Where Industry and Security Intersect
What's New | Sitemap | Search
Critical Technology Assessment of the U.S. Artificial Intelligence Industry
Artificial intelligence (AI) is an emerging technology of strategic importance to the military and of increasing importance to the international competitiveness of U.S. corporations. AI's greatest value is automating and increasing the utilization of expert or organizational knowledge (i.e., knowing what to do with given information or circumstances). Although AI technology is still in its early stages of development, many spectacular success stories in both military and major corporation applications serve as testimonials to AI's potential. The best AI systems save companies and the government millions of dollars a year.
The United States leads the world in nearly all aspects of AI technology, largely due to over 30 years of patronage by the Department of Defense (DoD). Currently, the United States alone accounts for over 60 percent of an estimated $900 million global AI market. Cutting edge (usually very expensive) AI applications, accounting for almost two-thirds of the U.S. market, are normally developed internally by DoD and a handful of major corporations (mostly) in the information technology field (AT&T, DEC, IBM, Apple, Intel, etc.).
More mature and proven AI technologies are packaged and marketed by merchant vendors. Most merchant AI systems are small in scope, and range in price from about $100 to $250,000. An estimated 70-80 percent of the Fortune 500 companies use AI technology to varying degrees. The merchant market, comprising about one-third of the global market, is intensely competitive and innovative, but not profitable. Between 1988-1993, merchant revenues more than doubled. However, vendors came and went by the dozens. About five or six leaders have emerged, but in 1993 only one had sales of over $40 million.
The United States leadership position in AI is eroding as the governments and companies in Japan, and as well as in Western European, working together, have gained ground. In select areas of AI, Japan and Western Europe now surpass the U.S. Two major conditions threaten the country's leading position: 1) the slow rate of AI commercialization relative to extensive research and development expenditures, and 2) declines in Defense research and development funding. Cuts in the military budget have reduced overall AI R&D, shifted research away from basic science and long-term projects, and threaten to delay AI's potential from being realized. The commercial sector will not come close to replacing these lost R&D funds, particularly at the basic research level.
The establishment of an AI experts group, with industry, academic, and government participation (representing commercial and defense interests), is needed to provide focus and direction to AI R&D and commercialization issues.
This AI assessment was undertaken under the Defense Authorization Act for Fiscal Years 1991 (Public Law No. 101-310, Section 825) and the National Defense Authorization Act for Fiscal Year 1993 (Public Law No. 102-484, Section 4215). This legislation requires the Departments of Defense and Commerce (acting through the Under Secretary for Export Administration) to submit reports to the Armed Services Committees of the Senate and the House of Representatives on the status of technologies deemed essential to the performance of current and next generation weapon systems, and crucial to the commercial sector's ability to compete in the global economy.
The goal of this assessment is to provide industry executives and government policy makers with comprehensive information and analysis about research into and production and application of AI technologies in the military and the U.S. economy. This includes analysis of the economic performance and international competitiveness of private sector firms and academic institutions involved in the creation, distribution, and use of AI, and the impact declining defense budgets have on the technology. In achieving this goal, the Department of Commerce's Office of Industrial Resource Administration (OIRA) collected information from the public and private sectors with a survey questionnaire and sought expert advice as necessary.
AI refers to highly engineered computer software programs used to make computers do things that appear intelligent - such as reason, learn, create, understand human speech, or solve problems. As a science, AI studies the nature of intelligence, and tries to make computers simulate intelligent behavior. As a technology, AI is used to automate (or extract and synthesize) knowledge from information and databases. AI has evolved into many specialized areas and approaches. Major areas and sub-areas of AI research include:
| automatic programming |
knowledge representation |
planning |
| decision making | knowledge acquisition | robotics |
| expert data base systems | logic programming | search |
| expert systems | machine learning | speech recognition |
| fuzzy logic | natural language processing | theorem proving |
| game playing | neural networks | uncertainty |
| general problem solving | understanding systems | |
| intelligent computer-aided instruction | vision |
(See Appendix A for definitions.)
