Five levels of autonomy in autonomous driving were defined and standardized a few years ago to categorize the state of the art of technology. Now, it’s the turn of what we call AGI (artificial general intelligence). The fact that there is an AGI ranking today should make us reflect on the state of maturity we are entering.

OpenAI has unveiled its plan for generalizing its technology, which can be interpreted as a road map, a statement of intent, or a way to maintain momentum. There are no big surprises, as it aligns with what was expected to happen; we understand it as a confirmed standardization.

However, after ten years of development in self-driving, experts indicate that reaching level five of autonomy, that is, fully self-reliant driving in any situation and place without human intervention, is almost impossible. Although autonomous cars have been in San Francisco for years, these vehicles have had millions of miles of training on the city’s closed circuit.

Nowadays, you can’t export that technology to Spain for the summer trip you want to make, you just haven’t trained. The fact that autonomous vehicles are already in circulation in that city does not mean that it is technically and economically feasible to deploy this technology globally in the short or medium term.

Hopefully, we get it wrong with the AGI, and the fact that there is this ranking of levels, a public standardization, does not become a barrier to getting to the last level, to the fifth, the so-desirable Artificial General Intelligence.

Five levels to reach AGI

Level 1: Chatbots, what we currently have. Virtual assistants: our digital counterparts.
Level 2: Reasoners, problem solvers at the doctoral level. Logical analysts, sharp minds that unravel the most complex puzzles.
Level 3: Agents, AI systems that can spend days taking action for you. Unremitting executors, your digital assistants who work without rest.
Level 4: Innovators, your version of DaVinci AI. Technology visionaries are digital inventors that illuminate the way to the future.
Level 5: Organizations, a single AI system doing the work of an entire company. Virtual corporations are the artificial intelligence that manages everything from accounting to business strategy.

Today, we find ourselves halfway between levels 1 and 2. What is expected in future versions of GPT-4o is that reason. It is very different from reason to organize text with meaning, a little hue that changes everything. Virtually, with GPT-4, we have already used all the data or knowledge existing. Then, evolution from here will be by happy ideas, synthetic data, or a new architecture that they invent.

What’s coming right away are the agents. We’ve been working with agents for a year now. However, they’re still beginning, and the early versions we’re seeing aren’t convincing us, be they very specialized bots in a knowledge base or more realistic interfaces for voice interaction.

They’re there, but they haven’t slept yet. What you expect from an agent is not to answer a question as an assistant, it’s to do an entire task from start to finish, to open a Spotify list for you and play it. It is a completed task, yes, but we are becoming more and more demanding, and what we want is that the agents complete more complex and sophisticated tasks, which require internal interoperability between agents of the same technological product and external interoperability, that is, coordination between different agents of other products.

This problem is not trivial. Through intercommunicating APIs by TCP/IP, you can solve certain tasks in series, but we want to go further. It is a good time for a new exclusive telecommunication protocol to be born and dedicated to solving the problem of communication between agents. Here, the telco could have a golden opportunity to innovate in this emerging AI scenario.

Anyway, what would be my ideal agent? I would answer that a replica of myself; I don’t want to have 20 agents specialized in specific tasks, one per task or area of knowledge; I want one multitasking that solves all my tasks and problems throughout my day, my dream as I commented on you in the article of GPT-3 changes everything, is to replicate all my knowledge in one agent, so that I can go to the beach, and let it work all day to the agent independently.

How would this product be solved? The idea is not to implant a Neuralink chip and wrap all the information in my brain; it’s much more straightforward. Your operating system records every action you do, every software you open, every document, everything you browse, and the routines and sequences of your day-to-day work. Microsoft with Windows is the company that could make you this agent, as it is the firm that has these low-level data. In the future, your operating system will learn enough about you to become an autonomous agent that can replace you under minimal supervision. This I mentioned is not the latest version of Windows Copilot; it is another different product.

On level 4, which is expected to be very sophisticated with respect to AI’s inventive capacity, today, we already have inventions with my product Vizologi. You can create project innovation portfolios for your company. In one hour, you can create 60 new ideas for your business. Of the 60, there will be 6 that will be extraordinarily good, applicable, and executable as developers of innovation and growth in your business.

At level 4, however, you are not going to work in the same way, what is expected is to offer solutions on the unknown, that you invent beyond the human knowledge with which you have been trained. As I mentioned in previous articles, a very high level of personalization will also be achieved at this level. If you need project management software ultra-personalized to your needs, you can ask this AI to do it for you. At this level, we can tailor-made, ultra-personalized technological products that probably do not require programming or technical development.

Finally, you would get to level 5, which is theoretically there; at the peak, there will be the AGI. If you honestly expected a little more of a definition, even this milestone could be resolved at level 3 with the agents. In theory, at this last level, you would have a level of coordination and intelligence so high that the AI could perform all the work that is carried out in a company.

Pedro Trillo

CEO at Vizologi | Website

Pedro Trillo is a tech entrepreneur, telecommunications engineer, founder of the startup Vizologi, specialist in Generative Artificial Intelligence and business strategy, technologist, and author of several essays on technology.

Share: