Artificial intelligence (AI), in its broadest sense, is intelligence exhibited

by machines, particularly computer systems. It is a field of research in computer

science that develops and studies methods and software that enable machines
to perceive their environment and use learning and intelligence to take actions that
maximize their chances of achieving defined goals.[1] Such machines may be called
AIs.

Some high-profile applications of AI include advanced web search

engines (e.g., Google Search); recommendation systems (used
by YouTube, Amazon, and Netflix); interacting via human speech (e.g., Google
Assistant, Siri, and Alexa); autonomous
vehicles (e.g., Waymo); generative and creative tools (e.g., ChatGPT, Apple
Intelligence, and AI art); and superhuman play and analysis in strategy
games (e.g., chess and Go).[2] However, many AI applications are not perceived as
AI: "A lot of cutting edge AI has filtered into general applications, often without being
called AI because once something becomes useful enough and common enough
it's not labeled AI anymore."[3][4]

Alan Turing was the first person to conduct substantial research in the field that he
called machine intelligence.[5] Artificial intelligence was founded as an academic
discipline in 1956,[6] by those now considered the founding fathers of AI, John
McCarthy, Marvin Minksy, Nathaniel Rochester, and Claude Shannon.[7][8] The field
went through multiple cycles of optimism,[9][10] followed by periods of disappointment
and loss of funding, known as AI winter.[11][12] Funding and interest vastly increased
after 2012 when deep learning surpassed all previous AI techniques,[13] and after
2017 with the transformer architecture.[14] This led to the AI boom of the early 2020s,
with companies, universities, and laboratories overwhelmingly based in the United
States pioneering significant advances in artificial intelligence.[15]

The growing use of artificial intelligence in the 21st century is influencing a societal
and economic shift towards increased automation, data-driven decision-making, and
the integration of AI systems into various economic sectors and areas of
life, impacting job
markets, healthcare, government, industry, education, propaganda,
and disinformation. This raises questions about the long-term effects, ethical
implications, and risks of AI, prompting discussions about regulatory policies to
ensure the safety and benefits of the technology.

The various subfields of AI research are centered around particular goals and the
use of particular tools. The traditional goals of AI research
include reasoning, knowledge representation, planning, learning, natural language
processing, perception, and support for robotics.[a] General intelligence—the ability to
complete any task performable by a human on an at least equal level—is among the
field's long-term goals.[16]

To reach these goals, AI researchers have adapted and integrated a wide range of
techniques, including search and mathematical optimization, formal logic, artificial
neural networks, and methods based on statistics, operations research,
and economics.[b] AI also draws
upon psychology, linguistics, philosophy, neuroscience, and other fields.[17]
Goals
The general problem of simulating (or creating) intelligence has been broken into
subproblems. These consist of particular traits or capabilities that researchers expect
an intelligent system to display. The traits described below have received the most
attention and cover the scope of AI research.[a]

Reasoning and problem-solving

Early researchers developed algorithms that imitated step-by-step reasoning that
humans use when they solve puzzles or make logical deductions.[18] By the late
1980s and 1990s, methods were developed for dealing with uncertain or incomplete
information, employing concepts from probability and economics.[19]

Many of these algorithms are insufficient for solving large reasoning problems
because they experience a "combinatorial explosion": They become exponentially
slower as the problems grow.[20] Even humans rarely use the step-by-step deduction
that early AI research could model. They solve most of their problems using fast,
intuitive judgments.[21] Accurate and efficient reasoning is an unsolved problem.

Knowledge representation

An ontology represents knowledge as a set of

concepts within a domain and the relationships between those concepts
Knowledge representation and knowledge engineering[22] allow AI programs to
answer questions intelligently and make deductions about real-world facts. Formal
knowledge representations are used in content-based indexing and retrieval,[23] scene
interpretation,[24] clinical decision support,[25] knowledge discovery (mining "interesting"
and actionable inferences from large databases),[26] and other areas.[27]

A knowledge base is a body of knowledge represented in a form that can be used by

a program. An ontology is the set of objects, relations, concepts, and properties used
by a particular domain of knowledge.[28] Knowledge bases need to represent things
such as objects, properties, categories, and relations between objects;[29] situations,
events, states, and time;[30] causes and effects;[31] knowledge about knowledge (what
we know about what other people know);[32] default reasoning (things that humans
assume are true until they are told differently and will remain true even when other
facts are changing);[33] and many other aspects and domains of knowledge.
Among the most difficult problems in knowledge representation are the breadth of
commonsense knowledge (the set of atomic facts that the average person knows is
enormous);[34] and the sub-symbolic form of most commonsense knowledge (much of
what people know is not represented as "facts" or "statements" that they could
express verbally).[21] There is also the difficulty of knowledge acquisition, the problem
of obtaining knowledge for AI applications.[c]

Planning and decision-making

An "agent" is anything that perceives and takes actions in the world. A rational
agent has goals or preferences and takes actions to make them happen.[d]
[37]
In automated planning, the agent has a specific goal.[38] In automated decision-
making, the agent has preferences—there are some situations it would prefer to be
in, and some situations it is trying to avoid. The decision-making agent assigns a
number to each situation (called the "utility") that measures how much the agent
prefers it. For each possible action, it can calculate the "expected utility": the utility of
all possible outcomes of the action, weighted by the probability that the outcome will
occur. It can then choose the action with the maximum expected utility.[39]

In classical planning, the agent knows exactly what the effect of any action will be.
[40]
In most real-world problems, however, the agent may not be certain about the
situation they are in (it is "unknown" or "unobservable") and it may not know for
certain what will happen after each possible action (it is not "deterministic"). It must
choose an action by making a probabilistic guess and then reassess the situation to
see if the action worked.[41]

In some problems, the agent's preferences may be uncertain, especially if there are
other agents or humans involved. These can be learned (e.g., with inverse
reinforcement learning), or the agent can seek information to improve its
preferences.[42] Information value theory can be used to weigh the value of
exploratory or experimental actions.[43] The space of possible future actions and
situations is typically intractably large, so the agents must take actions and evaluate
situations while being uncertain of what the outcome will be.

