What is ar­ti­fi­cial in­tel­li­gence (AI)?

Ar­ti­fi­cial in­tel­li­gence is a branch of computer science that aims to create a tech­no­lo­gic­al equi­val­ent to human in­tel­li­gence. But what exactly is in­tel­li­gence and how can it be re­pro­duced using tech­no­logy? Numerous theories and meth­od­o­lo­gies have been developed to address these questions, however, es­tab­lish­ing a precise defin­i­tion of ar­ti­fi­cial in­tel­li­gence has proven difficult due to the complex nature of in­tel­li­gence itself.

How is AI defined? A look at different defin­i­tions

Most ar­ti­fi­cial in­tel­li­gence has been developed to carry out technical tasks. The focus has been less on mastering human com­mu­nic­a­tion, and more on per­form­ing highly spe­cial­ised tasks ef­fi­ciently. For these types of tech­no­lo­gies, a re­stric­ted Turing test is used to test if a system possesses the same abilities as humans in a specific field, for example, in medical dia­gnostics or chess. If it does, it’s con­sidered an ar­ti­fi­cially in­tel­li­gent system. This is just one type of ar­ti­fi­cial in­tel­li­gence though. A dis­tinc­tion is made between this type of ar­ti­fi­cial in­tel­li­gence, which is con­sidered ‘weak’ and another type which is con­sidered ‘strong’. Below, we’ll take a look at these two different defin­i­tions of AI.

The vision: Strong AI

The defin­i­tion of strong ar­ti­fi­cial in­tel­li­gence, also referred to as general AI, refers to a type of in­tel­li­gence that, due to its diverse cap­ab­il­it­ies, is in a position to replace humans. In­tel­li­gence has various di­men­sions, en­com­passing cognitive, sensory, motor, emotional and social cap­ab­il­it­ies. Most current ap­plic­a­tions of ar­ti­fi­cial in­tel­li­gence are in the area of cognitive in­tel­li­gence, i.e., logic, planning, problem-solving, self-suf­fi­ciency and per­spect­ive formation.

The reality: Weak AI

On the other hand, weak ar­ti­fi­cial in­tel­li­gence, also known as narrow AI, is defined dif­fer­ently and refers to the de­vel­op­ment and ap­plic­a­tion of ar­ti­fi­cial in­tel­li­gence for clearly defined use cases. This is the current state of ar­ti­fi­cial in­tel­li­gence. Nearly all of the current uses of ar­ti­fi­cial in­tel­li­gence can be defined as weak AI but also un­doubtedly spe­cial­ised. A good example of weak AI is the de­vel­op­ment of self-driving cars, AI for medical dia­gnostics and in­tel­li­gent search and auto­ma­tion al­gorithms.

Over the last few years, research has made ground­break­ing success in the area of weak AI. The de­vel­op­ment of in­tel­li­gent systems in in­di­vidu­al sectors has shown itself as not just immensely practical but also as less con­tro­ver­sial, ethically speaking than the research into su­per­in­tel­li­gence.

How does ar­ti­fi­cial in­tel­li­gence work?

How ar­ti­fi­cial in­tel­li­gence works depends on how knowledge is rep­res­en­ted within the AI system. There are two fun­da­ment­al ap­proaches to rep­res­ent­ing knowledge:

1. Symbolic AI: With this approach, knowledge is rep­res­en­ted by symbols and operates with symbol ma­nip­u­la­tion. Symbolic AI ap­proaches the pro­cessing of in­form­a­tion using a top-down approach, operating with symbols, abstract cor­rel­a­tions and logical keys.
2. Neural AI: With this approach, knowledge is depicted using ar­ti­fi­cial neurons and con­nect­ors. Neural AI ap­proaches the pro­cessing of in­form­a­tion from the bottom up, sim­u­lat­ing in­di­vidu­al ar­ti­fi­cial neurons, which organise them­selves into larger groups and together form an ar­ti­fi­cial neural network.

Symbolic AI

Symbolic AI is con­sidered the classical approach. It is grounded in the idea that human thought can be re­con­struc­ted from a higher-level framework based on logic and concepts and doesn’t need to rely on concrete ex­per­i­ences (top-down approach). Knowledge is rep­res­en­ted by abstract symbols, including written and spoken language. Machines learn to recognise, un­der­stand and use these symbols on the basis of al­gorithms. The in­tel­li­gent system retrieves its in­form­a­tion from expert systems.

