Artificial Intelligence and Life in 2030: The One Hundred Year Study on Artificial Intelligence

Substantial increases

• The panel expects substantial increases in self-driving vehicles, healthcare diagnostics and treatments, elder-care assistance, industrial robotics, and automated delivery, while applications remain specialized and require years of tailored research.Examples include agriculture, food processing, fulfillment centers, factories, drones, self-driving trucks, and delivery robots.Source: Substantial increases · pp. 4–5

EXECUTIVE SUMMARY

• AI applications are already changing daily life, while the panel expects increasingly useful technologies to emerge across transportation, healthcare, education, employment, and other domains by 2030.The report identifies eight domains and considers both progress over the past fifteen years and likely developments over the next fifteen.Source: Substantial increases · p. 4, EXECUTIVE SUMMARY · p. 4

OVERVIEW

• AI is already improving health, safety, and productivity, and advances in vision, speech recognition, and natural-language processing are making future interactions with machines more fluid, personalized, and service-oriented.The report projects self-driving transportation with on-time movement of people and packages, potentially making traffic jams and parking challenges obsolete in typical North American cities.Source: OVERVIEW · p. 6, Many have already grown · p. 6

Many have already grown

• The report focuses on eight domains because AI’s effects and challenges differ across them, including hardware reliability, expert interaction, public trust, fears of human marginalization, and diminished interpersonal interaction.The domains are transportation, healthcare, education, low-resource communities, public safety and security, employment and workplace, home/service robots, and entertainment.Source: Substantial increases · p. 4, Many have already grown · p. 6
• AI’s benefits are accompanied by privacy concerns, job displacement, bias and civil-liberties risks, and uncertainty over how economic gains are shared, making research, deployment, and regulation central to its social impact.The panel finds no imminent threat from self-sustaining machines, but warns that near-term design and policy choices will shape whether AI’s benefits are broadly distributed.Source: Substantial increases · pp. 4–5, Many have already grown · p. 6
• AI research is shifting toward large-scale and deep learning, reinforcement learning, robotics, human collaboration, and more capable interaction, supported by abundant data, cloud computing, and improved hardware.The report highlights progress in perception, sequential decision-making, robot-world interaction, dialogue, and systems that collaborate with people.Source: Many have already grown · pp. 8–9
• AI applications should be designed so people can understand them, participate in their use, and build trust in them.The panel defines success by the value AI creates for human lives.Source: Many have already grown · p. 10
• Public policy should ease society’s adaptation to AI, extend its benefits, and mitigate inevitable errors and failures.The panel also calls for debate over deployment, privacy protection, and fair distribution of AI’s benefits.Source: Many have already grown · p. 10
• Because AI technologies are advancing rapidly, governments should acquire technical expertise and support research on fairness, security, privacy, and societal implications.The panel recommends removing impediments and increasing private and public spending for this research.Source: Many have already grown · p. 10

Misunderstandings

• AI’s rapid advances require new legal and policy concepts addressing responsibility for failures, discrimination, cheating, worker displacement, best practices, and regulatory adaptation.Source: Misunderstandings · p. 10
• AI systems can be used harmfully, and ensuring their training data avoid biases that produce discrimination remains a deep technical challenge.Source: Misunderstandings · p. 10
• Policies should promote informed debate, transparency, accountability, innovation, and equitable access so AI benefits do not widen inequalities or concentrate power. [2]Broad mandates with transparency and enforcement are presented as preferable to narrow controls, while applications should respect human and civil rights.Source: Misunderstandings · pp. 10–11

SECTION I: WHAT IS ARTIFICIAL INTELLIGENCE?

• The report defines what AI is and is not and surveys thriving research and application areas as groundwork for analyzing AI’s impacts and public-policy recommendations.Source: SECTION I: WHAT IS ARTIFICIAL… · p. 12

DEFINING AI

• AI spans a multidimensional spectrum: calculators, software, animals, and autonomous systems differ by kind, speed, scale, autonomy, and generality.Nilsson defines AI as making machines intelligent, where intelligence enables appropriate, foresighted functioning.3Source: DEFINING AI · p. 12
• AI’s frontier advances incrementally as technologies become commonplace and cease to be called AI, a recurring “AI effect” that shifts the boundary forward.5Current development focuses on improving, generalizing, and scaling intelligence found in smartphones.Source: DEFINING AI · p. 12
• Human intelligence remains the natural benchmark because of its unmatched versatility, although systems can already exceed people in speed for tasks such as scheduling thousands of flights.6Matching a human ability is sufficient but not necessary for counting a system’s behavior as intelligent.Source: DEFINING AI · p. 13
• Deep Blue defeated Garry Kasparov 3.5-2.5 in 1997, then was portrayed as a collection of “brute force methods” that wasn’t “real intelligence.”8IBM’s subsequent publication described its search and evaluation procedures but did not mention “intelligent” even once.9Source: DEFINING AI · p. 13

