What Exactly is Artificial Intelligence (AI)

byrobys53

263 followers

1 year ago

What Exactly is Artificial Intelligence (AI)

The branch of computer science devoted to artificial intelligence (AI) is enormously concerned with creating intelligent machines capable of doing activities that typically need human intelligence.

While AI is an interdisciplinary discipline with many methodologies, breakthroughs in machine learning and deep learning, particularly, are driving a paradigm shift in almost every aspect of the technology sector.

Machine learning allows machines to mimic or even enhance the capabilities of the human mind.

From self-driving cars to the emergence of generative AI tools like Google's ChatGPT and Bard, AI is fast becoming a part of everyday life - and a sector in which businesses of all sizes are investing.

Understanding Artificial Intelligence

In general, artificially intelligent systems can execute tasks associated with human cognitive capabilities, such as language interpretation, gameplay, and pattern recognition. They typically learn to accomplish this by analysing enormous volumes of data and looking for patterns that they can apply to their own decisions. Humans often supervise AI learning by praising sound conclusions and discouraging bad ones. Some AI systems, on the other hand, are meant to learn without supervision, for example, by playing a video game repeatedly until they figure out the rules and know how to win.

AI strength vs. AI weakness

Because intelligence is challenging to quantify, AI experts typically distinguish between strong and weak AI.

Effective AI

Artificial general intelligence (also known as "strong AI") is a machine that, like humans, can find solutions to problems for which it has never been trained.

This is the type of artificial intelligence we see in movies, such as the machines in Westworld or the character Data in Star Trek: The Next Generation. This type of AI does not currently exist.

Many people working in artificial intelligence consider it their ultimate goal to design a machine on par with the human mind. Yet, the pursuit of artificial general intelligence is plagued with obstacles.

Some argue that research into strong AI should be restricted due to the risks of developing powerful AI without proper safeguards.

Strong AI, in contrast to weak AI, represents a machine with complete cognitive skills - and an equally extensive range of use cases - yet the complexity of completing such a feat has not diminished over time.

AI failure

Weak AI, narrow AI or specialised AI is a simulation of human intelligence applied to a tightly defined problem (for example, driving a car, transcribing human speech, or managing content on a website).

Weak AI is frequently intended to excel at a specific task. While these computers appear to be intelligent, they are subject to many more limits and limitations than even the most basic human intelligence.

Here are some examples of poor AI:

Siri, Alexa, and other artificial intelligence assistants

Automobiles that drive themselves

Google lookup

Bots that converse

Spam filters for email

Netflix recommendations

Deep learning vs. machine learning

Although "machine learning" and "deep learning" are frequently used in discussions about artificial intelligence, they are not interchangeable. Deep learning is a specific subject of AI known as machine learning.

Learning by machine

A computer feeds data to a machine learning algorithm, which then employs statistical methods to "learn" how to execute a task better and better without being coded particularly for that purpose.

ML algorithms, on the other hand, use historical data as input to anticipate new output values. To that purpose, machine learning (ML) includes both supervised (where the output for the input is known because of labelled data sets) and unsupervised (where the outputs are uncertain because of the use of unlabeled) learning.

Deep understanding

In deep learning, a particular type of machine learning, inputs are processed using a biologically inspired neural network architecture.

The neural networks include several hidden layers that process the data, allowing the machine to go "deep" in learning by creating connections and weighting the inputs for the best results.

The four forms of artificial intelligence

There are four broad categories of AI defined by the tasks they can successfully do. They are as follows:

Machines that react

Memory limitations

Psychology theory

Self-awareness

Machines that react

A reactive machine adheres to the most fundamental principles of artificial intelligence and, as the name implies, can only utilise its intellect to observe and react to the world around it.

A reactive computer cannot store memory, and so cannot make judgments in real time based on previous experience.

Because of their rapid observation of the world, reactive machines are only suitable for a few specific activities.

However, intentionally restricting a reactive machine's perception of the world offers advantages: This AI is more trustworthy and reliable, responding consistently to the same stimuli.

Reactive machine examples

Deep Blue, a chess-playing supercomputer built by IBM in the 1990s, defeated international grandmaster Gary Kasparov in one game.

Deep Blue could only recognise the pieces on a chessboard and know how each piece moves according to chess rules, recognise each piece's present position and choose the most logical move.

The machine did not anticipate its opponent's following movements or attempt to position its pieces better. Each move was seen as a separate reality, separate from all prior moves.

AlphaGo, developed by Google, is also incapable of predicting future moves, instead depending on its neural network to assess developments in the current game, allowing it to outperform Deep Blue in a more challenging game format.

AlphaGo has also defeated world-class opponents in this game, most recently, Go champion Lee Sedol in 2016.

Memory limitations

Memory limitations: As it gathers information and weighs prospective options, AI can store historical data and forecasts, effectively looking to the past for hints as to what might happen next.

AI with limited memory is more complicated and versatile than reactive machines.

Limited-memory AI occurs when a team regularly trains a model to analyse and use fresh data or when an AI ecosystem is developed in which models can be automatically taught and regenerated.

When employing restricted memory AI in ML, there are six stages to take:

Obtaining training data

Developing a machine-learning model

Ascertain that the model is capable of making predictions.

Check that the model can receive human or environmental feedback and save the feedback as data.

Repeat the preceding steps in a loop.