AI systems can today: 1) help organizations manage knowledge assets and deal with complexity; 2) help experts solve difficult analysis problems and design new devices, 3) learn from examples; and 4) provide answers to English questions using both structured data and free text.
AI is a dual-use technology. The same AI system shells or tools can be readily applied across most military and civilian applications. In terms of sales, the most successful and dominant AI tools to date have included knowledge-based systems, neural networks, fuzzy logic, and natural language systems.
In 1993, the global market for AI systems was estimated at about $900 million. The North American (NA) portion of the global market was roughly $600 million of this total. Slightly over $200 million of the NA portion, or about one-third, was sales and license fees collected by AI vendors (the merchant market). An estimated 20-30 percent of this merchant market was exported. Major AI tools include the following:
Knowledge-based (or expert) systems (KBS) have experienced the most commercial success to date. They are used for diagnostic, scheduling, planning, data synthesis, and tutoring purposes, and for automating manuals and running factories. The NA market for KBS totaled an estimated $350 million in 1993. This included estimates of $171 million developed internally by corporations, about $30 million for off-the-shelf systems and embedded components, and $150 million in AI vendor sales and licensing fees.
Neural networks were long neglected, but are projected to grow at a fast rate in the next decade, and perhaps overtake KBS systems. Neural networks efficiently handle huge databases. They are exceptionally good at pattern recognition. Neural network systems are used by the brokerage houses to predict stock fluctuations, by banks to detect fraud, by insurance companies to appraise applicants, and are finding their way into the factory, where, for example, they can sort fruit at incredible speeds. Neural networks and fuzzy logic systems lumped together totaled $150 million in 1993. Neural net vendor revenues totaled about $26 million, and fuzzy logic vendors' about $6 million. As the numbers attest, the bulk of these systems were developed internally by government and corporations, or used as embedded components in other software programs.
Fuzzy logic is the latest rage, although total sales are still small. Invented in America 30 years ago, the Japanese embraced the technique and now lead the world in nearly all aspects of fuzzy logic. Fuzzy logic is used mostly as a control mechanism in camcorders, anti-lock braking systems, elevators, transmission controls, washing machines, and many other products.
Natural language systems (NLS), particularly speech recognition, have benefited from significant recent advances in the technology. NLS systems are used to interface between humans and machines, and to allow human-machine communication in English in place of a rigid set of commands. This includes two-way communication. Using NLS, a manager can dictate a letter into a machine and have it printed or simply transmitted (paperless option) across the country. The NLS market was estimated at $64 million in 1993.
AI is not a "stand-alone" technology, and to an extent has lost identity as a distinct product in the marketplace. The technology is increasingly embedded or integrated in multi-purpose software packages where it can, for example, increase the productivity, performance, and user friendliness of the package. As an emerging technology with customized applications across the economy, AI is difficult to define as a distinct industry. Based on what it does, however, AI is combining with several other software technologies and emerging as the Knowledge Automation Industry. This industry is an important component of the rapidly expanding information age, playing a unique role in automating the process of converting information into knowledge.
(A software industry that facilitates and improves the ability to leverage the value of an organization's in-house knowledge and experience.)
Artificial Intelligence makes -- it easier to manage knowledge and experience assets
Object Oriented Programming -- makes it easier to manipulate code
Computer Assisted Software Engineering -- makes it easier to manage software development
Client-Server Networks -- makes it easier to connect people and computers in an organization and promotes teamwork
The AI sector can be divided into three overlapping components. These are research, commercialization, and applications. The research component is predominantly funded by the Federal Government. The commercialization component transfers the technology to end-users for a price, or more often, a cost if done internally. The applications component is made up of customers -- the businesses and government agencies that purchase AI technology packaged in a variety of ways. Its true value lies in the increased competitiveness conferred on businesses using the technology. This value is difficult to quantify, but substantial.
