Evolutionary AI Agents, a Prerequisite for AGI
To understand how AI agents works, it's useful to use the opposable thumb as an analogy. Scientists claims that without our opposable thumbs, we'd be little more than Chimpanzees or Dolphins. If you create a tool for your AI agent, it becomes the equivalent of an "opposable thumb".
To understand the difference between an agent with and without tools, imagine Stephen Hawkins. Regardless of how smart he was, he couldn't even get himself a glass of water without somebody else helping him. And in fact, without his compuers he couldn't even talk to people the last 20+ years of his life.
While any human being capable of understanding English having functioning arms and legs, could easily wash the dishes, go with the garbage, and walk the dog. Hence, "usefulness" from a practical point of view, ignoring theorethical physics, is not about IQ, but rather about having arms and legs, or "the opposable thumb". And the same is true for AI agents; As in, it doesn't matter how "smart" it is - Without tools, it's just sitting there and can't do anything!
Growing Software
When I explain our Hyperlambda Generator, I often use words such as "growing software". From a conceptual point of view, that's actually how it feels like when it is doing its thing. Instead of "creating" software, we "grow software". And in fact, this is an analogy I've had around Hyperlambda for many years. Below is an example where I'm using the Hyperlambda Generator to crawl and scrape websites.
The above of course is a "toy thing", simply because we cannot for security reasons allow random visitors to modify the state of our server with generated code. But with the full experience from the Magic Cloud dashboard, you can create APIs, connect to databases, the file system, and anything you'd expect from a modern programming language. With the difference being it's not a programming language, only a small DSL.
And the last parts is crucial, since it delivers the security features you'd need to be able to use it in production. If you'd try to create something like our natural language API with Python or JS, you'd need to deploy the code into a container of some sort before you could execute it. This process is about 20 billion times more expensive than simply executing the function in-process, literally - And hence therefore for all practical concerns 100% useless.
With Hyperlambda, you could deliver code-gen solutions creating new tools on the fly, as illustrated by our natural language API thing, and each new tool would be ready in some 1 to 3 seconds. With Python the timeframe would be several minutes!
The path to AGI
I've been criticized a million times for my opinions about AGI, but really you think about what humans are, and you realise quite rapidly that the difference between a Chimpanzee and a human being is quite simple ...
We are tool builders!
With the Hyperlambda Generator all of a sudden the machine becomes a "tool builder". Notice, you're still technically in control, since it creates tools according to its security settings, and natural language input - But it completely reverses the equation.
To imagine AGI without something similar as our Hyperlambda Generator is quite frankly absurd! I can create tools, all day long if I wanted to - If the AI can't, it's not "generally" intelligent.
Wrapping up
Without the ability to build tools, we can never have AGI. Today's programming languages such as Python and JS are terrible vehicles for delivering a "tool builder" - Simply because you're one AI hallucination away from doing the format C thing.
If you want a credible path to AGI, you must have a programming language that facilitates for it. To believe anything else is delusional. The language mechanisms is how we "control the thing, avoiding it from going crazy, and destroying the world".