Classic uses of symbolic AI are word pro­cessing and speech re­cog­ni­tion but it has also been used for other logical activ­it­ies like playing chess. Symbolic AI works based on set rules, and with in­creas­ing computing power, can solve problems of in­creas­ing com­plex­ity. With the help of symbolic AI, IBM’s Deep Blue was able to win a game of chess against Garry Kasparov, who was the world champion at the time.

Neural AI

In 1986, Geoffrey Hinton and two of his col­leagues revived research into neural AI and with it the research field of ar­ti­fi­cial in­tel­li­gence. The further de­vel­op­ment of the back­propaga­tion algorithm created the basis for deep learning, which nearly all AI works with these days. Thanks to this learning algorithm, deep neural networks can con­tinu­ally learn and grow by them­selves.

Neural ar­ti­fi­cial in­tel­li­gence splits up knowledge into tiny func­tion­al units known as ar­ti­fi­cial neurons. These neurons then form groups, which become in­creas­ingly larger (bottom-up approach), resulting in a diverse and branched network of ar­ti­fi­cial neurons. Unlike with symbolic AI, the neural network is trained. In robotics, for example, this is done with sensory motor data. With the help of machine learning, the AI generates a knowledge base that con­tinu­ously grows. And this is exactly where the big break­through happens. While this training requires a sig­ni­fic­ant amount of time, the system is now in a position to learn in­de­pend­ently.

What are some examples of ar­ti­fi­cial in­tel­li­gence?

Whether it’s facial re­cog­ni­tion, voice as­sist­ants or trans­la­tion software, AI has become a part of our everyday lives. Even if you con­sciously avoid using such tools, it’s difficult to escape the influence of ar­ti­fi­cial in­tel­li­gence in digital en­vir­on­ments. For example, AI systems play a sig­ni­fic­ant role in shaping the product re­com­mend­a­tions you receive from online stores as well as re­com­mend­a­tions from platforms like YouTube and Netflix. These systems are designed to provide you with sug­ges­tions that are in­creas­ingly tailored to your pref­er­ences.

Below are some examples of how ar­ti­fi­cial in­tel­li­gence is currently being used:

• ChatGPT: ChatGPT is an AI chatbot that was developed by Open AI. The large language model (LLM) can un­der­stand text inputs and answer questions as well as generate, rewrite and translate texts.
• RankBrain: RankBrain is an ar­ti­fi­cially in­tel­li­gent algorithm from Google that was ori­gin­ally developed to better un­der­stand search queries that may be unknown at the time of the first search. In 2015, Google announced that Rankbrain, after links and content, was the third most important factor of over 200 ranking factors in Google Search. This means that RankBrain has a big influence on SEO.
• DeepMind: Purchased by Google in 2014, DeepMind is a company that has created many in­nov­at­ive AI tech­no­lo­gies including AlphaGo, the computer program that mastered the board game Go. In April 2023, Google announced that they were merging the company with their in-house AI division, Google Brain. DeepMind has dis­tin­guished itself in the field of AI research by equipping their AI systems with short-term memory.
• DALL-E: The AI system DALL-E can create im­press­ive and unique 2D and 3D images from written input in a matter of seconds. The open beta version of OpenAI’s software has been available since September 2022. According to the de­vel­op­ment team, over two million images are created with the ap­plic­a­tion every day.
• Amazon’s Alexa and Apple’s Siri: AI as­sist­ants Alexa and Siri use voice control to help users with everyday tasks like re­triev­ing in­form­a­tion. Using speech synthesis, they can provide answers in natural language.

What op­por­tun­it­ies and risks does AI pose?

Pre­dic­tions about how AI will change our lives are both positive and negative. Below, we outline the key ad­vant­ages and dis­ad­vant­ages of AI as well as the op­por­tun­it­ies and risks as­so­ci­ated with it.

What are the ad­vant­ages and pos­sib­il­it­ies of AI?

There is a whole range of ad­vant­ages and pos­sib­il­it­ies when it comes to AI. The most important ad­vant­ages are un­doubtedly in the world of work, where it can be highly efficient and dra­mat­ic­ally improve economic prospects.

Job creation and reduced workload

AI could bring about valuable new jobs and in general, lead to an economic upsurge. One thing that all experts agree on is that the tech­no­logy will have a radical impact on the job market as a whole. The im­prove­ments and sim­pli­fic­a­tions that AI is capable of bringing about could also mean more free time for people.