Research Component - Total research has hovered just above $200 million per year since 1990. The top five research organizations represent almost 40 percent of all AI research and over 50 percent of total basic research in AI in the United States. In terms of research capabilities, the United States is preeminent, but several foreign schools (e.g., University of Edinburgh) and companies (e.g., Siemens) are the equal of the best research organizations in the United States.
The Federal Government sponsored 75 percent (about $150 million per annum) of the AI research in the United States between 1989-1994. Over 80 percent ($125 million per annum) of the Federal total came from the Defense Department. The Advanced Research Projects Agency (ARPA) has been the champion of AI research and development almost from the Agency's inception in 1958.
In the past few years, pressures mounted within the Defense community to show results amid declining defense budgets. As a result, basic research, which accounted for about half of total research expenditures in 1993, is on the decline relative to applied research and development. Basic research is categorized as high-risk, long-term and is often the first area to be cut when budgets and priorities tighten. U.S. businesses sponsored about 15-20 percent of the research, while funding from foreign sources accounted for most of the remainder.
The share of overall AI research undertaken by universities and firms is about equal. Private internal research, which is more focused on short-term applied research and development, could not be quantified, but may range from $50-100 million. This research is undertaken by major corporations involved in the information technologies (AT&T, IBM, Intel, etc.).
Commercialization Component - The merchant market is intensely competitive in marketing primarily proven AI technologies. Most AI vendors are small and entry into the business is low cost. Because of easy entry, the market is perhaps overpopulated. Most vendors lodge themselves in a "niche" where they try to survive. Many have gone out of business while others continue entering the business. The largest vendor had sales of less than $50 million in 1993.
The overall AI market (merchant and internal development) has a high and low end. The high end or leading edge of the industry develops highly sophisticated AI systems that often merge the best of several AI technologies into a single system. These systems are super-engineered expert systems that in a very real sense exhibit capabilities near the human expert level. Leading-edge AI systems are typically developed for and often by major corporations, or the Department of Defense and other large organizations where the savings and productivity opportunities are the greatest. The development of leading-edge systems frequently includes university or think tank participation.
The lower end of the market, the merchant market, is comprised mostly of AI vendors marketing prepackaged proven AI technologies. The AI tools range from simple to complex, are usually inexpensive and, as with advanced systems, can include several integrated AI techniques. An AI vendor may simply sell or license an empty "shell" (or programming format). Vendors typically also offer development services at the customer's option.
With some notable individual company exceptions, the merchant market has not been profitable. Survey respondents as a group for the years 1989-92 showed profits in only one year, 1990. Total revenue for the four year period totaled more than $224 million, with losses of about $11 million. Much of the loss was due to the collapse of the mainframe business. To supplement the survey data, a review of the financial statements of four major publicly held AI vendors (not part of the survey sample) from 1988-1993 provided a similar picture of the financial difficulties many AI vendors have experienced. Total revenues by the four during the 6-year period were $473 million. Losses on those revenues were $106 million (i.e., a net loss of 22 percent). The financial health of the entire industry should improve over the next few years as market conditions improve. AI techniques are being integrated with other software (client-server, object oriented programming) as these vendors adapt to the changing needs of corporate computing.
Applications Component - Military - The Department of Defense is by far the single largest user of AI in the world. In using AI as a strategic asset and management tool, the military has demonstrated the feasibility of the technology, pushed its development, and, through its use, improved military effectiveness. The military uses AI systems for diagnostics, testing, robotics, target recognition, tutoring, war planning, logistics, nuclear test-ban monitoring, database management, and defense- related manufacturing.
Expert diagnostic systems are used extensively by the military to troubleshoot, and maintain military equipment operational. An important value of these systems is facilitating repairs by non-experts on equipment that otherwise could be out of commission for extended periods. The military has also been the major sponsor of research and development in the robotics and machine vision areas of AI. Major projects include autonomous land and aircraft vehicles, automatic target recognition systems, sonar discrimination systems, and navigation aids.
