Table of ContentsageSectionNumberAbout the editorIntroductionNational artific ial inte lligence research and deve lopment s trate gic planPreparing For The Future Of Artificial IntelligenceArtificial Intelligence, Automation, and the EconomyAcronyms
fill those gaps The al r&d strate gic Plan identifies stratefor both near -term andlong-term support of Al that address important technical and soc ital challenges The al r&dStrategic Plan, however, does not define specific research agendas for individual Federalnstead, it sets objectives for the Executive Branch, wit hin which agencies may pursueaut horities, and bud gets so that theresearch portfolio is consistent with the Al r&D Strategic Plancan increase economic prosperity by the introduction of new products and services which cancreate new markets, and impro ve the quality and efficiency of existing goods and services acrossmultiple industries including: Manufacturing, Logistics, Finance, Transportation, AgricMarketing, Communications, and Science and Technology
Al can also improve educationalopportunities andllbeing and enhanced national and home land securitySince its beginnings, Al research has ad vanced in three technology waves The first wavecrafted knowledge, with a strous in the 1980s on rule-based e xpertsystems in we ll-defined domains, in which know ge was collected from a human expert,expressed in"if-then"rules, and then imp lemented in hardware Such systems-enab led reasoninwas applied successfully to narrowly defined problems, but it had no ability to learn or to dealwith uncertainty Nevertheless, they still led to important solutions, and the deve lopmenttechniques that are still actively usedThe second wave of Al research from the 2000s to the present is characterized by the ascent ofinability of significantly larger amounts of digital daive massively parallel computational capabilities, and improved learning techniqhave brought significant ad vances in Al when app lied to tasksuch as image arecognition, speech understanding, and human language translation The fruits of these adancesare everywhere: smartphones perform speech recognition, ATMs perform handrecognition on written checks, email applications perform spam filtering, and free online services
Khe deve lopment of dAl systems now regularly outperform humans on specialized tasks Major milefirst surpassed human performance inc lude: chess (1997), trivia(2011), Atari games(2013)mage recognition (2015), speech recognition(2015), and Go(2016) The pace of suchmilestones appears to be increasing, as is the de gree to which the best-performing systems arebased on machine learning methods rather than sets of hand-coded ruleSuch achie vements in al have been fue led by a strong base of funda me nta l research
tuthe number of web of science- inde xedI articleatoning"deep learningld The trends also reveal the increasingly global nature of research, with the United States nolonger leading the world in publication numbers, or even publications receiving at least onecitatIomodels with an explanation and correction interface, to clarify the bas is for and reliabilityoutputs, to operate with a high degreemo vecapabilities that can generalize across broader task domains If successful, engineers could createdelsd phenatural communication with people, learn and reason as they encounter ne w tasks and situationsproblems by generalizing frorsystems might be construc ted automatically through ad vanced methods These models couldenable rapid learning in AI systems They may supply"meaning" or"unde rstanding"to the Alsystem, which could then enable the Al systems to ac hie ve more general capabilities
Theutlined in this ai r&d Strate gic plan focusunlikely to address and thus areas that are most likely to benefit from Federal investment Thesepriorities cut across all of al to include needs common to the Al sub-fields of perception,utomated reasoning/planning, cognitive systems, machine learning,I languageentire field, rather than only focusing on individual research challenges specific to each subomain To imp lement the plan, detailed roadmaps should be de veloped that address thecapability gaps consistent with the planOne of the most important Federal research priorities, outlined in Strategy l, is for sustainedlong-term research in aI to drive disco very and insight many of the investments by the u
sFederal government in high-risk, high-re ward fundamertechno logical advances we depend on today, including the Internet, GPS, smchrecognition, heart monitors, solar pane ls, ad vanced batteries, cancer therapies, and much, muchmore The promise of Al touches nearly every aspect of society and has the potentiasignificant positive societal and economic benefits Thus, to maintain a world leadership positionin this area, the United States must focus its investments on high-priority fundamental and longerm al researchallenges in how to best create AI systems that work with people in intuitive and helpful waysThe walls between humans and aI systems are slowly beginning to erode, with AI systemsffective methods for human- AI interaction and collaboration, as outlined in Strategy 2AI ad vancement are providing many positive benefits to society and are increasing UStional competit8 However, as with most transformative technologies, Al presentskssafety, ethical, and leAI science and techno lo gy develop, the federal government must also invest in research to bette