Comfort

Sup­port­ers of AI view each technical ad­vance­ment as an op­por­tun­ity for greater ease and comfort in everyday life. Examples of this include self-driving cars and in­tel­li­gent trans­la­tion software. In general, such de­vel­op­ments make life con­sid­er­ably easier for consumers.

Ex­traordin­ary per­form­ance

When it comes to tasks for the greater public good, ar­ti­fi­cial in­tel­li­gence also provides sig­ni­fic­ant benefits. There is no denying the fact that machines have a much lower error rate than humans, and their per­form­ance potential is enormous. In the health­care and legal sectors, in par­tic­u­lar, the ver­sat­il­ity of in­tel­li­gent machines is seen as es­pe­cially promising. While experts don’t expect that judges will one day be replaced by machines, ar­ti­fi­cial in­tel­li­gence can help judges to more quickly recognise patterns in a court case and reach objective con­clu­sions.

Economic ad­vant­ages

There is also the promise of large financial gains for the in­dus­tries that are creating the tech­no­logy. The AI industry is ex­per­i­en­cing re­mark­able growth worldwide, with the Global Ar­ti­fi­cial In­tel­li­gence Markets Report citing that global funding for the industry had doubled in 2021, reaching $66.8 billion (around £53 billion). In the subfield of gen­er­at­ive AI, funding increased eightfold from 2022, reaching $25.2 billion (around £20 billion) in 2023.

Fu­tur­ist­ic projects

Last but not least, ar­ti­fi­cial in­tel­li­gence inspires the natural curiosity of humans. Already it’s being used for exploring oil sources and con­trolling robots on Mars. It’s safe to assume that the continued de­vel­op­ment of the tech­no­logy will lead to an increase in the number of fields and use cases that it can be used for.

What are the dis­ad­vant­ages and risks as­so­ci­ated with AI?

Prominent experts have warned about the risks of ar­ti­fi­cial in­tel­li­gence, despite being directly involved in its de­vel­op­ment. Such criticism has also found support amongst larger or­gan­isa­tions and ini­ti­at­ives. The Future of Life Institute (FLI), for example, regularly mobilises renowned critics to call for a re­spons­ible approach to tech­no­logy.

Here are just some of the risks as­so­ci­ated with ar­ti­fi­cial in­tel­li­gence:

Human in­feri­or­ity

One potential risk that many people fear, and which has often been a favourite subject of science fiction writers, is the de­vel­op­ment of a su­per­in­tel­li­gence. This term refers to a tech­no­logy that optimises itself to the point where it is no longer reliant on humans. Although most re­search­ers view an in­ten­tion­ally malicious AI as being highly unlikely, many view the pos­sib­il­ity of ar­ti­fi­cial in­tel­li­gence becoming competent enough to carry out activ­it­ies in­de­pend­ently as highly plausible.

Tech­no­lo­gic­al de­pend­ency

An ever-growing de­pend­ency on tech­no­logy is another cause for concern. One example is in the area of health­care, where the use of nurse robots is already being tested. In this context, humans are in­creas­ingly becoming the monitored subjects of tech­no­lo­gic­al systems. As a result, people may be in danger of losing some of their personal privacy and autonomy.

Data pro­tec­tion and the dis­tri­bu­tion of power

In­tel­li­gent al­gorithms are now able to process growing datasets more ef­fi­ciently than ever. This is par­tic­u­larly good news for the online retail sector. However, the pro­cessing of data through AI tech­no­logy is becoming more and more difficult for consumers to un­der­stand and keep track of.

Filter bubbles and selective per­cep­tion

Online activist Eli Pariser has drawn attention to what he sees as a further risk of ar­ti­fi­cial in­tel­li­gence: filter bubbles. If al­gorithms only show content to a user based on their previous online behavior (per­son­al­ised content), it is very likely that their view of the world will get narrower and narrower. Or at least this is the concern. AI tech­no­lo­gies could promote selective per­cep­tion, re­in­for­cing a growing distance between in­di­vidu­als who have different ideo­lo­gic­al views.

Influence how opinions are formed

Ad­di­tion­ally, AI tech­no­lo­gies have the ability to control public opinion. The reason for this sort of thinking is the existence of tech­no­lo­gies that have very detailed in­form­a­tion on their users, as well as the presence of social bots that can influence public dis­cus­sions.