An operational definition

• The report operationally treats AI as computer science studying intelligence by synthesizing it, with software emphasis increasingly complemented by hardware tailored for neural-network computing.10 11This reflects tighter coupling between hardware and software in advancing AI.Source: An operational definition · pp. 13–14
• AI draws on psychology, economics, neuroscience, biology, engineering, statistics, and linguistics; artificial neural networks were inspired by information flow in biological neurons.12 13 14These allied fields have contributed progress to AI and its varied approaches to intelligence.Source: An operational definition · p. 14

• AI technologies now pervade daily life, and the field is shifting toward intelligent systems that are human-aware and trustworthy [15].Source: AI RESEARCH TRENDS · p. 14
• AI’s recent expansion has been fueled by mature machine learning, cloud computing, web-scale data, deep learning, improved hardware, and data-driven market incentives [16].Deep learning uses adaptive neural networks trained with backpropagation [16].Source: AI RESEARCH TRENDS · p. 14

Large-scale machine learning

• Large-scale machine learning research focuses on adapting established algorithms to extremely large datasets, often using single-pass or sublinear methods.Traditional methods could make several data passes, whereas modern methods may inspect the data once or only a fraction of it.Source: AI RESEARCH TRENDS · p. 14

Deep learning and reinforcement learning

• Deep learning has advanced computer vision and other perception areas, while reinforcement learning targets real-world sequential decision-making and helped produce AlphaGo’s success.AlphaGo was initialized with human expert games, then improved through self-play and reinforcement learning.Source: AI RESEARCH TRENDS · pp. 14–15

Robotics

• Robotics research is moving beyond largely solved static navigation toward generalizable interaction and manipulation, constrained by limited labeled data and the need for safe exploration.Reliable visual, force, and tactile perception remain key enablers of robotic capability.Source: AI RESEARCH TRENDS · p. 15

Computer vision

• Computer vision has improved dramatically through GPUs, large datasets, and neural-network refinements, enabling some narrowly defined visual classifications to exceed human performance.Current work also emphasizes automatic image and video captioning.Source: AI RESEARCH TRENDS · p. 15

Natural Language Processing

• Natural language processing is becoming commoditized for data-rich languages, while research shifts from stylized requests toward dialogue and real-time translation.Google reported that 20% of current mobile queries were voice searches [18].Source: AI RESEARCH TRENDS · p. 15

Collaborative systems

• Collaborative-systems research develops autonomous systems that cooperate with humans and other systems by combining complementary strengths and augmenting human abilities.Source: AI RESEARCH TRENDS · p. 16

Natural Language

• Crowdsourcing uses human intelligence to solve tasks computers handle poorly, supplying large amounts of labeled or interaction data that advances computer vision and NLP.Research also examines task divisions between humans and machines based on differing capabilities and costs; examples include Wikipedia, citizen science, and Amazon Mechanical Turk.Source: Natural Language · p. 16

Processing is a very

• Algorithmic game theory and computational social choice address incentive structures for systems involving self-interested people, firms, and automated agents with potentially misaligned incentives.Key topics include mechanism design, rank-order aggregation, information elicitation, prediction markets, scoring rules, peer prediction, market equilibria, network games, and Poker.Source: Processing is a very · p. 16
• AI can process the large sensory datasets generated by interconnected appliances, vehicles, buildings, and cameras, while potentially helping resolve incompatible IoT communication protocols.IoT research focuses on connecting devices so they can collect and share sensory information for intelligent purposes.Source: Processing is a very · p. 16
• Neuromorphic computers aim to improve hardware efficiency and robustness through architectures inspired by biological neural networks, but they have not yet demonstrated major wins or broad commercial viability.Dedicated hardware could eventually train and execute deep neural networks more effectively than today’s simulated von Neumann implementations.Source: Processing is a very · pp. 16–17