understand what the imp licatiofor ai for all these realms and to addreoutlined in StrategA critical gap in current Al technology is a lack of methodologies to ensure the safety andpredictable performance of AI systems Ensuring the safety of AI systems is a challenge becausef the unus ual complexity and evolving nature of these systems Several research prioritiesaddress this safety challenge
First, Strategy 4 emphasizes the need for explainable andtransparent systems that are trusted by their users, perform in a manner that is acceptable to theusers,and can be guaranteed to act as the user intendedThe potential capabilities and complexity of AI systems, combined withalth of possibleions with human users and the environment, makes it critically important to invest inrch that increases the security and control of al techno gies Strategy 5 calls on the Federahared public datasets for al trnd testinsder to adheprogress of Al research and to enable a more effective comparison of alternative solutiongaps, and drive inno vative solutions for specific problems and challenges Standards andbenchmarks are essential for measuring and evaluating AI systems and ensuring that Altechnologies meet critical objectives for functionality and interoperabilityFinally, the growing prevalence of al technologies across all sectors of soc iety creates newpressures for Al R&Ddeep understanding of the technology who can generate new ideas for ad vancing the boundarieskno wledge in the field The Nation should take action to ensure a sufficient pipeline ofapable talent, Strategy 7 addresses this challengwhile the ultimate gonany Al algorithms is to address open challenges with human-likeolutions, we do not have a good understanding of what the theoretical capabilities and
for ai and the extent to which such human-like solutions areAl algorithms Theoretical work is needed to better understand why Al techniques-especiallymachine learning-often work well in practicewhile different disc ip lines (including mathematics, control sciences, and computer science)arestud ying this issue, the field currently lacks unified theoretical models or frameworks tounderstand Al system performance Additional research is needed on computational solvability,which is an understanding of the classes of problems that Asolving, and likewise, those that they are not capable of solving This understanding must bevelopedt of existing hard ware in order to see how the hard ware affects thperformance of these algorithms Understanding which problems are theoretically unsolvablecanlead researchers to de velop approximate solutions to these problems, or even open up new linesof research on new hardware for AI systems
For example, when invented in the 1960s, ArtificialNeural Networks(ANNS)could only be used to solve very simple problems It only becamefeasible to use ANNs to solve complex problems after hard ware improvements sucharallelization were made, and algorithms were adjusted to make use of the new hardware Suchdevelopments were key factors in enabling today's significant advances in deep learningGeneral al has been an ambition of researchers since the ad vent of al but current systestill far from achie ving this goal The relationship between narrow and general Al is currentbeing explored; it is possible that lessons from one can be applied to improve the other and viceversa While there is no general consensus, most AI researchers be lie ve that general Al is stildecades away, requiring a long-term, sustained research effort to achieveHowe ver, groups and networks of AI systems may be coordinated or autonomously collaboratem tasks not possible with a single AI system, and mayleading the team The development and use osignificant research challenges in planning, coordination, control, and scalability of suchsystems, Planning techniques for multi-AI systems must be fast enough to operate and adapt in
time tont They should adapt in a fluidvecused on centralized plaand coordination technwever, these approaches aresub ject to single points of failure, such as the loss of the planner, or loss of the communicationsalgorithmically, and are often less efficient and incomplete but potentially offer greaterrobustness to single points of failure Future research must discover more efficient, robust, andscalable techniques for planning, control, and collaboration of teams of multip le aI systems andhumansAttaining human-like AI requires systems to exp lain the mselves in ways that peop le canunderstand this will resultgeneration of intelligent systems, such as intelligent tutoringystems and intelligent assistants that are effective in assisting people when performing theirtasks There is a significant gap, however, bet ween the way current al algorithms work and howle learn and perform tasks