A Markov decision process has a transition model that describes the probability that
a particular action will change the state in a particular way and a reward function that
supplies the utility of each state and the cost of each action. A policy associates a
decision with each possible state. The policy could be calculated (e.g., by iteration),
be heuristic, or it can be learned.[44]

Game theory describes the rational behavior of multiple interacting agents and is
used in AI programs that make decisions that involve other agents.[45]

Learning
Machine learning is the study of programs that can improve their performance on a
given task automatically.[46] It has been a part of AI from the beginning.[e]

There are several kinds of machine learning. Unsupervised learning analyzes a

stream of data and finds patterns and makes predictions without any other guidance.
[49]
Supervised learning requires a human to label the input data first, and comes in
two main varieties: classification (where the program must learn to predict what
category the input belongs in) and regression (where the program must deduce a
numeric function based on numeric input).[50]

In reinforcement learning, the agent is rewarded for good responses and punished
for bad ones. The agent learns to choose responses that are classified as "good".
[51]
Transfer learning is when the knowledge gained from one problem is applied to a
new problem.[52] Deep learning is a type of machine learning that runs inputs through
biologically inspired artificial neural networks for all of these types of learning.[53]

Computational learning theory can assess learners by computational complexity,

by sample complexity (how much data is required), or by other notions
of optimization.[54]

Natural language processing

Natural language processing (NLP)[55] allows programs to read, write and
communicate in human languages such as English. Specific problems
include speech recognition, speech synthesis, machine translation, information
extraction, information retrieval and question answering.[56]

Early work, based on Noam Chomsky's generative grammar and semantic networks,
had difficulty with word-sense disambiguation[f] unless restricted to small domains
called "micro-worlds" (due to the common sense knowledge problem[34]). Margaret
Masterman believed that it was meaning and not grammar that was the key to
understanding languages, and that thesauri and not dictionaries should be the basis
of computational language structure.

Modern deep learning techniques for NLP include word embedding (representing
words, typically as vectors encoding their meaning),[57] transformers (a deep learning
architecture using an attention mechanism),[58] and others.[59] In 2019, generative pre-
trained transformer (or "GPT") language models began to generate coherent text,[60]
[61]
and by 2023 these models were able to get human-level scores on the bar
exam, SAT test, GRE test, and many other real-world applications.[62]

Perception
Machine perception is the ability to use input from sensors (such as cameras,
microphones, wireless signals, active lidar, sonar, radar, and tactile sensors) to
deduce aspects of the world. Computer vision is the ability to analyze visual input.[63]

The field includes speech recognition,[64] image classification,[65] facial

recognition, object recognition,[66]object tracking,[67] and robotic perception.[68]

Social intelligence
 Kismet, a robot head which was made in the 1990s; a
machine that can recognize and simulate emotions[69]
Affective computing is an interdisciplinary umbrella that comprises systems that
recognize, interpret, process, or simulate human feeling, emotion, and mood.[70] For
example, some virtual assistants are programmed to speak conversationally or even
to banter humorously; it makes them appear more sensitive to the emotional
dynamics of human interaction, or to otherwise facilitate human–computer
interaction.

However, this tends to give naïve users an unrealistic conception of the intelligence
of existing computer agents.[71] Moderate successes related to affective computing
include textual sentiment analysis and, more recently, multimodal sentiment
analysis, wherein AI classifies the affects displayed by a videotaped subject.[72]

General intelligence
A machine with artificial general intelligence should be able to solve a wide variety of
problems with breadth and versatility similar to human intelligence.[16]

Techniques
AI research uses a wide variety of techniques to accomplish the goals above.[b]

Search and optimization

AI can solve many problems by intelligently searching through many possible
solutions.[73] There are two very different kinds of search used in AI: state space
search and local search.

State space search

State space search searches through a tree of possible states to try to find a goal
state.[74] For example, planning algorithms search through trees of goals and
subgoals, attempting to find a path to a target goal, a process called means-ends
analysis.[75]

Simple exhaustive searches[76] are rarely sufficient for most real-world problems:
the search space (the number of places to search) quickly grows to astronomical
numbers. The result is a search that is too slow or never completes.[20] "Heuristics" or
"rules of thumb" can help prioritize choices that are more likely to reach a goal. [77]

Adversarial search is used for game-playing programs, such as chess or Go. It

searches through a tree of possible moves and counter-moves, looking for a winning
position.[78]
Local search

Illustration of gradient descent for 3 different starting

points; two parameters (represented by the plan coordinates) are adjusted in order to
minimize the loss function (the height)
Local search uses mathematical optimization to find a solution to a problem. It
begins with some form of guess and refines it incrementally.[79]

Gradient descent is a type of local search that optimizes a set of numerical

parameters by incrementally adjusting them to minimize a loss function. Variants
of gradient descent are commonly used to train neural networks.[80]