Overall trends and the future of AI research

• Data-driven methods have displaced theorem proving, logic-based reasoning, planning, model-based approaches, Bayesian reasoning, and graphical models because deep learning succeeds with abundant data.Traditional approaches face difficulties with real-world grounding or restrictive modeling assumptions, while sensing-based data-driven systems close the loop on task outcomes.Source: Overall trends and the future of AI… · p. 17
• The Study Panel expects increasing focus on human-aware systems, interactive robot teaching, IoT systems, AI’s social and economic dimensions, safer robots, and data-driven products over fifteen years.These systems model or are designed around the characteristics of people with whom they interact.Source: Overall trends and the future of AI… · p. 17
• The panel expects traditional AI approaches to reemerge as practitioners recognize the inevitable limitations of purely end-to-end deep learning.It encourages researchers to retain awareness of progress in earlier AI and related fields, including control theory, cognitive science, and psychology.Source: Overall trends and the future of AI… · p. 17

SECTION II: AI BY DOMAIN

• AI research and implementation vary across transportation, robotics, healthcare, education, low-resource communities, public safety, employment, and entertainment, with projected everyday-life changes through 2030.The analysis targets distinct sectoral impacts and challenges and predicts trends in a typical North American city over the next fifteen years.Source: SECTION II: AI BY DOMAIN · p. 18

TRANSPORTATION

• Autonomous transportation is expected to become commonplace and strongly shape public perceptions of AI as people encounter physically embodied systems in daily life.The transition may occur suddenly once hardware is sufficiently safe and robust, requiring public adjustment.Source: TRANSPORTATION · p. 18

Self-driving vehicles

• Self-driving vehicles could reduce car ownership, alter commuting and city design, expand mobility for youth, elderly, and disabled people, and shift transportation toward shared services.Their effects on total travel remain uncertain because empty trips and increased willingness to travel could increase miles, while shared autonomous vehicles and incentives could reduce them.Source: TRANSPORTATION · p. 18, Self-driving vehicles · p. 21

TRANSPORTATION

• Smartphones, GPS, sensors, connectivity, and machine learning enable real-time traffic prediction, route calculation, ridesharing, autonomous driving, and increasingly capable vehicle perception.Current vehicles already support automated parking, adaptive cruise control, stop-and-go steering, blind-spot alerts, collision braking, object detection, and sound recognition.Source: TRANSPORTATION · pp. 18–19

Self-driving vehicles

• 1,500,000 miles were logged by Google’s autonomous vehicles, including 300,000 miles without an accident, while Tesla deployed semi-autonomous driving requiring human oversight.The report notes that semi-autonomous systems may become less safe if driver attention declines, especially after the first autonomous-car traffic fatality in June 2016.Source: Self-driving vehicles · p. 20
• Autonomous vehicles and connected infrastructure raise unresolved requirements for safe testing, cybersecurity, ethical decision-making, accident responsibility, and legal regulation.As of 2016, several jurisdictions allowed public-road testing, but existing laws generally did not assign blame for autonomous or semi-autonomous crashes.Source: Self-driving vehicles · pp. 20–21

Transportation planning

• Cities use sensors, cameras, and AI to optimize transit schedules, traffic flows, speed limits, road pricing, and traffic-light timing through dynamic strategies.These methods aim to use limited transportation resources more efficiently and depend on widespread data and connectivity.Source: Transportation planning · pp. 21–22
• Dynamic transportation pricing can improve demand management but raises fairness concerns because high-demand market conditions may make services unavailable to some segments of the public.Before the 2000s, planners primarily relied on static pricing tied to particular days or times.Source: Transportation planning · p. 22

Ethical questions arise

• Predictive models of individuals’ movements, preferences, and goals are expected to emerge as transportation data become more available, creating ethical issues.The report explicitly refers these ethical issues to its policy discussion in Section III.Source: Ethical questions arise · pp. 22–23
• Connected vehicles, drones, and autonomous transportation raise unresolved questions about safety rules, regulations, and the appropriate framework for deployment.Car-to-car communication may improve safety, while drone delivery and broader transportation applications still require appropriate safety rules and regulations.Source: Ethical questions arise · p. 23
• Privacy concerns over sensing, connected infrastructure, and private transportation data require preemptive industry action or legal safeguards.The report states that how these issues are addressed will affect the pace and scope of AI-related advances in transportation.Source: Ethical questions arise · pp. 22–23
• On-demand transportation uses dynamic pricing to ration access by willingness-to-pay while creating policy concerns about competition, regulation, and safety.These services match drivers and passengers by location and suitability, encourage additional driver supply, and appear likely to support self-driving cars.Source: Ethical questions arise · p. 23

