he raised questions about the involvement of human musicians in shaping the rollout of AI music, considering that AI models are trained on music created by humans. There are similar concerns regarding medicine. Goertzel feels the individuals supplying medical data should have a say in research and the deployment of therapies.
Goertzel emphasized that as AI continues to transform various vertical markets such as manufacturing, supply chain, and finance, decisions should be made by those whose data feeds the AI and those who are directly impacted by its outcomes, and ideally “as much diversity of human hearts and minds as possible”.
To achieve this, his vision is that decentralized governance mechanisms should underlie the platforms and algorithms facilitating these market functions.
There are therefore ongoing efforts to develop decentralized governance mechanisms, with various projects incorporating decentralized autonomous organizations as part of their frameworks. However, achieving effective decentralized governance in the blockchain world is still a work in progress and a challenging task.
Challenges of decentralized governance
The current state of decentralized governance presents some challenges.
While there are existing mechanisms and organizations exploring decentralized governance, many face issues such as token concentration (i.e. too many tokens in some people’s hands, giving them more power) or low participation in voting. These problems, however, should be solvable with focused attention and effort.
Inspired by projects like Cardano's Catalyst program, SingularityNET aims to enhance its decentralized governance gradually. The goal is to empower the community to have a more significant role in shaping the network's future. While progress has been slower than anticipated, the importance of robust decentralized governance, especially in the context of advancing towards AGI, remains a top priority.
The vision is for the community to participate in more decisions regarding the development and direction of the decentralized SingularityNET AI network. However, achieving full decentralization requires time and careful planning. The goal is to strike a balance between decentralization and maintaining order and functionality while progressing steadily towards AGI.
Goertzel aims to radically transform SingularityNET's governance over the next two years, moving towards decentralization and embracing the concept of a SingularityNET Community DAO. This entity will gradually assume functions currently handled by the SingularityNET Foundation and other parts of the ecosystem, ultimately evolving into a fully decentralized DAO. Additionally, the possibility of creating other DAO-like entities to collaborate with the SingularityNET Community DAO is being considered as part of this transition process.
Comments
Post a Comment