Protocols for a big network of AI agents

In 1991, 24 years ago at the time of writing this article, Tim Berners-Lee created the first lightweight, flexible, and universal protocol that would allow us to connect all of humanity to the internet. We are now in the age of Artificial Intelligence, where machines autonomously navigate the internet freely, interact, and generate content, raising many questions and challenges, both technical and ethical, almost straight out of old science fiction movies from the 90s.

In a recent report by Adobe for Business, traffic generated by AI has increased tenfold, e-commerce businesses are already seeing sales referred by ChatGPT, and social media is flooded with content generated by artificial intelligence that goes viral faster, where some see many opportunities, others see big problems.

Security and control. If a model has free access to the Internet, it could read or publish things without supervision, including private or harmful information. Giving it a direct connection is like giving it the keys to the world without filters.

Current AI models face challenges that might be trivial for a human, such as the accuracy of information, as the web is full of contradictory, false, or manipulated data. A model would have to learn not only to search, but also to distinguish what to believe, something that even humans do not do entirely well. Constant updating: the internet changes every second. Integrating information in real time without breaking consistency with what the model already “knows” is a technical and conceptual challenge. Cost and speed: searching, reading, interpreting, and summarizing thousands of pages consumes a lot of energy and time. Processing fixed data is not the same as navigating a changing ocean. Legal and ethical issues, accessing or using copyrighted content, personal data, or sensitive material raise dilemmas that still lack clear frameworks. All this on the same protocol created by Tim B., who simply wanted to share documents with embedded links to each other via a short URL on the internet.

Great opportunities are coming for companies, which are seeing their capabilities increase exponentially when they adopt AI in their day-to-day operations, even reducing costs, for example in customer service with ChatBot, automatic responses, virtual agents, and various call centers in India replacing part of the human staff. In turn, OpenAI announces fully integrated purchases in ChatGPT that do not require the user to interact directly with our site, as well as offering developers the possibility of creating applications that integrate directly into the conversation and can be invoked by the user or suggested by the agent, a small step towards something that seems inevitable: the evolution of operating systems to LLMs.