HOME/SERVICE ROBOTS

• Over the next fifteen years, home robots are expected to become safer, more reliable, and more useful, while commercial opportunities remain limited to narrowly defined applications by technical constraints and high hardware costs.Potential applications include package delivery, office cleaning, and enhanced security.Source: HOME/SERVICE ROBOTS · pp. 24–25
• Home-robot progress has produced better navigation, self-charging, obstacle handling, and 3D mapping, but real-home mobility, reliable hardware, and limited consumer demand still restrict application diversity.VSLAM enables complete 3D house models and more efficient cleaning coverage, while many homes remain difficult because of steps and staircases.Source: HOME/SERVICE ROBOTS · pp. 24–25
• Cloud software and shared household data will enable faster software releases, cloud-based machine learning, and improvements to already deployed home robots.These capabilities build on advances in onboard processors, specialized silicon, and low-cost sensing.Source: HOME/SERVICE ROBOTS · p. 25
• Deep-learning advances in speech understanding and image labeling will improve robots’ interactions with people at home.Low-cost 3D sensors and increasingly capable embedded processors are also expected to accelerate onboard AI and robot adoption.Source: HOME/SERVICE ROBOTS · p. 25
• Low-cost safe robot arms and numerous startups are expanding manipulation and social-interaction research that may become applicable in homes around 2025, while raising potential ethics and privacy issues.The robot arms have reached hundreds of research laboratories worldwide, and more than half a dozen startups are developing AI-based home robots.Source: HOME/SERVICE ROBOTS · p. 25

HEALTHCARE

• Healthcare AI could improve health outcomes and quality of life for millions through clinical support, monitoring, coaching, surgery assistance, and healthcare-system management, if trust and institutional obstacles are addressed.Recent successes include inferring health risks from social media, predicting at-risk patients, and supporting surgery with robotics.Source: HEALTHCARE · pp. 25–26
• Healthcare AI deployment is slowed by difficult data use, outdated regulations, misaligned incentives, poor human-computer interaction, and the risks of implementing technology in a complex system.Data sources have expanded across monitoring devices, mobile apps, electronic health records, and medical robots, but finer-grained diagnostics and treatments remain difficult.Source: HEALTHCARE · pp. 25–26
• Clinical AI remains largely unrealized because healthcare delivery is structurally ill-suited to rapid advances, with poor EHR implementation, concentrated vendors, substandard interfaces, and regulatory barriers limiting analytics.Although Affordable Care Act incentives accelerated EHR adoption, implementation eroded clinicians’ confidence and physicians routinely dismiss intrusive pop-ups.Source: HEALTHCARE · p. 26

AI-based applications

• Healthcare AI could shift clinicians toward supervising automated symptom-pattern reasoning while preserving their hands-on experience, intuition, and responsibility for integrating human care with machine processes.Clinicians must be engaged from the outset so systems are well-engineered and trusted.Source: AI-based applications · p. 26
• Population-scale analytics could enable finer-grained diagnosis, genotype–phenotype discovery, “patients like mine” treatment support, and personalized recommendations from clinical, genomic, social, wearable, and environmental data.These systems may create self-defined subpopulations supported by automated recommendation and monitoring ecosystems.Source: AI-based applications · pp. 26–27
• Regulatory, legal, privacy, and social barriers—including slow FDA approval, HIPAA constraints, and absent privacy standards—can impede healthcare AI despite its potential for major improvements.Faster approval may become possible if effective post-marketing reporting is treated as a hedge against some safety risks.Source: AI-based applications · pp. 26–27
• Medical-image AI is more difficult than object recognition because it requires fine-grained, high-stakes judgments, continued radiologist involvement, and data federation across institutions.Strict regulation and limited digital archives have constrained progress, leaving very large integrated-care organizations best positioned to attack these problems.Source: AI-based applications · p. 27
• Over the next fifteen years, image triage and second-level checking are expected to improve imaging speed and cost-effectiveness, while deep networks can already produce basic radiological findings reliably from linked scans and records.The report does not expect fully automated radiology during this period.Source: AI-based applications · p. 27

The problem in medicine

• Healthcare robotics is advancing from surgical platforms such as da Vinci toward sensing, analytics, augmentation, and automation ecosystems, while hospital-delivery automation remains less commercially successful.The da Vinci system provides 3D visualization and wristed instruments and is used in nearly three quarters of a million procedures annually.Source: AI-based applications · p. 28, The problem in medicine · p. 28
• Many healthcare tasks will be augmented rather than fully automated, requiring intimate human–machine collaboration for activities such as delivery, patient walking, and surgical manipulation.Automation platforms can also generate data for quality assessment, error detection, optimization, and closed-loop medicine.Source: The problem in medicine · pp. 28–29