Another type of local search is evolutionary computation, which aims to iteratively

improve a set of candidate solutions by "mutating" and "recombining"
them, selecting only the fittest to survive each generation.[81]

Distributed search processes can coordinate via swarm intelligence algorithms. Two
popular swarm algorithms used in search are particle swarm optimization (inspired
by bird flocking) and ant colony optimization (inspired by ant trails).[82]

Logic
Formal logic is used for reasoning and knowledge representation.[83] Formal logic
comes in two main forms: propositional logic (which operates on statements that are
true or false and uses logical connectives such as "and", "or", "not" and "implies")
[84]
and predicate logic (which also operates on objects, predicates and relations and
uses quantifiers such as "Every X is a Y" and "There are some Xs that are Ys").[85]

Deductive reasoning in logic is the process of proving a new statement (conclusion)

from other statements that are given and assumed to be true (the premises).
[86]
Proofs can be structured as proof trees, in which nodes are labelled by sentences,
and children nodes are connected to parent nodes by inference rules.

Given a problem and a set of premises, problem-solving reduces to searching for a

proof tree whose root node is labelled by a solution of the problem and whose leaf
nodes are labelled by premises or axioms. In the case of Horn clauses, problem-
solving search can be performed by reasoning forwards from the premises
or backwards from the problem.[87] In the more general case of the clausal form
of first-order logic, resolution is a single, axiom-free rule of inference, in which a
problem is solved by proving a contradiction from premises that include the negation
of the problem to be solved.[88]
Inference in both Horn clause logic and first-order logic is undecidable, and
therefore intractable. However, backward reasoning with Horn clauses, which
underpins computation in the logic programming language Prolog, is Turing
complete. Moreover, its efficiency is competitive with computation in other symbolic
programming languages.[89]

Fuzzy logic assigns a "degree of truth" between 0 and 1. It can therefore handle
propositions that are vague and partially true.[90]

Non-monotonic logics, including logic programming with negation as failure, are

designed to handle default reasoning.[33] Other specialized versions of logic have
been developed to describe many complex domains.

Probabilistic methods for uncertain reasoning

A simple Bayesian network, with

the associated conditional probability tables
Many problems in AI (including in reasoning, planning, learning, perception, and
robotics) require the agent to operate with incomplete or uncertain information. AI
researchers have devised a number of tools to solve these problems using methods
from probability theory and economics.[91] Precise mathematical tools have been
developed that analyze how an agent can make choices and plan, using decision
theory, decision analysis,[92] and information value theory.[93] These tools include
models such as Markov decision processes,[94] dynamic decision networks,[95] game
theory and mechanism design.[96]

Bayesian networks[97] are a tool that can be used for reasoning (using the Bayesian
inference algorithm),[g][99] learning (using the expectation–maximization algorithm),[h]
[101]
planning (using decision networks)[102] and perception (using dynamic Bayesian
networks).[95]

Probabilistic algorithms can also be used for filtering, prediction, smoothing, and
finding explanations for streams of data, thus helping perception systems analyze
processes that occur over time (e.g., hidden Markov models or Kalman filters).[95]
 Expectation–maximization clustering of Old
Faithful eruption data starts from a random guess but then successfully converges on an
accurate clustering of the two physically distinct modes of eruption
Classifiers and statistical learning methods
The simplest AI applications can be divided into two types: classifiers (e.g., "if shiny
then diamond"), on one hand, and controllers (e.g., "if diamond then pick up"), on the
other hand. Classifiers[103] are functions that use pattern matching to determine the
closest match. They can be fine-tuned based on chosen examples using supervised
learning. Each pattern (also called an "observation") is labeled with a certain
predefined class. All the observations combined with their class labels are known as
a data set. When a new observation is received, that observation is classified based
on previous experience.[50]

There are many kinds of classifiers in use. The decision tree is the simplest and
most widely used symbolic machine learning algorithm.[104] K-nearest
neighbor algorithm was the most widely used analogical AI until the mid-1990s,
and Kernel methods such as the support vector machine (SVM) displaced k-nearest
neighbor in the 1990s.[105] The naive Bayes classifier is reportedly the "most widely
used learner"[106] at Google, due in part to its scalability.[107] Neural networks are also
used as classifiers.[108]

Artificial neural networks

A neural network is an interconnected group of nodes, akin

to the vast network of neurons in the human brain
An artificial neural network is based on a collection of nodes also known as artificial
neurons, which loosely model the neurons in a biological brain. It is trained to
recognise patterns; once trained, it can recognise those patterns in fresh data. There
is an input, at least one hidden layer of nodes and an output. Each node applies a
function and once the weight crosses its specified threshold, the data is transmitted
to the next layer. A network is typically called a deep neural network if it has at least
2 hidden layers.[108]

Learning algorithms for neural networks use local search to choose the weights that
will get the right output for each input during training. The most common training
technique is the backpropagation algorithm.[109] Neural networks learn to model
complex relationships between inputs and outputs and find patterns in data. In
theory, a neural network can learn any function.[110]

In feedforward neural networks the signal passes in only one direction.[111] Recurrent
neural networks feed the output signal back into the input, which allows short-term
memories of previous input events. Long short term memory is the most successful
network architecture for recurrent networks.[112] Perceptrons[113] use only a single layer
of neurons, deep learning[114] uses multiple layers. Convolutional neural
networks strengthen the connection between neurons that are "close" to each other
—this is especially important in image processing, where a local set of neurons
must identify an "edge" before the network can identify an object.[115]

