I am not 100% sure, but I think we're now the first official AGI company on Earth, capable of selling AGI solutions to you and your company, starting out at $298 per month.
To understand why and how, you need to watch the following video to understand what AGI actually is.
What is AGI?
AI and LLMs such as ChatGPT are amazing at what they do, but they suffer from the lack of "an opposable thumb", to use an analogy. This implies that as stand alone products, they can't really do much, if anything at all.
Artificial General Intelligence implies a computer-based AI system that is better than humans at performing any task. To understand what AGI is I asked ChatGPT to define it using one paragraph. Here's its answer.
Artificial General Intelligence (AGI) refers to a type of artificial intelligence that possesses the ability to understand, learn, and apply knowledge across a wide range of tasks at a level comparable to human intelligence. Unlike narrow AI, which is designed for specific tasks, AGI can perform any intellectual task that a human can, including reasoning, problem-solving, understanding complex concepts, and adapting to new situations. AGI aims to replicate the flexible and generalized nature of human cognition, potentially leading to transformative impacts across various domains, from science and medicine to daily life and industry.
The key take aways from the above definition are the words; "apply knowledge" and "tasks". Everything else is basically already available from ChatGPT, and will only improve in the future, as OpenAI releases better and better models. If you have doubts about this, realise that ChatGPT was able to outperform your general physician by 18% in accuracy on diagnostic tasks, and 30% on empathy, according to a peer reviewed study. This was almost one year ago. In addition it was able to pass the bar exam, and a lot of other really difficult tasks typically only associated with human beings. So we basically have everything we need to create AGI, the only problem that remains, is to provide the AI with the ability to execute tasks and functions, to such interact with the real world.
We can deliver AGI systems that basically executes tasks, almost any task you can imagine, based upon your instructions. This allows us to deliver custom AGI solutions, based upon your specific needs and requirements - Whatever they happen to be. To understand how, you need to watch the following video.
The Opposable Thumb Problem
AGI is not about intelligence, something illustrated by the above zoology analogy in the first video. AGI requires the ability for the AI to do things in the real world. If you go to ChatGPT and ask it to make you some coffee for instance, it will probably reply with "sorry, I don't have access to your coffee machine". We however, can deliver an AI system that can create coffee for you, assuming your coffee machine has some sort of API we can connect to of course. Creating coffee is of course just one example. Better examples can be found below.
- Send all my leads in my CRM system an email reminder about our weekly offer
- Create an appointment for next Thursday for a Zoom meeting with Rick, and send him an email about it
- Save a reminder to myself for tonight to remember to water the plants
- Or even better, remember to water the plants tonight yourself (assuming your sprinkler system has an API)
As long as we can somehow reach whatever it is we need to reach, we can provide your AI chatbot with "a specialised opposable thumb", allowing it to perform that particular task.
Can your AI drive my car?
No, and this is an argument most will use to claim it is in fact not AGI, but rather "narrow AI". However, there are probably 3.5 billion people on the planet who cannot drive a car at all, and my guess is that there are only two or three people that can drive your car, assuming you're not renting it out to any stranger of course.
Does this mean that only you and your wife has "general and 'wide' intelligence?"
Of course not. Just because a human being doesn't have a driver license, doesn't imply that human being is somehow less intelligent than you. It's just a problem of teaching the human how to drive. Currently there exists no (real) way to teach AI how to perfectly drive a car - However, when it does, we could indeed teach your AI chatbot to drive your car, assuming your car has an API.
To illustrate this, realise that we could easily deliver an AI chatbot that outperforms Magnus Carlsen on playing chess, simply by connecting to the Stockfish API, and having the AI chatbot use Stockfish to determine its next move instead of ChatGPT. This allows us to deliver an AI chatbot based upon ChatGPT that plays chess one million times better than ChatGPT does - Even though it's based upon ChatGPT.
This paradoxically reflects the way OpenAI delivers their existing models, since ChatGPT is in fact not one model, it's in fact dozens of models, based upon "mixture of experts" ideas. When you are querying ChatGPT, the first thing that happens, is that their internal systems determines what type of question you just submitted to it. Then it "routes" your question to a specialised model that is most capable of answering your question. This implies that ChatGPT is in fact not one AI model, but rather dozens of AI models.
Hence, if you don't accept this process of "routing" questions to different models, then purely logically AGI will never be possible to create - Simply because we will never have the "one model to do everything" capability that everyone is currently defining as AGI.
