AGI is likely around the corner. We need to mentally prepare for it. Here are three avenues as I see this unfold.
Recently at a talk in Boston, I walked through these concepts with the audience.
Quick Step Back:
The adult brain has approximately 100 billion neurons, it orchestrates over 100 trillion connections—far outstripping the complexity of what we have seen with all connected devices and content on the Internet. That said, our current most powerful transformer model, the T in GPT, only handles 1.2 trillion connections, but seriously, looky at what it can do now with only 1.2% of the connections of our human brain.
This staggering neural network we are seeing unfold continues to confound the best scientists. We still have the most basic questions left to answer, such as, how do we experience intuition? That sudden, seemingly magical understanding of something without conscious reasoning is a puzzle. Some researchers argue that intuition is the brain’s way of drawing on subconscious memories and knowledge, while others suggest it’s simply a cognitive shortcut our minds take to save time. Regardless, it remains a mystery how we process such insights with so little conscious effort.
Then there’s memory—how can a melody or scent suddenly transport me back to 2nd grade when on the playground and someone shouts, “rover red rover, send Joe right over,” and I ran full speed into my teacher’s back? She returned later that afternoon to the classroom, thankfully. Now every time I smell Bengay, I think of that moment. Despite extensive research into brain structures involved in memory, we are only just barely sniffing the surface of how memories are truly coded, saved, and recalled.
This is where we all must lean into a thought. It’s important to begin preparing ourselves mentally for what’s about to happen. So, if that’s not too much to wrap your head around that we don’t know how our brain processes smell, there is more, much more. As we grapple with the depth of our own minds and how they work, we must now confront the path toward artificial general intelligence (AGI)—an AI that matches or surpasses human cognitive abilities. It’s no longer a question of “if,” but of “how” and “when.” The answer lies in three key trends (in my presentations, this is the “Trinity of Forces”) that are guiding us toward this future. Each of these trends alone has the power to drive significant change, but together, they might just lead to the birth of AGI.
1. The Scaling Phenomenon: Bigger, Faster, Stronger AI
When the first Transformer was cobbled together at Google in 2017 by Geoffery Henton’s team, I doubt they saw how quickly it would hit. This all changed in 2022, and can you imagine what Google thought when the chat function exploded in 2022, by an upstart leveraging their initial idea? If you think GPT-4 is impressive, imagine a model like GPT-5 or even GPT-10. This is the core of what’s known as the LLM “scaling hypothesis,” essentially, the bigger and faster these models get, the better they perform. Yuval Noah Harari has said he expects GPT-5 to be 1000 times more powerful than v4.
We’re witnessing my favorite phrase “exponential growth” improvements as AI models increase larger in size, complexity, and processing power. Some at OpenAI have said they don’t foresee a ceiling to this scaling. Beyond what people say, it is not about pinpointing an exact date for AGI, it is more about recognizing that these models are already excelling at many tasks beyond human capability. One could argue, based on how Claud 3.5, GPT 4-o1, or Grok perform, that we might already be there. Imagine a scientist waking up from coma that begin in 2015 and playing with GPT-4o from her hospital bed. She might think we had already hit AGI with its ability to reason and think. The goal posts seem to keep moving on how we define it.
As with chips doubling in speed every six to twelve months, i.e. Moore’s Law, scaling has a theoretical limit. It’s possible that at some point, models will hit a plateau due to constraints in energy, data availability, or simply the cost-to-benefit ratio. At the moment, I don’t see these models going away, because they seem to be the easiest way for us to connect with incredibly complex systems. These interactions will only grow more intense. As I have predicted, I see future AI becoming best friends for some children, even adults. It may seem odd or unusual now, but it will just make sense eventually. Something that knows everything about you, also has the world of knowledge behind it, will be able to sympathize, interact with compassion, and provide thoughtful guidance. It’s surreal, but it’s right around the corner.
So, the future of generative AI is not just an assistant but a core driver of creativity, friendship, productivity, and decision-making.