Deep learning

Deep learning[114] uses several layers of neurons between the network's inputs and
outputs. The multiple layers can progressively extract higher-level features from the
raw input. For example, in image processing, lower layers may identify edges, while
higher layers may identify the concepts relevant to a human such as digits, letters, or
faces.[116]

Deep learning has profoundly improved the performance of programs in many

important subfields of artificial intelligence, including computer vision, speech
recognition, natural language processing, image classification,[117] and others. The
reason that deep learning performs so well in so many applications is not known as
of 2023.[118] The sudden success of deep learning in 2012–2015 did not occur
because of some new discovery or theoretical breakthrough (deep neural networks
and backpropagation had been described by many people, as far back as the 1950s)
[i]
but because of two factors: the incredible increase in computer power (including the
hundred-fold increase in speed by switching to GPUs) and the availability of vast
amounts of training data, especially the giant curated datasets used for benchmark
testing, such as ImageNet.[j]

GPT
Generative pre-trained transformers (GPT) are large language models that are
based on the semantic relationships between words in sentences (natural language
processing). Text-based GPT models are pre-trained on a large corpus of text which
can be from the internet. The pre-training consists in predicting the next token (a
token being usually a word, subword, or punctuation). Throughout this pre-training,
GPT models accumulate knowledge about the world, and can then generate human-
like text by repeatedly predicting the next token. Typically, a subsequent training
phase makes the model more truthful, useful and harmless, usually with a technique
called reinforcement learning from human feedback (RLHF). Current GPT models
are still prone to generating falsehoods called "hallucinations", although this can be
reduced with RLHF and quality data. They are used in chatbots, which allow you to
ask a question or request a task in simple text.[127][128]

Current models and services include: Gemini (formerly

Bard), ChatGPT, Grok, Claude, Copilot and LLaMA.[129] Multimodal GPT models can
process different types of data (modalities) such as images, videos, sound, and text.
[130]

Specialized hardware and software

Main articles: Programming languages for artificial intelligence and Hardware for
artificial intelligence
In the late 2010s, graphics processing units (GPUs) that were increasingly designed
with AI-specific enhancements and used with specialized TensorFlow software had
replaced previously used central processing unit (CPUs) as the dominant means for
large-scale (commercial and academic) machine learning models' training.
[131]
Specialized programming languages such as Prolog were used in early AI
research,[132] but general-purpose programming languages like Python have become
predominant.[133]

Applications
Main article: Applications of artificial intelligence
AI and machine learning technology is used in most of the essential applications of
the 2020s, including: search engines (such as Google Search), targeting online
advertisements, recommendation systems (offered by Netflix, YouTube or Amazon),
driving internet traffic, targeted advertising (AdSense, Facebook), virtual
assistants (such as Siri or Alexa), autonomous
vehicles (including drones, ADAS and self-driving cars), automatic language
translation (Microsoft Translator, Google Translate), facial recognition (Apple's Face
ID or Microsoft's DeepFace and Google's FaceNet) and image labeling (used
by Facebook, Apple's iPhoto and TikTok). The deployment of AI may be overseen by
a Chief automation officer (CAO).

Health and medicine

Main article: Artificial intelligence in healthcare
The application of AI in medicine and medical research has the potential to increase
patient care and quality of life.[134] Through the lens of the Hippocratic Oath, medical
professionals are ethically compelled to use AI, if applications can more accurately
diagnose and treat patients.
For medical research, AI is an important tool for processing and integrating big data.
This is particularly important for organoid and tissue engineering development which
use microscopy imaging as a key technique in fabrication.[135] It has been suggested
that AI can overcome discrepancies in funding allocated to different fields of
research.[135] New AI tools can deepen the understanding of biomedically relevant
pathways. For example, AlphaFold 2 (2021) demonstrated the ability to approximate,
in hours rather than months, the 3D structure of a protein.[136] In 2023, it was reported
that AI-guided drug discovery helped find a class of antibiotics capable of killing two
different types of drug-resistant bacteria.[137] In 2024, researchers used machine
learning to accelerate the search for Parkinson's disease drug treatments. Their aim
was to identify compounds that block the clumping, or aggregation, of alpha-
synuclein (the protein that characterises Parkinson's disease). They were able to
speed up the initial screening process ten-fold and reduce the cost by a thousand-
fold.[138][139]

Games
Main article: Game artificial intelligence
Game playing programs have been used since the 1950s to demonstrate and test
AI's most advanced techniques.[140] Deep Blue became the first computer chess-
playing system to beat a reigning world chess champion, Garry Kasparov, on 11 May
1997.[141] In 2011, in a Jeopardy! quiz show exhibition match, IBM's question
answering system, Watson, defeated the two greatest Jeopardy! champions, Brad
Rutter and Ken Jennings, by a significant margin.[142] In March 2016, AlphaGo won 4
out of 5 games of Go in a match with Go champion Lee Sedol, becoming the
first computer Go-playing system to beat a professional Go player
without handicaps. Then in 2017 it defeated Ke Jie, who was the best Go player in
the world.[143] Other programs handle imperfect-information games, such as
the poker-playing program Pluribus.[144] DeepMind developed increasingly
generalistic reinforcement learning models, such as with MuZero, which could be
trained to play chess, Go, or Atari games.[145] In 2019, DeepMind's AlphaStar
achieved grandmaster level in StarCraft II, a particularly challenging real-time
strategy game that involves incomplete knowledge of what happens on the map.
[146]
In 2021, an AI agent competed in a PlayStation Gran Turismo competition,
winning against four of the world's best Gran Turismo drivers using deep
reinforcement learning.[147]