Mobile health

• Mobile health is expanding through biometrics, specialized motion tracking, connected homes, and apps that mine social and healthcare data for behavior change, monitoring, prediction, and coaching.Examples include clinician-trained Tele-Language and Lifegraph alerts from passively collected smartphone data.Source: Mobile health · p. 29

Personalized rehabilitation

• AI-enabled health management could monitor movement, activities, mood, and behavior, recommend healthier practices, alert caregivers, and help manage co-morbidities and treatment interactions.These functions are described as combining mobile applications, social platforms, in-home monitoring, and health-information access.Source: Personalized rehabilitation · p. 30
• Personalized rehabilitation and in-home therapy could reduce hospital or care-facility stays while extending elderly people’s capabilities through assistive devices and low-cost sensing.Intelligent walkers, wheelchairs, exoskeletons, hearing aids, visual aids, and home-based sensing are identified as key supports.Source: Personalized rehabilitation · p. 30
• Useful elderly-facing social robots will require integrating natural-language processing, reasoning, learning, perception, and robotics, while raising privacy questions and challenges for an active aging population.The proposed systems must be both useful and usable by elderly people, and innovations may affect privacy among friends, family, and caregivers.Source: Personalized rehabilitation · p. 30

EDUCATION

• AI has advanced substantially in education, enabling personalization at scale through intelligent tutors, natural-language technologies, online learning, and learning analytics, while human teachers remain necessary.The report emphasizes that effective education still requires active engagement by human teachers and face-to-face learning.Source: EDUCATION · p. 31
• Over the next fifteen years, intelligent tutors, classroom and home AI assistance, and virtual-reality learning are expected to expand, but computer-based systems are unlikely to replace school teachers.Adoption has been slowed by limited funding and insufficient evidence that AI improves students’ learning objectives.Source: EDUCATION · p. 31

Teaching robots

• Educational robot kits such as Ozobot, Cubelets, Dash and Dot, and PLEO rb offer coding, reasoning, programming, and biology activities, but widespread adoption requires evidence of improved academic performance [75, 76, 77, 78].The kits use sensing technologies and programmable or interactive behaviors to support learning activities.Source: Teaching robots · p. 31

Intelligent Tutoring Systems (ITS) and online learning

• Intelligent tutoring systems have moved from laboratory research into practical language, mathematics, and subject-specific instruction, using ASR, NLP, hints, and context-specific feedback [79, 80, 81].Applications include foreign-language training, mathematics, geography, circuits, medical diagnosis, computer literacy, programming, genetics, and chemistry.Source: Intelligent Tutoring Systems (ITS) and… · pp. 31–32

It can be argued that AI

• Higher-education applications include SHERLOCK82 for aircraft troubleshooting, avatar-based cultural training, and Bayesian Knowledge Tracing83 for personalized tutoring and sequencing.These systems support domain-specific practice, behavioral preparation, individualized mastery learning, and problem sequencing.Source: It can be argued that AI · p. 32
• MOOCs and online education combine NLP, machine learning, crowdsourcing, adaptive tools, and automated assessment for essays, short answers, and programming assignments.These systems are expanding in professional and lifelong learning, where reduced reliance on face-to-face interaction benefits working professionals and career changers.Source: It can be argued that AI · pp. 32–33
• AI-supported education has scaled classrooms, with some higher-education classes reaching a few tens of thousands of students.The report attributes this multiplication of classroom size to AI as a key enabling technology.Source: It can be argued that AI · p. 32

Learning analytics

• Learning analytics uses large-scale online-course data with deep learning and NLP to study student engagement, behavior, and outcomes.Current projects model misconceptions, predict failure risk, provide real-time feedback, and connect cognitive research to educational technology.Source: Learning analytics · p. 33
• Educational AI adoption is slowed by institutional resource constraints and limited evidence of effectiveness, while hype has often exceeded the reality of MOOCs and related technologies.Foundations and teacher-training programs are addressing these problems gradually, and experiences such as San Jose State’s Udacity experiment have prompted more cautious assessments.Source: Learning analytics · p. 33

While formal education

• Over the next fifteen years, AI is expected to assist human teachers through improved interaction and increasingly sophisticated virtual-reality learning environments.Examples include interactive historical documents, archaeological sites, and difficult-to-access environments in biology, anatomy, geology, and astronomy.Source: Learning analytics · p. 34