Peopleble of learning from just a few examples, or byreceiving formal instruction and/or"hints"to performing tasks, or by observing other peopleperforming those tasks Medical schools take this approach, for example, when medical studentslearn by observing an estab lished doctor performing a complex medical procedure Even in higlerformance tasks such as world-championship Go games, a master-level player would havead games to train him/ herself In contrast, it would take hundredsears for a human to play the number of games needed to train Alpha Go More foundationalresearch on new approaches for achieving human-like Al would bring these systems closerSignificant ad vances in robotic techno gies over the last decade are leading to potential impactsin a multiplicity of applications, including manufacturing, lo gistics, medicine, healthcaredefense and national security, a griculture, and consumer products While robots were historicallyenvision static industriarobots and humans Robotics technologies are now showing promise in their ability to
lent enhHowever, scientists need to make these robotic systems more capable, reliable, and easy-to-users need to better unders tand robotic perception to extract information from a varietysensors to provide robots with real-time situational awareness Progress is needed in cognitioand reasoning to allow robots to better understand and interact with the physical world animproved ability to adapt and learn will allo w robots to generalize their skills, perform self-assessment of their current performance, and learn a repertoire of physical mo vements fromhuman teachers Mobility and manipulation are areas for further investigation so that robots cannove across rugged and uncertain terrain and handle a variety of objects dexterously
robotseed to learn to team to gether in a seamless fas hion and co lla borate with humans in a way that istrustworthy and predictableitude of goalsh those goals constraintsthose actions, and other factors, as well as easily adapt to modifications in the goals In addition,aspects of their currerded to generalize these facetof hAI systems to de velop systemWhile much of the prior focus of AI research has been on algorithms therrorle perfowork is needed to de ve lop systehuman capabilities across many domains Human augmentation research inc ludesnms thatork on a stationary device(such as a computer); wearable devices(such as smart glanted devices(such as brain interfaces); and in specific user environments(such as
About the editorMichael Erbschloe has worked for over 30 years performing analys is of theeconomics of information techno logy, public policy relating to technology, andutilizing techno logy in reengineering organization processes He has authoredseveral books on social and management issues of information technology thatere published by Mc Graw Hill and other major publishers
He has also taught atseveral univers ities and developed technology-related curriculum His career hasfocused on several interrelated areasTechnology strategy, analysis, and forecastingTeaching andcurrIcelopmentWriting books and articlesPublishing and editingPublic policy analys is and program evaluatiBooks by michael ErbschloeSocial Media Warfare: Equal Weapons for All(Auerbach Publications)Walling Out the Insiders: Controlling Access to Improve Organizational Security(Auerbach PublicPhysical Security for IT(Elsevier Science)Trojans, Worms, and Spyware(Butterworth-Heinemann)Implementing Homeland Security in Enterprise IT(Digit al Press)Guide to Disaster Recovery(Course TechnologySocially responsible IT Management Digital Press)Information Warfare: How to Survive Cyber Attacks(Mc Graw Hill)s Guide to Privacy Management(Mc Graw HillNet Privacy: A Guide to Developing Implementing an e-bus iness Privacy Pl(Mc Graw Hill)
introductioArtificial intelligence(Al) is a transformative techno lo gy that holdoc ital and economic benefit al has the potential to re vo lutionize how wek, learneconomic prosperity, improved educational opportunities and quality of life, and enhancedational and home land security Because of these potential benefits, the Us government hasinvested in al researcnany years