Finance
Finance is one of the fastest growing sectors where applied AI tools are being
deployed: from retail online banking to investment advice and insurance, where
automated "robot advisers" have been in use for some years. [148]

World Pensions experts like Nicolas Firzli insist it may be too early to see the
emergence of highly innovative AI-informed financial products and services: "the
deployment of AI tools will simply further automatise things: destroying tens of
thousands of jobs in banking, financial planning, and pension advice in the process,
but I’m not sure it will unleash a new wave of [e.g., sophisticated] pension
innovation."[149]

Military
Main article: Military artificial intelligence
Various countries are deploying AI military applications.[150] The main applications
enhance command and control, communications, sensors, integration and
interoperability.[151] Research is targeting intelligence collection and analysis, logistics,
cyber operations, information operations, and semiautonomous and autonomous
vehicles.[150] AI technologies enable coordination of sensors and effectors, threat
detection and identification, marking of enemy positions, target acquisition,
coordination and deconfliction of distributed Joint Fires between networked combat
vehicles involving manned and unmanned teams.[151] AI was incorporated into military
operations in Iraq and Syria.[150]

In November 2023, US Vice President Kamala Harris disclosed a declaration signed

by 31 nations to set guardrails for the military use of AI. The commitments include
using legal reviews to ensure the compliance of military AI with international laws,
and being cautious and transparent in the development of this technology.[152]

Generative AI
Main article: Generative artificial intelligence

Vincent van Gogh in watercolour created by generative AI

software
In the early 2020s, generative AI gained widespread prominence. In March 2023,
58% of U.S. adults had heard about ChatGPT and 14% had tried it.[153] The increasing
realism and ease-of-use of AI-based text-to-image generators such
as Midjourney, DALL-E, and Stable Diffusion sparked a trend of viral AI-generated
photos. Widespread attention was gained by a fake photo of Pope Francis wearing a
white puffer coat, the fictional arrest of Donald Trump, and a hoax of an attack on
the Pentagon, as well as the usage in professional creative arts.[154][155]

Other industry-specific tasks

There are also thousands of successful AI applications used to solve specific
problems for specific industries or institutions. In a 2017 survey, one in five
companies reported having incorporated "AI" in some offerings or processes.[156] A
few examples are energy storage, medical diagnosis, military logistics, applications
that predict the result of judicial decisions, foreign policy, or supply chain
management.
In agriculture, AI has helped farmers identify areas that need irrigation, fertilization,
pesticide treatments or increasing yield. Agronomists use AI to conduct research and
development. AI has been used to predict the ripening time for crops such as
tomatoes, monitor soil moisture, operate agricultural robots, conduct predictive
analytics, classify livestock pig call emotions, automate greenhouses, detect
diseases and pests, and save water.

Artificial intelligence is used in astronomy to analyze increasing amounts of available

data and applications, mainly for "classification, regression, clustering, forecasting,
generation, discovery, and the development of new scientific insights" for example
for discovering exoplanets, forecasting solar activity, and distinguishing between
signals and instrumental effects in gravitational wave astronomy. It could also be
used for activities in space such as space exploration, including analysis of data from
space missions, real-time science decisions of spacecraft, space debris avoidance,
and more autonomous operation.

Ethics
Main article: Ethics of artificial intelligence
AI has potential benefits and potential risks. AI may be able to advance science and
find solutions for serious problems: Demis Hassabis of Deep Mind hopes to "solve
intelligence, and then use that to solve everything else".[157] However, as the use of AI
has become widespread, several unintended consequences and risks have been
identified.[158] In-production systems can sometimes not factor ethics and bias into
their AI training processes, especially when the AI algorithms are inherently
unexplainable in deep learning.[159]

Risks and harm

Privacy and copyright
Further information: Information privacy and Artificial intelligence and copyright
Machine-learning algorithms require large amounts of data. The techniques used to
acquire this data have raised concerns about privacy, surveillance and copyright.

Technology companies collect a wide range of data from their users, including online
activity, geolocation data, video and audio.[160] For example, in order to build speech
recognition algorithms, Amazon has recorded millions of private conversations and
allowed temporary workers to listen to and transcribe some of them.[161] Opinions
about this widespread surveillance range from those who see it as a necessary
evil to those for whom it is clearly unethical and a violation of the right to privacy.[162]

AI developers argue that this is the only way to deliver valuable applications. and
have developed several techniques that attempt to preserve privacy while still
obtaining the data, such as data aggregation, de-identification and differential
privacy.[163] Since 2016, some privacy experts, such as Cynthia Dwork, have begun to
view privacy in terms of fairness. Brian Christian wrote that experts have pivoted
"from the question of 'what they know' to the question of 'what they're doing with
it'."[164]

Generative AI is often trained on unlicensed copyrighted works, including in domains

such as images or computer code; the output is then used under the rationale of "fair
use". Experts disagree about how well and under what circumstances this rationale
will hold up in courts of law; relevant factors may include "the purpose and character
of the use of the copyrighted work" and "the effect upon the potential market for the
copyrighted work".[165][166] Website owners who do not wish to have their content
scraped can indicate it in a "robots.txt" file.[167] In 2023, leading authors
(including John Grisham and Jonathan Franzen) sued AI companies for using their
work to train generative AI.[168][169] Another discussed approach is to envision a
separate sui generis system of protection for creations generated by AI to ensure fair
attribution and compensation for human authors.[170]

