To Prevent the Monopolization of AI Technology by a Few Companies, Humanity Needs Open-Source AI

Current AI systems, particularly AR-LLMs, lack true reasoning, planning, and common sense understanding. They are not controllable or safe by design. To achieve human-level AI, we need systems that can learn world models from sensory inputs, have persistent memory, plan actions to fulfill objectives, and are controllable and safe.

Yann LeCun, the Chief AI Scientist at Meta, speaks towards Objective-Driven AI architecture which consists of modules for perception, world modeling, cost computation, and action planning, configured by a configurator module. Yann believes that self-supervised learning, using joint embedding architectures like JEPA, can enable machines to learn world models from sensory data without the limitations of generative architectures. Meta AI research suggests energy-based models (EBMs) provide a flexible framework for learning and inference, with advantages over probabilistic models.

The idea that future AI assistants will require human-level intelligence and serve as a shared infrastructure, highlighting open-source AI platforms and crowd-sourced training, is a forward-looking vision at Meta. Let's break it down.

Human-level intelligence

=> To be truly effective and versatile, AI assistants of the future will need to possess intelligence comparable to that of humans.

=> This includes the ability to understand and process natural language, reason, learn, adapt, and exhibit common sense.

=> Human-level AI would enable these assistants to engage in more complex and nuanced interactions, providing users with a more intuitive and helpful experience.

Shared infrastructure

=> AI assistants will become a common, shared infrastructure, similar to the internet today.

=> AI assistants will be widely accessible and integrated into various aspects of our daily lives, such as communication, information retrieval, and problem-solving.

=> As a shared infrastructure, AI assistants would need to be compatible, interoperable, and adhere to common standards to ensure seamless functionality across different platforms and devices.

Open-source AI platforms

=> To prevent the monopolization of AI technology by a few companies and to ensure transparency and accountability, humanity needs open-source AI platforms.

=> Open-source AI would allow for collaborative development, auditing, and improvement of AI systems by a global community of researchers, developers, and users.

=> This approach could help mitigate the risks associated with proprietary AI systems, such as biases, privacy concerns, and potential misuse.

Crowd-sourced training

=> Training AI assistants to achieve human-level intelligence would require vast amounts of diverse, high-quality data.

=> Crowd-sourced training involves leveraging the collective knowledge and experiences of a large number of people to train AI systems.

=> By engaging a diverse global community in the training process, AI assistants can learn from a wide range of perspectives, cultures, and domains, ultimately becoming stronger, unbiased, and adaptable.

Democratization of AI

Making AI assistants accessible to everyone as a shared resource could help bridge the digital divide and empower individuals and communities with advanced technological capabilities. Open-source AI platforms would enable researchers and developers worldwide to contribute to the advancement of AI technology, accelerating innovation and cross-disciplinary collaboration. Transparency and accountability provided by open-source AI platforms, along with diverse crowd-sourced training, will help to ensure that future AI assistants are developed and deployed ethically, aligned with human values and societal interests.

Impact

The widespread adoption of human-level AI assistants as a shared infrastructure could revolutionize various aspects of our lives, including education, healthcare, scientific research, and government, leading to significant advancements and improved quality of life. However... realizing Meta's vision would require ensuring quantum-safe data privacy and security (hint, hint... like QuantumGuard+), developing fortress-like governance frameworks, and requires international cooperation and standards for AI development and deployment.