2. The Fusion of Specialized AI: Toward Human-Like Capabilities
While scaling is near limitless, for the moment, an important second trend focuses on specialization and the fusion of AI systems to mimic human-like functions. Instead of creating one AI that does it all, companies are developing specialized models that excel in specific tasks—think reasoning, planning, creativity, memory, and even sensory capabilities. Since the LegalTech world is near and dear to me, you are starting to see it finally have a moment in this world of specialization around reasoning. For years, LegalTech lagged most industries with leveraging new technology to keep pace. They didn’t need to keep up. Clients paid high fees because, on the whole, no one pushed back. Now, we are at a moment of inflection, where the capabilities of AI, coupled with data, and expertise can virtually transform the entire legal ecosystem. What is core to LLMs? Language. What do lawyers specialize in? Language. Now it is only a matter of time when the culmination of legal data (considered finite comparatively), combined with an LLM trained on the data, and expertise imparted will alter the legal landscape forever. What we will witness is a rapid movement to systems doing much of the research, writing, analysis, comparisons, and reasoning, eventually that will move up the ladder from first year associates to the managing partner. In a short time, we will witness a major shift away from human attorney legal work, to legal AI agent work. It’s happening now, and will only rapidly increase. Your next lawyer will be a Legal AI Agent esquire in a few years. Honestly, the agent can likely pass the Bar now. The mega 5000 attorney law firm is coming to an end. These firms will transition to legal AI agents servicing clients.
Now, what if you want to combine a law firm with a technology consulting firm. When synthesized, these specialized systems could create an AI that’s more general in its abilities. Take multimodality, for example—the fusion of data from different sensors, akin to how humans integrate information from their senses. Picture an AI system that can “see” through cameras, “hear” through microphones, and “feel” through pressure sensors. AI systems today are advancing into multimodal territory, allowing them to navigate and interact with the physical world in much more intuitive ways.
Chain-of-thought reasoning is another pivotal advancement, where AI systems break complex problems into smaller, manageable steps, mimicking the way we process difficult tasks. By combining these capabilities with planning and strategic decision-making—what the military first introduced as “AI scaffolding”—we’re inching closer to AI systems that can orchestrate their own problem-solving. Maybe we are already there?
Even more intriguing is agentic AI. These autonomous systems act as general-purpose agents, capable of making decisions, executing tasks, and even cooperating with other AIs to solve complex problems. Agentic AI is like having a tireless intern with expertise in virtually everything—no micromanagement needed.
This specialization and coordination could allow small and medium-sized enterprises to harness cutting-edge AI tools without needing to match the deep pockets of tech giants. A dynamic AI ecosystem will foster innovation across industries, all while pushing us closer to AGI.
3. Fundamental Technological Shifts: Building the Backbone of AGI
Lastly, we must consider the broader technological landscape that underpins AI development. These fundamental technologies—processing power, data infrastructure, and energy systems—are evolving rapidly, setting the stage for AGI. It’s that exponential growth that doesn’t seem to want to stop anytime soon.
The most exciting development might be in processing power, where quantum computing could revolutionize AI’s capacity to model real-world systems. I’ve been talking about quantum for too long, and based on a recent conference visit, seems like we are finally moving closer to real effective use. Quantum computers, with their ability to solve problems beyond classical computing’s reach, could be a game-changer for AI systems. Though quantum computing itself won’t directly drive AI forward, its eventual integration with classical systems will create a hybrid computational environment optimized for AI performance.
The energy sector will also play a crucial role. As energy becomes cheaper and more renewable—imagine fusion breakthroughs reducing costs to near-zero—the operational expenses of running large AI systems will plummet, enabling wider and more sustainable deployment of advanced AI technologies. We will be at a point where electricity could be a nominal price for most, and these systems will eventually use most of the electricity produced.
Finally: Convergence and the Road to AGI
While each of these three trends will drive progress in its own right, the real breakthrough will come from their convergence. It’s easy to imagine a future where we combine a scaled LLM, advanced multimodal capabilities, and quantum-powered compute systems to create a truly general AI. This convergence could redefine industries and, at the risk of sounding dramatic, humanity itself.
The path forward is clear: today’s AI systems are just the beginning. As we progress through the stages of scaling, specialization, and technological shifts, we are moving closer to a reality where AGI isn’t science fiction but a tangible part of our world. It will be a defining moment in human history—a moment when machines no longer simply assist, but think, reason, and create alongside us.
We’re witnessing the dawn of a new era, and the journey to AGI is already underway. Whether it comes in 2 years or 10, the question is not whether we’ll reach AGI but how ready we’ll be when it arrives. Humor will help, but really acceptance that intelligence will continue beyond us as an evolutionary steppingstone will be key.
Now head out on a vacation and touch grass.