Misinformation
See also: YouTube § Moderation and offensive content
YouTube, Facebook and others use recommender systems to guide users to more
content. These AI programs were given the goal of maximizing user engagement
(that is, the only goal was to keep people watching). The AI learned that users
tended to choose misinformation, conspiracy theories, and extreme partisan content,
and, to keep them watching, the AI recommended more of it. Users also tended to
watch more content on the same subject, so the AI led people into filter
bubbles where they received multiple versions of the same misinformation.[171] This
convinced many users that the misinformation was true, and ultimately undermined
trust in institutions, the media and the government.[172] The AI program had correctly
learned to maximize its goal, but the result was harmful to society. After the U.S.
election in 2016, major technology companies took steps to mitigate the problem [citation
needed]
.

In 2022, generative AI began to create images, audio, video and text that are
indistinguishable from real photographs, recordings, films, or human writing. It is
possible for bad actors to use this technology to create massive amounts of
misinformation or propaganda.[173] AI pioneer Geoffrey Hinton expressed concern
about AI enabling "authoritarian leaders to manipulate their electorates" on a large
scale, among other risks.[174]

Algorithmic bias and fairness

Main articles: Algorithmic bias and Fairness (machine learning)
Machine learning applications will be biased if they learn from biased data.[175] The
developers may not be aware that the bias exists.[176] Bias can be introduced by the
way training data is selected and by the way a model is deployed.[177][175] If a biased
algorithm is used to make decisions that can seriously harm people (as it can
in medicine, finance, recruitment, housing or policing) then the algorithm may
cause discrimination.[178] Fairness in machine learning is the study of how to prevent
the harm caused by algorithmic bias. It has become serious area of academic study
within AI. Researchers have discovered it is not always possible to define "fairness"
in a way that satisfies all stakeholders.[179]

On June 28, 2015, Google Photos's new image labeling feature mistakenly identified
Jacky Alcine and a friend as "gorillas" because they were black. The system was
trained on a dataset that contained very few images of black people,[180] a problem
called "sample size disparity".[181] Google "fixed" this problem by preventing the
system from labelling anything as a "gorilla". Eight years later, in 2023, Google
Photos still could not identify a gorilla, and neither could similar products from Apple,
Facebook, Microsoft and Amazon.[182]

COMPAS is a commercial program widely used by U.S. courts to assess the

likelihood of a defendant becoming a recidivist. In 2016, Julia
Angwin at ProPublica discovered that COMPAS exhibited racial bias, despite the fact
that the program was not told the races of the defendants. Although the error rate for
both whites and blacks was calibrated equal at exactly 61%, the errors for each race
were different—the system consistently overestimated the chance that a black
person would re-offend and would underestimate the chance that a white person
would not re-offend.[183] In 2017, several researchers[k] showed that it was
mathematically impossible for COMPAS to accommodate all possible measures of
fairness when the base rates of re-offense were different for whites and blacks in the
data.[185]

A program can make biased decisions even if the data does not explicitly mention a
problematic feature (such as "race" or "gender"). The feature will correlate with other
features (like "address", "shopping history" or "first name"), and the program will
make the same decisions based on these features as it would on "race" or "gender".
[186]
Moritz Hardt said "the most robust fact in this research area is that fairness
through blindness doesn't work."[187]

Criticism of COMPAS highlighted that machine learning models are designed to

make "predictions" that are only valid if we assume that the future will resemble the
past. If they are trained on data that includes the results of racist decisions in the
past, machine learning models must predict that racist decisions will be made in the
future. If an application then uses these predictions as recommendations, some of
these "recommendations" will likely be racist.[188] Thus, machine learning is not well
suited to help make decisions in areas where there is hope that the future will
be better than the past. It is descriptive rather than prescriptive.[l]

Bias and unfairness may go undetected because the developers are overwhelmingly
white and male: among AI engineers, about 4% are black and 20% are women.[181]

At its 2022 Conference on Fairness, Accountability, and Transparency (ACM FAccT

2022), the Association for Computing Machinery, in Seoul, South Korea, presented
and published findings that recommend that until AI and robotics systems are
demonstrated to be free of bias mistakes, they are unsafe, and the use of self-
learning neural networks trained on vast, unregulated sources of flawed internet data
should be curtailed.[190]

Lack of transparency
See also: Explainable AI, Algorithmic transparency, and Right to explanation
Many AI systems are so complex that their designers cannot explain how they reach
their decisions.[191] Particularly with deep neural networks, in which there are a large
amount of non-linear relationships between inputs and outputs. But some popular
explainability techniques exist.[192]

It is impossible to be certain that a program is operating correctly if no one knows

how exactly it works. There have been many cases where a machine learning
program passed rigorous tests, but nevertheless learned something different than
what the programmers intended. For example, a system that could identify skin
diseases better than medical professionals was found to actually have a strong
tendency to classify images with a ruler as "cancerous", because pictures of
malignancies typically include a ruler to show the scale.[193] Another machine learning
system designed to help effectively allocate medical resources was found to classify
patients with asthma as being at "low risk" of dying from pneumonia. Having asthma
is actually a severe risk factor, but since the patients having asthma would usually
get much more medical care, they were relatively unlikely to die according to the
training data. The correlation between asthma and low risk of dying from pneumonia
was real, but misleading.[194]

