What is Project Q*? What is artificial general intelligence(AGI)? Is it Dangerous!

What is Project Q*?

Before we continue, it's important to mention that all the information about Project Q* comes from recent reports related to the controversy surrounding Altman's firing. According to Reports at Reuters on November 22, the details were shared by "two people familiar with the matter." This gives us some insight into what was happening internally in the weeks leading up to the firing.

The article mentions Project Q*, a new model that's really good at learning and doing maths. Right now, it's said to be on par with solving maths problems at a grade-school level. Even though it's just starting, it shows promise and could potentially demonstrate a kind of intelligence that researchers haven't seen before.

It might seem okay at first, but hold on. There's something called Q* That's a bit alarming. Some researchers got so worried about it that they wrote a letter to the board, saying it could “threaten humanity.”

On a different note, some explanations for Q* are not as groundbreaking. The Chief AI scientist at Meta, Yann LeCun, tweeted that Q* involves replacing "auto-regressive token prediction with planning" to enhance the reliability of large language models (LLMs). LeCun stated that this is a problem all of OpenAI's competitors have been working on, and OpenAI hired someone specifically to tackle this issue.

LeCun is expressing that the mentioned development isn't groundbreaking or unique; other AI researchers are already talking about similar things. However, in response to Altman, LeCun is critical, saying Altman has a history of being overly optimistic about his ideas. LeCun isn't convinced that there has been a significant improvement in solving the problem of planning in learned models based on the information about Q*.

What is artificial general intelligence(AGI)?

Artificial General Intelligence (AGI) is a kind of AI research where scientists try to make software that thinks like humans and can learn on its own. The goal is to create software that can do tasks even if it wasn't specifically taught how to do them. It's like making a computer system that can be really smart, just like people.

Right now, artificial intelligence (AI) works based on specific rules set by humans. For instance, AI programs that recognize images can't create websites. But, there's this idea called Artificial General Intelligence (AGI), which is like teaching AI to be more independent. AGI would be able to control itself, understand things about itself, and learn new stuff, even in situations it wasn't prepared for. The goal is to have AGI with abilities similar to humans, but right now, it's mostly a theory and something researchers are working towards.

What is the difference between artificial intelligence and artificial general intelligence?

Over the years, AI researchers have achieved big milestones, making machines smarter, almost like humans in specific tasks. For instance, AI summarizers use fancy models to pick out important information from documents and make a simple summary. AI is like a computer science superhero, helping software tackle tough tasks as well as humans.

Now, let's talk about AGI, the superhero's even cooler cousin. AGI can solve problems in different areas, just like humans, all on its own. Unlike regular AI that needs a lot of training, AGI can teach itself and solve tricky problems it wasn't specifically trained for. AGI is like the dream of having a super-smart computer that can handle all kinds of tasks with human-like smarts. Some computer scientists think of AGI like a computer program with human-like understanding and thinking skills. Unlike today's AI, which needs special training for specific tasks, AGI is like a brainy computer that can adapt to new things without extra training. For example, imagine a smart language model that talks about medicine—it can learn to chat about medical stuff without lots of special training.

Strong AI compared with weak AI

Strong AI is like a super-smart computer that can do things just like a human, even without knowing much beforehand. It's often portrayed in science fiction as a machine that thinks like a human, understanding everything without any limits.

On the other hand, weak AI, or narrow AI, is like a computer that's good at specific tasks it's designed for. It follows certain rules and doesn't remember much from the past. Even the fancier AI with better memory is still limited to the tasks it was made for and can't switch to doing something completely different.

What are the theoretical approaches to artificial general intelligence research?

To make AGI (Artificial General Intelligence), we need a lot more technology, data, and connections than what we use for AI now. Things like creativity, understanding, learning, and memory are crucial to make AI act like humans. People who know a lot about AI have suggested different ways to study and develop AGI. Here are some approaches to artificial general intelligence.

Symbolic

The symbolic approach assumes computers can learn AGI by using logic networks to represent human thoughts. It's like giving the computer a set of rules (if-else logic) to understand ideas better. However, this method may struggle to copy some detailed thinking skills, like perception.

Connectionist

The connectionist approach tries to copy how the human brain works using something called neural networks. In our brains, neurons can change how they pass along information when we interact with things around us. Scientists believe that using this approach in AI models can make them more like human intelligence, especially in basic thinking abilities. Examples of this kind of AI include big language models that understand languages naturally.

Universalists

Researchers who follow the universalist approach are working on dealing with the complicated parts of AGI at the calculation level. They're trying to come up with theoretical solutions that they can later use in real, practical AGI systems.

Whole organism architecture

The whole organism architecture approach means connecting AI to a model of the human body. Scientists who like this idea think that AGI can only happen when the system learns by interacting with the physical world.

Hybrid

The hybrid approach combines different ways of representing human thoughts in AI to get better results. Researchers try to mix and match various ideas and methods to make progress in developing AGI.

What are the technologies driving artificial general intelligence research?

Making AGI is still something far away that researchers are working on. They're trying hard, and new things keep coming up that help them. Let's talk about some of these new technologies. Here are some technologies driving AGI research.

Deep learning

Deep learning is a part of AI that trains computer networks with many layers to figure out complicated patterns from raw information. Experts use deep learning to create systems that understand text, sound, images, video, and more. For instance, Ecotence developers use Amazon SageMaker to make simple deep learning models for IoT (Internet of Things) and mobile devices.

Generative AI

Generative Artificial Intelligence (Generative AI) is like a smart tool within deep learning. It can create new and realistic content based on what it has learned. Imagine it's like a super-smart assistant that can understand and answer questions using text, audio, or visuals, just like how humans do. Companies use Generative AI, like LLMs from AI21 Labs, Anthropic, Cohere, and Meta, to tackle tricky problems. Ecotence deploys these smart models quickly using various service platforms.

NLP

Natural Language Processing (NLP) is a part of AI that helps computers understand and create human language. NLP uses computer language and learning technology to turn words into simple forms called tokens and understand how they relate.

Computer vision

Computer vision is a tech that helps systems understand and use information from images. Like how self-driving cars use it to see the road and avoid things. Deep learning makes this tech smart, letting it recognize and understand lots of things in pictures.

Robotics

Robotics is a type of engineering where companies create machines that can move on their own. In AGI, these machines help AI systems do things in the physical world. This is important for giving AI the ability to sense and interact with things physically. For instance, if we attach a smart robotic arm to AGI, it could sense, grab, and peel oranges, similar to what humans do.

What are the challenges in artificial general intelligence research?

Computer scientists face some of the following challenges in developing AGI.

Make connections

Regular AI works well in one area but can't connect ideas between different areas. Humans, on the other hand, use what they know in one area to understand and solve problems in another. For instance, ideas from education help in making fun video games. People can also use what they learn in books to solve real-life problems. But, to make computer models do this, they need a lot of practice with specific information.

Emotional intelligence

While deep learning models show potential for AGI, they still can't match human creativity. Creativity involves emotions, something current neural networks can't fully understand. For instance, humans respond to feelings in a conversation, but models like NLP generate responses based on patterns they learned from language data, not real emotions.

Sensory perception

For AGI to work, AI systems need to physically engage with the outside world. Apart from being good at robotics, the system has to see and understand the world like humans do. The current computer tech needs to get better before it can recognize shapes, colours, taste, smell, and sounds just like humans.