MOOCs and other forms

• MOOCs and adaptive systems will blend formal and self-paced learning from K-12 through university, enabling individualized pacing and faster advances in learning analytics.Online systems are expected to personalize instruction as they learn from students, while formal education remains part of blended classrooms.Source: MOOCs and other forms · p. 34
• Digital reading devices, machine translation, online professional degrees, and certifications will make educational materials and continuing education more accessible and affordable.These tools are expected to provide supplementary subject information, translate materials, and help adults and working professionals retrain as fields evolve.Source: MOOCs and other forms · p. 34
• Online education may expand access where schooling is scarce, but reduced face-to-face contact may harm social development, even as some autistic children benefit from AI interaction.Positive effects depend on people having tools to access online resources, and some technologies may also produce neurological side effects.Source: MOOCs and other forms · pp. 34–35

LOW-RESOURCE COMMUNITIES

• Targeted incentives could direct AI toward low-resource communities through predictive models for lead poisoning, prenatal care, code violations, and food distribution.These applications help agencies prioritize limited resources and route surplus food before it spoils.Source: LOW-RESOURCE COMMUNITIES · p. 35
• AI could use trusted peer networks to spread health information among homeless youth, but deployment must prevent discriminatory proxies and preserve auditability.Dynamic networks complicate research, while careful implementation could reduce discrimination because AI systems are more readily audited than humans.Source: LOW-RESOURCE COMMUNITIES · p. 36

PUBLIC SAFETY AND SECURITY

• By 2030, North American cities will rely heavily on AI for surveillance, drones, anomaly detection, and predictive policing, with trust determining whether policing becomes targeted or overbearing.Careful deployment may reduce human bias, but expanded surveillance also creates privacy and safety concerns.Source: PUBLIC SAFETY AND SECURITY · pp. 36–37
• AI will improve fraud detection, cybersecurity, crime-scene and rescue prioritization, video analysis, and social-media monitoring, although many systems are not ready to automate operations.Transfer learning may accelerate adaptation to new scenarios, while privacy risks and possible law-enforcement overreach remain significant.Source: PUBLIC SAFETY AND SECURITY · pp. 36–37
• Security agencies will expand AI use for efficiency, including vision, speech, gait analysis, and TSA’s DARMS airport-screening project, but biased risk datasets require safeguards.AI simulations may also train international law-enforcement collaboration before tools support actual investigations.Source: PUBLIC SAFETY AND SECURITY · p. 37

Employment and workplace

• AI-driven changes in digital systems are expanding the range of routine tasks and increasingly reaching professional services, with prior technologies affecting skilled-middle workers more than the lowest- or highest-skilled workers.110Examples include automation of parts of first-year lawyers’ jobs through legal information extraction and topic modeling.111 112Source: EMPLOYMENT AND WORKPLACE · p. 38
• AI will likely replace tasks rather than jobs in the near term, gradually affecting sectors from legal services to radiology, trucking, and gardening while creating new kinds of jobs.The effects may range from limited augmentation to complete replacement, and emerging jobs are harder to anticipate than jobs likely to disappear.Source: EMPLOYMENT AND WORKPLACE · p. 38
• AI may enable smaller organizations by reducing the need to scale through human labor and may create markets and employment by lowering entry barriers and enabling new modes of interaction.Examples include app stores, AirBnB, and taskrabbit.Source: EMPLOYMENT AND WORKPLACE · pp. 38–39
• AI systems can lower the cost of goods and services, but job losses may remain more politically salient than diffuse gains in living standards.A sudden replacement of all human jobs within one generation is considered highly unlikely, although AI is expected to gradually enter nearly all employment sectors.Source: EMPLOYMENT AND WORKPLACE · p. 39
• Without mitigating policies, AI could reduce the value of labor, widen inequality, and leave many workers unable to attain acceptable living standards, requiring political responses and stronger social safety nets.113Short-run measures include education, retraining, and new goods and services; longer-term options include expanded social services or guaranteed basic income, alongside debate over sharing AI-produced wealth.114Source: EMPLOYMENT AND WORKPLACE · p. 39