AI systems proved their strategic value in support of operations Desert Shield and Desert Storm. For example, DART (Dynamic Analysis and Replanning Tool) solved the logistical nightmare of moving the U.S. military assets to the Saudi Desert. The application was developed to schedule the transportation of all U.S. personnel and materials such as vehicles, food, and ammunition from Europe to Saudi Arabia. This one application alone reportedly more than offset all the money the Advanced Research Projects Agency had funneled into AI research in the last 30 years.
Another example is AALPS (Automated Airload Planning System), a military airlift load planner used by the Army and Air Force to maintain the aircraft's center of gravity, through evaluation of the shape and weight of each piece of cargo. AALPS was designed using a graphical interface depicting the aircraft, and point and click-on icon representations of helicopters, trucks, and other cargo are used to position cargo in the aircraft hold. AALPS reduced the time required to generate and modify cargo loads from about a week to about one hour.
AI is also used to monitor the Nuclear Test Ban Treaty through an intelligent system, IMS (Intelligent Monitoring System), that automatically detects, locates and identifies underground nuclear tests. It incorporates expert systems, fuzzy logic, neural networks and semi-automated knowledge acquisition.
Applications Component - Commercial - The profile of use in the private sector is different from the military. Where the military community develops a very large and expensive AI system from beginning to end to meet a particular objective, the commercial sector uses more off the shelf, less expensive AI systems to help them achieve more micro-efficiency objectives. AI technology is used in virtually every sector of the commercial economy. About 70-80 percent of the Fortune 500 firms now use AI to varying degrees. Major application areas are in manufacturing, diagnostics, tutoring, financial services, transportation, and data management.
The most applications (about 25 percent) were reported in the manufacturing sector. These included applications in the chemical, steel, auto, electronics, computer, aerospace, and plastics industries. They involved design and engineering, process control, scheduling and planning, part making, factory automation, inspection, and monitoring. AI is also a core technology used in computer integrated manufacturing. In the data management area, for example, an AI program automatically processes and indexes news wires into almost 700 categories for Reuters News Service. For this system alone, savings were estimated at more than $1.25 million in a recent year.
Another major commercial category is diagnostics and testing. Diagnostic systems are used to examine aircraft engines, human hearing, telephone networks, manufacturing machinery and other types of equipment, energy pipelines, ground water and hazardous materials.
A third major segment is transportation services. AI is utilized for traffic management systems, aircraft maintenance operations, airport gate scheduling, railroad planning and forecasting and barge to tow boat assignments. These AI systems are used to manage and draw optimal scheduling decisions from large volume, complex and dynamic databases that would overwhelm human beings.
AI has improved the competitiveness of its commercial users by increasing productivity, improving quality, and augmenting marketing. It has also expanded user capabilities and task performance in ways not previously feasible. It can also induce faster, more consistent and accurate communications, improve service, and improve such intangibles as company or organizational images and customer satisfaction.
AI technology is a small, but influential part of the much larger general software industry. It represents the leading edge of computer software. AI research undertaken years ago has contributed a great deal to the usefulness of today's conventional software. For example, AI research in the 1960s and 1970s was directly responsible for windowing, spreadsheets, e-mail, spell-checkers, chess programs, and many other system components.
AI is still an emerging technology. Continued research is essential to its long-term development. While many AI techniques have attained commercial viability, improvements are needed to further expand markets. In other cases, such as machine learning and robotics, major research remains undone.
Three major points need to be understood about AI as an emerging technology. First, it is revolutionary in that it potentially raises productivity 10-fold or more, and requires many changes in the mind-set of management and people using it. It can, therefore, have an enormous impact on international competitiveness. Second, as an emerging technology, AI requires sustained long-term research. However, AI lacks a solid constituency (critical mass) in the commercial market to support that research. Federal leadership and patronage, critical to AI's long-term development, is declining primarily due to DOD spending cuts.
