Artificial Intelligence (AI) - Real Buzz?

There has been a lot of buzz and excitement surrounding Artificial Intelligence (AI), particularly since late 2022. I aim to list the many forms of AI and evaluate where we are now, for which I did use my understanding of this topic and experience in practical applications of few use cases with my team.

I thought of compiling all the pertinent data into a single blog article that possibly would make a better business?sense and I truly hope that this benefits anyone in beginning to gain the desperately needed understanding of AI and its context.

The main categories of AI are outlined below, with the possibility of incorporating other types as subcategories that you might have read/heard about, as many terminologies are presently in use.

What is AI and what are the different types of AI?

Artificial intelligence (AI) is a broad term that encompasses the creation of machines or systems that can perform tasks that normally require human intelligence, such as reasoning, learning, decision making, perception, and creativity.

AI can be classified into different types based on the level of intelligence and the scope of tasks that they can achieve.

In this blog post, we will explore four main types of AI: narrow AI, general AI, broad AI, and super AI.

Narrow AI

Narrow AI, also known as weak AI or applied AI, is the most common and widely used type of AI today. It refers to AI systems that are designed to perform specific and well-defined tasks, such as face recognition, speech recognition, web search, spam filtering, product recommendation, self-driving cars, etc. Narrow AI systems are usually based on machine learning or deep learning techniques that enable them to learn from data and improve their performance over time. However, narrow AI systems are limited by their predefined goals and rules, and they cannot handle tasks that are beyond their scope or domain. For example, a narrow AI system that can play chess very well cannot play any other game or understand natural language.

Some examples of narrow AI systems are:

- Siri, Alexa, Cortana, and other virtual assistants that can understand voice commands and perform simple tasks.

- Google Translate, Duolingo, and other language translation and learning applications that can translate text or speech between different languages.

- Netflix, Amazon, Spotify, and other recommendation systems that can suggest products or content based on user preferences and behavior.

- Tesla, Waymo, Uber, and other self-driving car companies that use computer vision and sensor fusion to navigate roads and traffic.

- IBM Watson, Google DeepMind, OpenAI, and other research projects that use deep learning and reinforcement learning to achieve superhuman performance in games such as Jeopardy!, Go, Chess, etc.

General AI

General AI, also known as strong AI or human-level AI, is the hypothetical type of AI that can perform any intellectual task that a human can do. It refers to AI systems that can understand and reason about the world, learn from any data, communicate in natural language, and exhibit common sense, creativity, and emotions.

Specific domains or goals do not restrict general AI systems; they can adapt to new situations andchallenges. general AI or super AI is the ultimate goal of AI research but is also the most difficult and elusive. Many experts believe that achieving general AI is possible but not imminent.

Some examples of general AI systems are:

- HAL 9000 from 2001: A Space Odyssey , a fictional computer system that can control a spaceship and interact with humans in natural language.

- Data from Star Trek , a fictional android that can mimic human appearance and behavior.

- Sophia from Hanson Robotics , a real-life humanoid robot that can express emotions and converse with humans.

- AlphaZero from Google DeepMind , a real-life computer program that can learn to play any board game from scratch without human guidance.

Broad AI

It’s being used interchangeably with strong AI which again is used as general AI but few suggest it’s not general AI? So be it, we will keep this separate as the one more type for now which is an emerging area of AI research that aims to bridge the gap between narrow AI and general AI.

Broad AI is a type of AI that lies between narrow AI and general AI. It refers to AI systems that can perform multiple tasks across different domains or contexts. Broad AI systems are more flexible and versatile than narrow AI systems but less intelligent and capable than general AI systems. Broad AI systems can leverage transfer learning or meta-learning techniques to apply their knowledge or skills from one domain to another.

Some examples of broad AI systems are:

- GPT-3 from OpenAI , a real-life natural language processing system that can generate coherent text for various purposes such as answering questions, writing essays, creating chatbots, etc.

- DALL-E from OpenAI , a real-life computer vision system that can generate realistic images from text descriptions such as a pentagon made of cheese or a snail made of a harp.

- MuZero from Google DeepMind , a real-life reinforcement learning system that can learn to play any video game without knowing its rules or objectives.

- AIXI from Marcus Hutter , a theoretical model of an optimal general reinforcement learning agent that can learn anything from any environment.

Super AI

Super AI is the type of AI that can surpass human intelligence in every aspect. It refers to AI systems that can think, reason, solve problems, make judgments, plan, learn, and communicate better than any human or group of humans. Super AI systems can also have cognitive properties such as self-awareness, consciousness, and emotions. Super AI is a hypothetical concept of artificial intelligence that has not been realized yet. Some people think that super AI is inevitable and desirable, while others think that it is impossible or dangerous.

Where we are now?

As per experts currently, we are just scratching the surface.

As AI research and development progresses, we can expect to see more advances and breakthroughs in all types of AI in the near future.

Achieving general AI or an advanced super AI that can perform any intellectual task a human can do, is a very challenging and ambitious goal.

There is no consensus among AI experts on how close we are to reaching it, or even if it is possible at all.

I have read several surveys that confirm how fluid the current situation is:

One survey of AI experts have used different definitions and methods for asking about general AI.

For example, a survey conducted in 2018 asked about high-level machine intelligence (HLMI), which was defined as machines being able to achieve most human professions at least as well as a typical human. The median estimate for the probability of achieving HLMI by 2040 was 10%, and by 2060 was 50%.

A survey conducted in 2019 asked about artificial general intelligence (AGI), which was defined as machines being able to understand and perform any intellectual task that humans can. The median estimate for the probability of achieving AGI by 2035 was 10%, and by 2050 was 50%.

One recent survey of 300+ AI professionals performed in 2023, which says that the median estimate for the likelihood that a better general AI would exist by 2030 was 50%.

In other words, half of the experts provided a date before 2045 and the other half provided a date after 2045.

Ray Kurzweil, Google’s director of engineering and a pioneer of pattern recognition technology, believes that AI will reach “human levels of intelligence” in 2029 and surpass human intelligence by 2045.

As you can see, different surveys have different results, but they all suggest that there is a significant chance that general AI or super AI will be developed within the next few decades. However, these predictions are not based on solid evidence or guarantees but rather on expert opinions and intuitions, which can be influenced by many factors such as personal beliefs, biases, expectations, experiences, and preferences. Therefore, they should be taken with a grain of salt and not as definitive answers.

Moreover, achieving general AI or super AI is not only a technical problem but also a philosophical, ethical, and social one. Some many open questions and challenges need to be addressed before we can create and interact with general AI safely and responsibly. For example:

- How do we define and measure intelligence, both human and artificial?

- How do we ensure that general AI is aligned with human values and goals?

- How do we prevent or mitigate potential risks and harms, such as misuse, abuse, bias, discrimination, unemployment, inequality, conflict, etc.?

- How do we foster trust, cooperation, communication, and understanding between humans and general AI or super AI?

- How do we respect the rights, dignity, autonomy, and agency of general AI or super AI?

These are some of the questions we need to consider and discuss as a society before we can welcome general AI or super AI into our world. Achieving general AI or super AI is not only a matter of when, but also of how and why?

In conclusion, it is important to continue the AI journey, explore its applications, and not be overly concerned about reaching a final AI destination. The field of AI is constantly evolving, and there is much to discover and learn along the way. By embracing this mindset, we can fully appreciate the potential of AI and its impact on various domains.