Artificial general intelligence – the next step in AI?

Amazing art, convincing voice, and superefficient research: we've all seen what AI can do these past few months. However, it's nothing when compared to artificial general intelligence, the crown of AI development.

The purpose of AI has always been to fill in for us by performing tasks that are mundane, difficult, or for any reason, unwanted. Recent developments with generative AI have shown that we can rely on AI for tasks we can do ourselves, but AI just does it faster and more efficiently.

For now, though, the scope of AI, despite its recent lightning-fast growth, remains limited. Midjourney can paint, but it can't cook. ChatGPT can do your homework, but it's unable to move a chair. However apt a piece of AI is at what it does, give it a different task, and it becomes utterly useless.

But this is not the case with artificial general intelligence, which is equally capable as an average human – at everything. In the words of OpenAI, artificial general intelligence will empower and benefit all humanity with its capabilities.

What is Artificial General Intelligence?

Artificial general intelligence (AGI) is a type of AI that can understand, learn, and solve tasks that require a human level of intelligence. Just like us, it can adapt to different environments and solve various complex problems.

Unlike artificial narrow intelligence (ANI), which is the type of AI that we have today, AGI is not trained for a specific task, but it's made to observe, learn, and adapt on its own without being directly commanded by a specific code.

Nearly everything we do today is, in one way or another, linked to narrow AI. The order of content and ads that you see on the Internet is governed by a specialized AI. Your neighbor's kid is playing a video game against an AI opponent. Perhaps the customer service assistant you just talked to was a synthesized voice of narrow artificial intelligence. These all function seamlessly, but put any one of these narrow AI to do the task of the other, and it will fail.

Artificial general intelligence could learn and perform all these tasks and more. Eventually, when we develop an AGI in earnest, it would display the capabilities of the median human, but retain the potential to become an expert in the field, something we now consider reserved for the narrow AI. It would be possible for AGI to become a scientist, an engineer, or a doctor.

"Would be", as we have yet to reach the point of general AI. But some experts say that we're getting there.

Precursors of AGI

Last year brought us Optimus, Tesla's attempt at making a type of AGI that can get its bearings in a three-dimensional environment and perform simple mechanical tasks. In other words, walk, wave, and carry stuff from point A to point B. While Musk promises commercial use of Optimus in the near future, Optimus' current capabilities are extremely limited. In the future, perhaps?

On the other hand, ChatGPT is an interesting example of AI that treads the fine line between the two types of AI. Built on GPT-3.5, a large language model, it employs both supervised and reinforcement learning techniques in order to deal with a large scope of text-related tasks. It can write poetry, compose music, do programming and debugging, answer complex questions, and even simulate an entire chat room. Talk about a wide spectrum, right?

John Hennessy, a former Stanford University professor and Turing Award recipient, credits ChatGPT and other recently developed AI systems for propelling us towards the general AI "singularity", or the point in time where we come up with a full-fledged AGI.

ChatGPT has its limitations, though. It is still prone to "AI hallucination", or nonsense and false information that it displays as correct, and sometimes shows bias in its answers. These are all issues that are currently being dealt with but remain a reminder that we still have some way to go before we make the leap.

ETA for AGI?

The big question is: when will we create artificial general intelligence?

For proper general AI, there are three vital components of consciousness it must have:

  1. A model of surroundings with the body of the unit at the center
  2. Perception of time for projections of future outcomes
  3. An ability to consider and evaluate multiple actions and choose the most optimal one

In other words, it has to be able to explore, learn, interpret, and solve various problems in different environments. For example, just like a three-year-old child can. 

While current AI models efficiently tackle various tasks, we have failed to develop a system that would harmonize these tasks into a single AI solution. Our AI fails to overcome problems that three-year-old children routinely solve. Forget the toddler, we still have to reach the intelligence of a parrot or a mouse. 

But the necessary research is rapidly advancing, and investments are bigger than ever. John Hennessy estimates that we're 20 to 40 years away from artificial general intelligence. John Carmack, an ex-Meta executive, gives a 95% chance of AGI being developed by 2050 and 60% chance of this happening by 2030.

In our lifetimes! Should we buckle up?

General AI – a Singularity or a Gradual Shift?

When discussing artificial general intelligence, many people mention the singularity, or a point in time after which AI, as we know it, is transformed, a specific moment after which everything will be different for us.

However, as we discover new abilities of ChatGPT and similar models, it is becoming increasingly likely that the advances in AI will come gradually, with them being capable of more and more until we all acknowledge that we have reached something that can be classified as AGI. As OpenAI's Sam Altman says, we most likely would not be able to come to an agreement when that point in time exactly was.

First, it will reach some of our cognitive potential, then become our equal in capability, and ultimately surpass us. But without the fanfare we have come to expect.

But who knows for sure? Let's be patient for a decade or two and see for ourselves. 

An AI Companion

Rather than an omnipotent android, we should expect AGI and AI, in general, to come in the form of an AI companion.

As Hennessy says, AI today is an "amplifier" of abilities, a position similar to Microsoft's term "co-pilot". We can use AI to get a Microsoft PowerPoint presentation or a Google Doc article, and it will most likely not be perfect, but it will be able to get us something that we can work with and use human intelligence to perfect.

This symbiotic relationship most likely won't change with the arrival of general AI. It will still be our companion, just with better capabilities and more ways in which to assist us.

Skippet, using the power of AI, can help you organize your business and personal life, all you need to do is describe your problem, and Skippet will create a custom solution fit for your needs. Make sure to sign up for our waitlist to get access to our beta.

Check out Skippet in action.