ENTERTAINMENT

• AI is expanding entertainment creation and analysis through sports analytics, music composition, soundtrack recognition, computer vision, NLP, natural-language scene generation, stylometry, and painting analysis.These tools support both professional production and lay-user creativity across media and the arts.Source: ENTERTAINMENT · p. 40
• AI-powered entertainment is becoming more interactive, personalized, and engaging, while raising concerns about reduced interpersonal interaction and motivating research on societal benefits.Internet platforms use AI techniques including natural language processing, information retrieval, image processing, crowdsourcing, and machine learning; collaborative filtering recommends content from demographic details and browsing history.Source: ENTERTAINMENT · p. 40
• Future entertainment systems may combine virtual reality, haptics, companion robots, dialogue-based interaction, and characteristics such as emotion, empathy, and environmental adaptation.Improvements in Automatic Speech Recognition are expected to make interaction with robots and other entertainment systems progressively more human-like.Source: ENTERTAINMENT · p. 41
• AI tools will democratize high-quality content creation and enable inexpensive customization through ASR, dubbing, and Machine Translation, although proliferating AI-created media may make future tastes difficult to predict.Examples include composing music and choreographing dance with avatars.Source: ENTERTAINMENT · p. 41
• Micro-analysis and micro-serving of individual preferences could enable media conglomerates to control people’s ideas and online experiences, creating a need for societal safeguards.The report leaves open whether broader society will develop measures to prevent such “Big Brothers.”Source: ENTERTAINMENT · p. 41

SECTION III: PROSPECTS AND RECOMMENDATIONS FOR AI PUBLIC POLICY

• AI policy should create societal value by building interpretability, participation, trust, human capability, inclusion, helpful innovation, and broad corporate and civic responsibility.The report calls for new or retooled laws and policies to address widespread impacts and for social debate over sharing AI-created wealth.Source: SECTION III: PROSPECTS AND… · p. 42

AI POLICY, NOW AND IN THE FUTURE

• AI’s success will be measured by the value it creates for human lives and by how easily people use and adapt to applications developed incrementally.Small technical improvements can nevertheless produce novel, game-changing applications.Source: AI POLICY, NOW AND IN THE FUTURE · p. 42
• Errors in increasingly critical AI applications may trigger backlash, so explainability, human participation, and realistic expectations are needed to build trust and sustain use.Frustration with promised but undelivered functions diminishes trust and future willingness to use systems.Source: AI POLICY, NOW AND IN THE FUTURE · p. 42
• AI systems may erode people’s abilities on automated tasks, although humans and machines can complement one another through human reasoning and creative expression.Early exposure to AI may improve children’s interactions while widening generational differences in perceptions of its societal influence.Source: AI POLICY, NOW AND IN THE FUTURE · p. 42
• Unequal access to AI, computation, and data could widen opportunity gaps by improving the capabilities of advantaged users while disadvantaging groups lacking relevant technologies.Machine translation can expand access to learning resources, whereas English-only speech translation disadvantages non-English speakers.Source: AI POLICY, NOW AND IN THE FUTURE · p. 43
• Designer and user biases can deepen social inequality and undermine fairness, although AI decision tools could also reduce biases in lending or criminal justice.Some speech-recognition systems perform poorly for women and people with accents.Source: AI POLICY, NOW AND IN THE FUTURE · p. 43
• AI surveillance, privacy, bias, and wealth distribution will generate persistent ethical debates, especially as expertise and underlying datasets concentrate in large corporations.Personal and societal value judgments make these disputes resistant to quick resolution.Source: AI POLICY, NOW AND IN THE FUTURE · p. 43
• The panel recommends building AI expertise across government, removing barriers to research on fairness, security, privacy, and social impacts, and increasing interdisciplinary public and private funding.Such research should assess AI’s safety, privacy, fairness, and broader societal effects.Source: AI POLICY, NOW AND IN THE FUTURE · pp. 43–44
• AI’s integration raises unresolved questions about accountability, discrimination, displaced workers, and benefit-sharing, while existing regulatory regimes may require adaptation or fundamental reconfiguration.In the United States, oversight is currently sector-specific across agencies including the FDA, FAA, FTC, and SEC.Source: AI POLICY, NOW AND IN THE FUTURE · p. 44

Policy and legal considerations

• Regulating AI requires difficult tradeoffs between innovation and safety because software components come from multiple vendors, while AI’s effects vary across applications and jurisdictions.Critical infrastructure is defined by end-user application rather than the technology producing the software, and regulatory bodies will need greater technical expertise.Source: AI technologies. Private · pp. 44–45

Privacy

• AI can expose private information and reproduce bias through correlated proxies, although careful design, testing, and deployment may yield less biased decisions than typical human judgment.Applications such as credit-risk and recidivism prediction raise concerns about using race, sexual orientation, or correlated features such as zip code.Source: AI technologies. Private · p. 46

Liability

• AI’s unpredictable physical and criminal harms challenge existing liability rules, leaving courts to determine responsibility when injuries are unforeseeable or machine behavior would constitute a crime.Civil liability may shift toward manufacturers, while criminal law must address whom to hold accountable and under what theory.Source: AI technologies. Private · pp. 46–47