Yet, as with any significant techno logy in which theFederal go vernment has interest there are notremendopportunities but aof considerations that must be taken into account in guiding the o verall direction of Federallyfunded r&d in alIn 1956, researchers in computer science from across the United States met at DartmouthCollege in New Hampshire to discuss seminal ideas on anemerging branch of computing calledartificThey imagined a world in which"machines use language,and concepts, solve the kinds of problems now reserved for humans, and improvethemselves" This historic meeting set the stage for decades of government and industry researcin Al, including ad vances in perception, automated reasoning/planning, cognitive systems,achine learning, natural language processing, robotics, and related fields Today, these researchadvances have resulted in new sectors of the economy that are impacting our everyday livesfrom mapping technolo gies to voice-assisted smarthandwriting recognition for malivery, to financial trading,o spanmore Al ad vances are also providing great benefits to our social wellbeing in areas such astainability, education, and public welfareThe increased prominence of AI approaches over the past 25 years has been boosted in large partbfstad probabilistic me thods, the availability of large amounts of data,and increased computer processing power Over the past decade, the AI subfield of machinelearning, which enables computers to learn from experience or e xamples, has demonstrateincreasingly accurate results, causing much excitement about the near-term prospects of AlWhile recent attention has been paid to the importance of statistical approaches such as deep
learning, impactful Al adhacty of otcontrol theory, cognitive system architectures, search and optimization techniques, and mantheThe recent accomp lishments of al have generated important questions on the ultimate directionand implications of these technologies: What are the importantific and techno lo gical gapsin current AI technologies? What new Al advances would provide positive, needed economicand societal impac ts? How can Al techno lo gies continue to be used safely and bene ficially? Howbe designed to align with ethical, legal, and societal principles? wimplications of these ad vancement for the Al r&d work force?The landscape for ai r&d is becoming increasingly complex While past and presentha ve also becef industries andprofit organizations
This investment landscape raises major questions about the appropriate roleFederal investments in AL, especially regarding areas and timeframes where industry is unlikelyto invest? Are there opportunities for industrial and international r&d collaborations thatadvance US prioritiesill the new administration pursue this approach or dumb down thestrate gy because of the fear of facts, science, and opposing opinions?
About the national sciend Technology counciThe nationance and Tec hnology Council (NS TC) is the principal means by which theExecutive Branch coordinates science and techno logy policy across the diverse entities that makeup the Federalh and de velopment(R&D)enterprise One of the NSTC's primarygy investmThe nstc prepares R&d packages aimed at accomp lishing multip le national goals TheNSTC's work is organized under five committees: Environment, Natural Resources, andSus tainability: Home land and National Security: Science, Technology, Engineering, andMathematics (stem educecience;and Technolo gy Each of these committees overseessubcommittees and working groups that are focused on different aspec ts of science andtechnologyMoreinformationisavailableatwwwwhitehousegov/ostp/nstc
About the Office of Science and Technology policyThe Office of Science and Technolo gy Policy (OSTP) was established by the National Scienceand Techition and PrioAct of 1976 The misfaSTthreefold; first, to provide the president and his senior staff with accurate, re levant, and timelylesBranch are informed by sound science; and third, to ensure that the scientificand technical work of the Executive Branch is properly coordinated so as to provide the greahe director of ostp also serves as assistant to the president for science andTechnolo gy and manages the nstc more information is available atAbout the subcommittee on netwand Information TechnologyResearch and DevelopmentThe subcommittee on Net working and Information Technology Research and Developmen(NITRD) is a body under the Committee on Techno logy (CoT)of the National Science aechnology Council(NSTC) The NITRD Subcommittee coordinates multiagency re search andevelopment pro grams to help assure continued Us leadership in networking and informati
techno logy, satisfy the needs of the Federal Government for advanced networking andd accelerate de ve lopment and dep loand information technology It also implements relevant provisions of the High-PerformanceComputing Act of 1991(P L 102-194), as amended by the Next Generation Internet ResearchExce llence in Techno lo gy, Education and Science( COMPETES) Act of 2007(P L 110-69)
For(liNk:https:/wwwnitrdgpveww/naionalairdstrategicplamaspx
National artificiaResearchand De velopment strategic planSubcommittee on Machine Learning and Artificial intelligence, to help coordinate Federaactivity in Al This Subcommittee on June 15, 2016, directed the subcommittee on Networkingand Information Technolo gy Research and Deve lopment (NITRD) to create a National Artific ialIntelligence Research and Development Strategic Plan A NTRD Task Force on ArtificiaIntelligence was then formed to define the Federal strategic priorities for Al R&d, withparticular attention on areas that industry is unlikely to addressgenceFederally-funded Al research, both re search occurring within the go vernment as well aFederally-funded research occurring outside of government, such as in academia The ultimatoal of this research is to produce new ai know led ge and techno lo gies that provide a ran gepos itive benefits to society, while minimizing the negative impacts