People who have been harmed by an algorithm's decision have a right to an

explanation.[195] Doctors, for example, are expected to clearly and completely explain
to their colleagues the reasoning behind any decision they make. Early drafts of the
European Union's General Data Protection Regulation in 2016 included an explicit
statement that this right exists.[m] Industry experts noted that this is an unsolved
problem with no solution in sight. Regulators argued that nevertheless the harm is
real: if the problem has no solution, the tools should not be used.[196]

DARPA established the XAI ("Explainable Artificial Intelligence") program in 2014 to

try and solve these problems.[197]

There are several possible solutions to the transparency problem. SHAP tried to
solve the transparency problems by visualising the contribution of each feature to the
output.[198] LIME can locally approximate a model with a simpler, interpretable model.
[199]
Multitask learning provides a large number of outputs in addition to the target
classification. These other outputs can help developers deduce what the network
has learned.[200] Deconvolution, DeepDream and other generative methods can allow
developers to see what different layers of a deep network have learned and produce
output that can suggest what the network is learning.[201]

Bad actors and weaponized AI

Main articles: Lethal autonomous weapon, Artificial intelligence arms race, and AI
safety
Artificial intelligence provides a number of tools that are useful to bad actors, such
as authoritarian governments, terrorists, criminals or rogue states.

A lethal autonomous weapon is a machine that locates, selects and engages human
targets without human supervision.[n] Widely available AI tools can be used by bad
actors to develop inexpensive autonomous weapons and, if produced at scale, they
are potentially weapons of mass destruction.[203] Even when used in conventional
warfare, it is unlikely that they will be unable to reliably choose targets and could
potentially kill an innocent person.[203] In 2014, 30 nations (including China) supported
a ban on autonomous weapons under the United Nations' Convention on Certain
Conventional Weapons, however the United States and others disagreed.[204] By
2015, over fifty countries were reported to be researching battlefield robots.[205]

AI tools make it easier for authoritarian governments to efficiently control their

citizens in several ways. Face and voice recognition allow
widespread surveillance. Machine learning, operating this data, can classify potential
enemies of the state and prevent them from hiding. Recommendation systems can
precisely target propaganda and misinformation for maximum
effect. Deepfakes and generative AI aid in producing misinformation. Advanced AI
can make authoritarian centralized decision making more competitive than liberal
and decentralized systems such as markets. It lowers the cost and difficulty of digital
warfare and advanced spyware.[206] All these technologies have been available since
2020 or earlier—AI facial recognition systems are already being used for mass
surveillance in China.[207][208]

There many other ways that AI is expected to help bad actors, some of which can
not be foreseen. For example, machine-learning AI is able to design tens of
thousands of toxic molecules in a matter of hours.[209]

Reliance on industry giants

Training AI systems requires an enormous amount of computing power. Usually
only Big Tech companies have the financial resources to make such investments.
Smaller startups such as Cohere and OpenAI end up buying access to data
centers from Google and Microsoft respectively.[210]

Technological unemployment
Main articles: Workplace impact of artificial intelligence and Technological
unemployment
Economists have frequently highlighted the risks of redundancies from AI, and
speculated about unemployment if there is no adequate social policy for full
employment.[211]

In the past, technology has tended to increase rather than reduce total employment,
but economists acknowledge that "we're in uncharted territory" with AI.[212] A survey of
economists showed disagreement about whether the increasing use of robots and AI
will cause a substantial increase in long-term unemployment, but they generally
agree that it could be a net benefit if productivity gains are redistributed.[213] Risk
estimates vary; for example, in the 2010s, Michael Osborne and Carl Benedikt
Frey estimated 47% of U.S. jobs are at "high risk" of potential automation, while an
OECD report classified only 9% of U.S. jobs as "high risk".[o][215] The methodology of
speculating about future employment levels has been criticised as lacking evidential
foundation, and for implying that technology, rather than social policy, creates
unemployment, as opposed to redundancies.[211] In April 2023, it was reported that
70% of the jobs for Chinese video game illustrators had been eliminated by
generative artificial intelligence.[216][217]

Unlike previous waves of automation, many middle-class jobs may be eliminated by

artificial intelligence; The Economist stated in 2015 that "the worry that AI could do to
white-collar jobs what steam power did to blue-collar ones during the Industrial
Revolution" is "worth taking seriously".[218] Jobs at extreme risk range
from paralegals to fast food cooks, while job demand is likely to increase for care-
related professions ranging from personal healthcare to the clergy.[219]

From the early days of the development of artificial intelligence, there have been
arguments, for example, those put forward by Joseph Weizenbaum, about whether
tasks that can be done by computers actually should be done by them, given the
difference between computers and humans, and between quantitative calculation
and qualitative, value-based judgement.[220]

Existential risk
Main article: Existential risk from artificial general intelligence
It has been argued AI will become so powerful that humanity may irreversibly lose
control of it. This could, as physicist Stephen Hawking stated, "spell the end of the
human race".[221] This scenario has been common in science fiction, when a computer
or robot suddenly develops a human-like "self-awareness" (or "sentience" or
"consciousness") and becomes a malevolent character.[p] These sci-fi scenarios are
misleading in several ways.

