Why AGI Will Arrive in Just 2 Years — Not LLMs, But Real Learning AI

What we talked about.

In this episode of the Sunny Ray Show, host Sunny Ray talks with Peter Voss, the entrepreneur credited with coining the term AGI. Voss explains what his company, AIGO.ai, is building: a system capable of true autonomous learning, unlike current large language models that require massive pre training and cannot learn incrementally on the job. He argues AGI could arrive within two years using a human like learning approach rather than the big data scaling strategies pursued by major AI labs. The conversation covers the difference between AGI and the marketing term ASI, why Voss believes today's chatbots are not AGI despite appearing intelligent, and what a world of radical abundance without mandatory work might look like. Voss also shares his personal story, from a childhood split between East Germany and South Africa, dropping out of high school at sixteen, an overland journey through Africa and Asia, building and nearly losing a software company through IPO, and five years spent studying epistemology, psychology and cognition before designing his AGI architecture.

The man who coined AGI explains why real learning, not big data, will get us there in two years.

The questions, and the answers.

What is AIGO.ai building and what's the focus right now?

We are building true human level intelligence, AGI. We previously commercialized an earlier version in call center automation, including for 1-800-Flowers and Harry and David, where the system could understand and learn customer history. About a year and a half ago we paused commercial work to focus fully on making the system learn autonomously, the way a human would onboard into a new job, rather than needing custom engineering for every customer.

How far are we from AGI in your estimation?

With our approach, I believe we're only about two years away. Everyone else uses a big data approach with massive training runs costing billions and trillions of words, but that's not how human intelligence works. A child learns language and reasoning with a few million words. Our system works more like the human mind, needing much less data and compute, and it can continually learn on an ongoing basis without retraining from scratch.

What does AGI arriving mean for humanity, and are people ready?

Ready or not, it's going to happen. It's very unlikely humanity just stops developing AI, and whether it's us or another company, adaptive intelligence is only a few years away. I see it as very positive, both financially and for quality of life. Most people, if honest, would say they wouldn't work if they didn't have to. This points to a future of radical abundance where people have the freedom to choose.

How do we actually get to that radical abundance, through UBI or something else?

I do believe AGI itself will help us figure that out. Initially it might mean working fewer hours as the cost of goods and services drops dramatically, since human labor is the biggest cost factor in most things. I'm hoping for a solution that doesn't put arbitrary power in government hands, something more like everyone owning shares in AI companies that pay automatic dividends, not subject to political whims.

Is there a difference between AGI and ASI, artificial superintelligence?

It's frustrating how AGI has been distorted as a marketing term. Companies without a real path to AGI are now pivoting to say they're going straight for superintelligence, but that's usually just narrow superhuman ability in one domain, like a calculator. We coined AGI to mean the ability to learn a wide variety of things autonomously. Once you actually reach AGI, superhuman traits like photographic memory and constant uptime follow automatically.

Why do you feel AGI isn't already here, given how capable current systems seem?

The key difference is these systems cannot learn incrementally, it's essentially impossible with a frozen, pre-trained model. What looks like personalization is really external databases, not the model itself learning. If we had real AGI, you could put it in a call center or teach it to drive in about the same time a human needs, without custom hardware or software. That we still need those things shows we clearly don't have AGI yet.

Looking back at your path, which area of study gave you the biggest unlock for designing AGI?

Definitely epistemology, along with psychometrics. I helped develop a new type of culturally unbiased IQ test in South Africa, computerizing it to externalize how people actually solve problems. That work showed people without strong formal education still performed well, and it taught me that metacognition, thinking about your own thinking, is actually the most important component of high level intelligence.