To achieve this goal, this alR&d strategic Plan identifies the fo llo wing priorities for Federally-funded Al researchStrategy 1: Make long-term investments in Al research Prioritize investments in the nextgeneration of al that will drive disco very and insight and enable the United States to remain aworld leader in AlStrategy 2: Develop effective methods for human Al collaboration Rather than replachumans, most ai systems will collaborate with hResearch is needed to create effective interactions between humans and al systtrategy 3: Understand and address the ethical, legal, and societal implications of Al Weexpect Al technolo gies to behave according to the formal and informal norms to which we holdour fellow humans Research is needed to understand the ethical, legal, and social implications oAL, and to deve lop methods for designing AI systems that align with ethical, legal, and societaStrategy 4: Ensure the safety and security of Al systems Before AI systems are in widespreaduse,assurance is needed that the systems will operate safely and securely, in a controlled, we
defined and well-understood maFurthh is needed to address thhallenge of creating Al systems that are reliable, dependable, and trStrategy 5: Develop shared public datasets and environments for Al training and testing Thedepth, quality, and accuracy of training datasets and resources significantly affect aesponsib le access to high-quality datasets as well as to testing and training resourcesStrategy 6: Measure and evaluate Al technologies through standards and benchmarksga gement that guide and National artificial Intelligence Research and Development StrategicPlan evaluate pro gress in Al Additional re search is needed to develop a broad spectrum oevaluative techniques to evaluate progress in Al Additional research is needed to develop abroad spectrum of evaluative techniqueStrategy 7: Better understand the national Al R&D workforce needs Advances in Al wrequire a strong commuR&d workforce demands in AI is needed to help ensure that sufficient Al experts are availableto address the strate gic r&d areas out lined in this planThe ai r&d Strate gic Plan closes with two recommendationsRecommendation 1: De velop an AIR&d implementation frame work to identify s&pportunities and support effective coordination of ai r&d investments, consistent wStrategies 1-6 ofthis plan
Recommendation 2: Study the national landscape for creating and sustaining a healthy alR&d workforce, cons is tent with Strategy 7 of this plaThis plan makes se veral assumptions about the future of Al first, it assumes that Atechnologies will continue to grow in sophistication and ubiquity, thanks to Al R&d investmentsby government andthis plan assumes thatnt, educatdwell as the impact on US economic growth Third, it assumes that industry investment in Alwill continue to grow, as recent commercial successes have increased the perceived returns on
tment in R&D At thetime, this plan assumes thatrtantof researchunderinves tment problem surrounding public goods Lastly, this plan assumes that the demandfor Al expertise willtry, acadgovernment, lead iprIviDesired OutcomeThis Al R&D Strate gic Plan looks beyond near-term Al capabilities toward longer-termnpacts of AIcty and the world Recent ad vanAIsignificant optimism about the potential for Al, resulting in strong industry growth aialization of Al approaches
However, while the Federal government can leverageinvestments in Al, many application areas and long-term research challenges will notFederal government is the primary source of funding for long-term, high-risk research initiatives,as well as near-term develop mental work to achie ve department- or agency-specific requirementsto address important societal issues that private industry does not pursue The Federalgovernment should therefore emphasize Al investments in areas of strong socie tal importancethat are not aimed at consumer markets--areas such as AI for public health, urban systems andsmart communities, social welfare, criminal justice, environmental sustainability, and nationasecurity, as well as long-term research that accelerates the production of al know ledge andtechno lo giesA coordinated R&D effort in Al across the Federal go vernment will increase the positive impacf these technologiesthe kno wled:ded to addrespolicy challenges re lated to the use of Al A coordinated approach, moreo ver, will help theUnited States capitalize on the full potential of Al technolo gies for the betterment of societywork that can be used to identifyscientific and techno logical gaps in AI and track the Federal r&d investments that are design