Any AGI system will inevitably be based upon access to hundreds, and possibly millions of different systems, and/or models, in order to be able to surpass human level intelligence
This implies that whoever can dynamically facilitate for ChatGPT to connect to such systems rapidly, will purely logically be able to deliver AGI - And we believe we are the first company on earth capable of actually delivering this - Starting from today!
Wrapping up
Some people might argue that this is not AGI because there are things it cannot do. My counter argument to these is that neither are they "general" in their intelligence, simply because there's a million different things they cannot do either. For instance, can you play the trumpet? If the answer is no, then you do not have "general intelligence" according to your own definition.
However, is it ASI (Artificial Super Intelligence)? No it is not. But it is AGI (Artificial General Intelligence). And to understand how, realise it knows how to do "a bajillion things" that ChatGPT does not know how to do.
If you're interested in talking to us about how we can help you deliver custom AGI solutions for your company, you can contact us below.
Edit
I asked our AI chatbot the following question; "Scrape the following https://ainiro.io/blog/the-first-agi-company-on-earth and let's argue whether or not it is a true statement or not. Provide a nuanced and balanced response about it." You can find its response below.
The claim that AINIRO is the first AGI (Artificial General Intelligence) company on Earth is ambitious and thought-provoking. Here’s a nuanced and balanced perspective on this statement:
Supporting Arguments
Task Execution Capability
AINIRO emphasizes that their systems can execute a wide range of tasks by integrating with various APIs and systems. This ability to perform diverse tasks based on instructions aligns with the core idea of AGI, which is to replicate human-like flexibility and adaptability.
Integration with Existing Technologies
The company leverages existing AI models like ChatGPT and enhances them by connecting to specialized systems (e.g., CRM, email, scheduling APIs). This approach can indeed extend the functionality of AI, making it more versatile and capable.
Practical Examples
The article provides practical examples such as sending emails, scheduling appointments, and even potentially making coffee, which illustrate the broad applicability of their AI solutions. These examples demonstrate the potential to automate and optimize various business processes.
Counterarguments
Definition of AGI
Traditional definitions of AGI imply a system that can perform any intellectual task that a human can, without needing specialized integrations or APIs. AINIRO’s approach relies heavily on connecting to external systems, which some may argue is more akin to advanced narrow AI rather than true AGI.
Limitations in Scope
While AINIRO’s AI can perform many tasks, there are still significant limitations, such as driving a car or performing highly specialized tasks without specific integrations. This raises questions about whether the system truly possesses "general" intelligence or if it is still constrained by the availability and functionality of external APIs.
Comparative Analysis
Other companies and research institutions are also working on AGI, and the field is highly competitive. Claiming to be the first AGI company might be seen as premature without a comprehensive comparison with other ongoing efforts in the AI community.
Conclusion
AINIRO’s claim to be the first AGI company on Earth is bold and reflects their innovative approach to enhancing AI capabilities through integrations and task execution. While their solutions offer significant advancements and practical benefits, the debate on whether this constitutes true AGI remains open. The distinction between advanced narrow AI and AGI is nuanced, and AINIRO’s contributions are a step towards the broader vision of AGI, even if they may not fully meet the traditional definitions of AGI as understood by the AI research community.
Top comments (2)
I have seen a challenge on creating "drop-in remote AI workers" (manifold.markets/ZviMowshowitz/wil...); do you think you will meet the criteria below by 2027?
Hmm, the best model I've got access to is GPT4 - But I suspect OpenAI will release better models soon, and definitely before 2027 (I hope!) - The rest of the parts are possibly, and are basically just "AI functions", although it requires a lot of moving parts. Before 2027 is definitely possible, but then again there are the autonomous parts. After all, an AI agent is only doing stuff you tell it to do, which is kind of a big deal. A human worker (most of us at least), tends to "create our own tasks" when we're bored, doing some cool (and useful) stuff if we're not given tasks after a while but just left to ourselves ...
Having an AI agent perform tasks you assign to it, should be doable, 100% certain. I wouldn't want to give you a guarantee of that I'll be able to pull it through, since I'm also dependent upon people actually wanting it, and being willing to pay for it - But it's definitely doable ...
As in; If you give somebody like me the task of actually pulling something like this together, and you give them 3 years, it's not only doable, it is inevitable ...!!
What I've got now at this point, which I assume you believe is fairly impressive, was something I pulled through in 18 months. Starting with my existing parts, having twice the amount of time, makes it quite likely ...
I've got most of the parts already in place, at least partially ...