Agency and Certification

• AI’s legally salient activities complicate agency and certification because law must determine when software can bind people or corporations and who must demonstrate professional competency.Unresolved examples include autonomous surgical systems and applications providing legal advice.Source: AI technologies. Private · p. 47

Labor

• AI’s effects on employment are ambiguous but uneven, as some skills lose demand while outcomes depend substantially on government policy, work organization, and individual reskilling.AI may eliminate some jobs while creating others, with altered or terminated workers potentially seeking legislative or judicial recourse.Source: AI technologies. Private · p. 47

Taxation

• AI could shift public revenue from payroll and income taxes toward capital expenditure, destabilize dependent budgets, and prompt legislation that slows or redirects technological development.Autonomous cars may reduce revenue from speeding and parking tickets, requiring municipalities to find alternatives.Source: AI technologies. Private · pp. 47–48

Like other technologies,

• AI can be used for good or nefarious purposes, so society urgently needs informed debate about steering it to enrich lives while encouraging creativity.Policies should be judged by whether they democratically distribute AI’s benefits or concentrate power among a fortunate few.Source: Like other technologies · p. 48
• Policymakers should reject general AI regulation because AI lacks a clear definition and its risks differ across domains, requiring distinct industry-specific rules.Government needs expertise to scrutinize public- and private-sector standards and craft regulations where necessary.Source: Like other technologies · p. 48

Guidelines for the future

• Strict, detailed privacy rules produced corporate compliance mentalities, whereas ambiguous goals paired with transparency and enforcement better encouraged responsibility and privacy protections. 139The comparative study found that U.S. and German approaches fostered professional staff, internal processes, stakeholder engagement, and adaptation to technological change.Source: Guidelines for the future · pp. 48–49
• AI policies should strengthen accountability and transparency, respect human and civil rights, spread benefits fairly, and be continually reevaluated using societal challenges and fielded-system evidence.As AI advances through large datasets, sensory technologies, and deep learning, new applications in areas such as transportation and healthcare must build public trust and understanding.Source: Guidelines for the future · p. 49

APPENDIX I: A SHORT HISTORY OF AI

• This history is primarily based on Nilsson’s book and reflects a current emphasis on data-intensive methods and big data, rather than a complete or fully balanced account. [140]The report explicitly notes that data-intensive methods have not yet solved all problems.Source: Appendix I · p. 50
• AI was officially named and organized as a field at the 1956 Dartmouth workshop, whose goal was to make machines simulate aspects of intelligence. [141]John McCarthy is credited with first using the term in the workshop proposal co-authored with Marvin Minsky, Nathaniel Rochester, and Claude Shannon.Source: Appendix I · p. 50

The field of Artificial

• The field’s foundations preceded Dartmouth, including Bayesian probability, Boolean logic, statistics, electronic computing, and autonomous robots. [142, 143, 144, 145, 146]These developments supplied mathematical, computational, and physical bases for later AI research.Source: The field of Artificial · p. 50
• Turing’s formal model of computing and 1950 essay explored machine intelligence, intelligence testing, and automatic learning, but available computing resources prevented implementation. [147]His ideas inspired AI while remaining constrained by the hardware of his time.Source: The field of Artificial · p. 50
• Early AI developed heuristic search, character recognition, natural-language processing, mobile robotics, self-improving checkers, neural networks, and expert systems. [149, 150, 151, 152, 153, 154, 155, 156]Examples included Newell and Simon’s Logic Theorist and General Problem Solver, Shakey, Rosenblatt’s Perceptron, and domain-specific knowledge repositories.Source: The field of Artificial · p. 51
• By the mid-1980s, insufficient grounding in environmental data and overreliance on Boolean logic contributed to limited practical success, declining interest, and the “AI winter.”The period exposed the need to quantify uncertainty and connect systems more directly to physical signals and data.Source: The field of Artificial · p. 51
• The 1990s resurgence rejected Good Old-Fashioned AI as an adequate end-to-end approach and was enabled by cheaper sensing and actuation, Internet data, computing power, and storage. [157, 158]These changes made statistical techniques that derive solutions from data more feasible and helped AI influence daily life.Source: The field of Artificial · p. 51
• Traditional AI subfields include search and planning, knowledge representation and reasoning, machine learning, multi-agent systems, robotics, human-robot interaction, computer vision, natural-language processing, social-network analysis, and crowdsourcing. [159]The field’s separation into subfields supports deep technical progress, but large-scale intelligence requires integration; AlphaGo combined machine learning with search. [160, 161]Source: The field of Artificial · p. 52