First, AI does not require human-like "sentience" to be an existential risk. Modern AI

programs are given specific goals and use learning and intelligence to achieve them.
Philosopher Nick Bostrom argued that if one gives almost any goal to a sufficiently
powerful AI, it may choose to destroy humanity to achieve it (he used the example of
a paperclip factory manager).[223] Stuart Russell gives the example of household robot
that tries to find a way to kill its owner to prevent it from being unplugged, reasoning
that "you can't fetch the coffee if you're dead."[224] In order to be safe for humanity,
a superintelligence would have to be genuinely aligned with humanity's morality and
values so that it is "fundamentally on our side".[225]

Second, Yuval Noah Harari argues that AI does not require a robot body or physical
control to pose an existential risk. The essential parts of civilization are not physical.
Things like ideologies, law, government, money and the economy are made
of language; they exist because there are stories that billions of people believe. The
current prevalence of misinformation suggests that an AI could use language to
convince people to believe anything, even to take actions that are destructive.[226]

The opinions amongst experts and industry insiders are mixed, with sizable fractions
both concerned and unconcerned by risk from eventual superintelligent AI.
[227]
Personalities such as Stephen Hawking, Bill Gates, and Elon Musk,[228] as well as
AI pioneers such as Geoffrey Hinton, Yoshua Bengio, Demis Hassabis, and Sam
Altman, have expressed concerns about existential risk from AI.

In 2023, many leading AI experts issued the joint statement that "Mitigating the risk
of extinction from AI should be a global priority alongside other societal-scale risks
such as pandemics and nuclear war".[229]

Other researchers, however, spoke in favor of a less dystopian view. AI

pioneer Juergen Schmidhuber did not sign the joint statement, emphasising that in
95% of all cases, AI research is about making "human lives longer and healthier and
easier."[230] While the tools that are now being used to improve lives can also be used
by bad actors, "they can also be used against the bad actors."[231][232] Andrew Ng also
argued that "it's a mistake to fall for the doomsday hype on AI—and that regulators
who do will only benefit vested interests."[233] Yann LeCun "scoffs at his peers'
dystopian scenarios of supercharged misinformation and even, eventually, human
extinction."[234] In the early 2010s, experts argued that the risks are too distant in the
future to warrant research or that humans will be valuable from the perspective of a
superintelligent machine.[235] However, after 2016, the study of current and future risks
and possible solutions became a serious area of research.[236]

Ethical machines and alignment

Main articles: Machine ethics, AI safety, Friendly artificial intelligence, Artificial
moral agents, and Human Compatible
Friendly AI are machines that have been designed from the beginning to minimize
risks and to make choices that benefit humans. Eliezer Yudkowsky, who coined the
term, argues that developing friendly AI should be a higher research priority: it may
require a large investment and it must be completed before AI becomes an
existential risk.[237]

Machines with intelligence have the potential to use their intelligence to make ethical
decisions. The field of machine ethics provides machines with ethical principles and
procedures for resolving ethical dilemmas.[238] The field of machine ethics is also
called computational morality,[238] and was founded at an AAAI symposium in 2005.[239]

Other approaches include Wendell Wallach's "artificial moral agents"[240] and Stuart J.
Russell's three principles for developing provably beneficial machines.[241]

Open source
Active organizations in the AI open-source community include Hugging Face,
[242]
Google,[243] EleutherAI and Meta.[244] Various AI models, such as Llama
2, Mistral or Stable Diffusion, have been made open-weight,[245][246] meaning that their
architecture and trained parameters (the "weights") are publicly available. Open-
weight models can be freely fine-tuned, which allows companies to specialize them
with their own data and for their own use-case.[247] Open-weight models are useful for
research and innovation but can also be misused. Since they can be fine-tuned, any
built-in security measure, such as objecting to harmful requests, can be trained away
until it becomes ineffective. Some researchers warn that future AI models may
develop dangerous capabilities (such as the potential to drastically
facilitate bioterrorism), and that once released on the Internet, they can't be deleted
everywhere if needed. They recommend pre-release audits and cost-benefit
analyses.[248]

Frameworks
Artificial Intelligence projects can have their ethical permissibility tested while
designing, developing, and implementing an AI system. An AI framework such as the
Care and Act Framework containing the SUM values—developed by the Alan Turing
Institute tests projects in four main areas:[249][250]

 Respect the dignity of individual people

 Connect with other people sincerely, openly, and inclusively
 Care for the wellbeing of everyone
 Protect social values, justice, and the public interest
Other developments in ethical frameworks include those decided upon during
the Asilomar Conference, the Montreal Declaration for Responsible AI, and the
IEEE's Ethics of Autonomous Systems initiative, among others;[251] however, these
principles do not go without their criticisms, especially regards to the people chosen
contributes to these frameworks.[252]

Promotion of the wellbeing of the people and communities that these technologies
affect requires consideration of the social and ethical implications at all stages of AI
system design, development and implementation, and collaboration between job
roles such as data scientists, product managers, data engineers, domain experts,
and delivery managers.[253]

The AI Safety Institute in the UK has released a testing toolset called 'Inspect' for AI
safety evaluations available under a MIT open-source licence which is freely
available on GitHub and can be improved with third-party packages. It can be used
to evaluate AI models in a range of areas including core knowledge, ability to reason,
and autonomous capabilities.[254]

Voicing concerns about issues including the existential risk from artificial general
intelligence,[255] in May 2023, Geoffrey Hinton announced his resignation from Google
in order to be able to "freely speak out about the risks of A.I." [256] He stressed that in
order to avoid the worst outcomes, establishing safety guidelines will require
cooperation among those competing in use of A.I.[257]