Mind-Body Concept

The Theory of Mind is merely theoretical. We still need to develop the technological and scientific capabilities required to reach this next level of artificial intelligence.

The psychological foundation is that other living beings have ideas and feelings that impact their conduct.

This would imply that AI robots are capable of reflective reasoning and may draw conclusions based on their observations of human, animal, and machine behaviour.

To create a two-way relationship between people and AI, robots would need to be able to grasp and interpret the concept of mind,' the fluctuations of emotions in decision-making, and a slew of other psychological concepts in real time.

Self-awareness

Once the theory of mind has been established, which will be the case in the Future of AI, the next stage will be for AI to become self-aware.

This sort of AI has human-level consciousness and is aware of its presence in the world, as well as the presence and emotional state of others.

It could comprehend what others require based not only on what they convey to it but also on how they communicate it.

To build self-awareness in robots, human researchers must first understand the preconditions of consciousness and then learn how to mimic it.

Artificial intelligence examples

Chatbots, navigation apps, and wearable fitness trackers are all examples of artificial intelligence technologies.

The examples below demonstrate the breadth of possible AI applications.

ChatGPT

ChatGPT is an AI-powered chatbot that can compose text in a number of different styles, from essays to code to simple question replies.

OpenAI launched ChatGPT in November 2022, relying on a large language model to accurately mimic human writing.

Since May 2023, ChatGPT has also been accessible as a mobile app for iOS devices and, since July 2023, for Android smartphones.

It's only one of many chatbot instances. However, it's a potent one.

Google Maps.

Google Maps monitors traffic flow and determines the shortest route using location data from cellphones and user-reported data about roadworks and car accidents.

Intelligent helpers

Natural language processing (NLP) is used by AI personal assistants such as Siri, Alexa, and Cortana to receive user commands, set reminders, search for internet information, and control lighting in the home.

In many situations, these assistants are designed to learn the user's preferences and improve the experience over time by making better suggestions and providing personalised responses.

Snapchat effects

Snapchat filters employ machine learning algorithms to differentiate between the subject of an image and the backdrop, track facial movements, and modify the image on the screen to the user's activities.

Automobiles that drive themselves

Self-driving cars are a well-known example of deep learning since they employ deep neural networks to recognise objects in their environment, assess their distance from other cars, detect traffic signals, and perform other tasks.

Wearables

Deep learning is also employed in wearable sensors and gadgets used in the healthcare business to measure the patient's health, such as blood glucose levels, blood pressure, and heart rate.

They can also use patterns derived from a patient's previous medical data to anticipate future health issues.

MuZero

MuZero, a DeepMind computer program, is a promising predecessor to natural artificial general intelligence.

It has mastered games it was never taught, such as chess and various Atari games, through raw force, playing games millions of times.

The Benefits of Artificial Intelligence

Artificial intelligence has numerous applications, ranging from accelerating vaccine development to automatically detecting potential fraud.

According to CB Insights, AI startups will earn 66.8 billion US dollars in funding in 2022, more than doubling the level in 2020. AI is creating waves in various areas due to its rapid adoption.

Safer banking

According to Business Insider Intelligence's analysis of artificial intelligence in banking in 2022, more than half of financial service providers currently utilise AI solutions for risk management and revenue generation.

Using AI in banking might result in savings of $400 billion.

Improved medicine

According to World Health Organization research from 2021, while integrating AI into healthcare poses problems, the technology has "promise" in benefits such as better-informed healthcare policies and increased accuracy in diagnosing patients.

Media that is cutting-edge

AI has also made an impression on the entertainment industry. According to Grand View Research, the global market for artificial intelligence in media and entertainment is expected to reach $99.48 billion by 2030, up from $10.87 billion in 2021.

AI applications such as plagiarism detection and the generation of high-resolution visuals are part of this expansion.

AI's Challenges and Limitations

Although AI is unquestionably regarded as a vital and quickly increasing resource, this new discipline has drawbacks.

The Pew Research Center surveyed 10,260 U.S. citizens in 2021 to learn how they felt about artificial intelligence. About half of the respondents (45%) are equally excited and concerned, while 37% are more worried than excited.

Furthermore, more than 40% of respondents believe driverless automobiles are dangerous for society.

The idea of employing AI to detect the spread of disinformation on social media, on the other hand, was greeted more positively: over 40% of respondents believed it was a good idea.

AI is a panacea for increasing production and efficiency while decreasing the possibility of human error.

However, there are also disadvantages, such as the high research costs and the risk of automated machines displacing human workers.

It is worth mentioning, however, that the artificial intelligence business will also create jobs, some of which have yet to be established.

Artificial Intelligence's Future

Given the computational expenses and technical data infrastructure required for artificial intelligence, its practical implementation is a complex and costly business.

Fortunately, tremendous breakthroughs in computer technology have occurred, as seen by Moore's Law, which claims that the number of transistors on a microchip typically doubles every two years while the cost of computers halves.

Although many experts expect Moore's Law to expire in the 2020s, it has had a significant impact on present AI approaches - deep learning would be financially unimaginable without it.

According to a recent study, AI innovation outpaces Moore's Law, doubling roughly every six months rather than two years.

According to this rationale, artificial intelligence has significantly progressed in many industries in recent years.

And the possibility of an even more significant influence in the following decades appears certain.