Third, AI is experiencing market acceptance problems. It can take a major education effort to introduce a new technology, which from a vendor's standpoint means extensive customer consultations, and salesmanship. From a customer's standpoint it means investment in worker training and equipment. If the AI technology is revolutionary, it may even entail a re-engineering of entire business organizations, and require an often unwelcome paradigm shift in the thinking of business management.
Lack of market acceptance is also partly related to the fact that AI technology came out of the laboratory before the corporate computing world was ready. In 1980, corporate computing was mainframe based, centrally controlled, and focused on accounting and payroll. Early AI vendors tried to get corporate computing management to re-engineer the way they used these systems before corporate managers understood the dramatic changes taking place in computer technology. However, advances in computer technology have forced change. Now corporate computing is becoming multi-platformed, decentralized, and involved in every facet of the business. This change is favorable to the use of AI systems and the other components of the knowledge automation industry.
As previously noted, the Federal Government, notably DoD, has played the preeminent role in the development and commercialization of AI technology. In the future, as AI technology becomes more a part of mainstream software and simpler to use, non-Defense agencies can be expected to increase their use of AI relative to Defense usage. However, Defense will remain the major user for the foreseeable future. Perhaps more importantly, Defense will continue taking AI technology out of the lab and fielding first time applications.
The government's role can be divided generally into four elements --
1) Fund Research: Sponsor funding for basic and applied research.
2) Purchase the Product: Develop and deploy new and existing AI techniques that a) enable the government to accomplish its mission; b) improve government efficiency and services to the public; c) demonstrate the feasibility of the technology; and d) provide an initial market for the private sector.
3) Manage Business Environment: Provide legal, regulatory, and educational infrastructure to foster the development and use of AI technologies.
4) Leverage Risk: Provide leadership in promoting cooperative agreements. Promote dual use and technology transfer to private sector and stimulate private investments in new technologies.
Government Funded Research - Universities look to the Federal Government for about 80 percent of their research funding. The surveyed universities were in general agreement that the Government needs to continue funding basic and applied AI research if the United States is to maintain its leading position. AI companies (about 75 percent of funding from USG) also look for the Government to fund research. The consensus viewpoint of companies surveyed is that the commercial sector will not pay for the required AI research, particularly at the basic level; therefore, they look to the Government to fill that role.
Government AI Purchases - Statistics on the use of AI in the Government exist only on an partial basis. Comprehensive data are not available, and no system is in place to collect it. However, the Government is unmistakably the largest single user of the technology. In so doing, the Government has contributed steadily to AI's development as a commercial product. On numerous occasions DoD has been the first user of new AI technology, demonstrating its feasibility.
Many government agencies, or bureaus within those agencies, have formalized AI groups that understand the technology and actively promote the technology transfer of AI within their areas. The U.S. Army has the largest such group by far. The Army group, with several hundred people, has many ongoing AI projects. One particularly impressive AI project is called Blacksmith. Blacksmith, which manages over a terabyte (trillion bytes) of data, is a management decision tool that will enable military planners to simulate with high precision how a change in policy, however large or small, will effect Army force structure and capabilities throughout the world, and thus allow decisions to be made with foreknowledge of the consequences.
The Navy and Air Force were slower to adopt the concept, but they now too have such AI groups. Other agencies, such as NASA, the Internal Revenue Service, Social Security Administration, National Library of Medicine, and FBI, to name a few have also adopted the specialized group approach.
The DoD has long been the major patron of AI research, and the leading developer and user of AI in the world. A decline in the Department's AI activities could severely slow down basic and applied research as well as AI development. Such a slowdown would adversely affect the long-term competitiveness of the United States. A major concern is that as the Defense budget declines, other Federal agencies will not take up the slack.
University AI labs and private think tanks will absorb the brunt of Defense cutbacks in AI research. A major AI institution characterized federally funded research as a national resource in terms of transfer of government supported technology developments to the commercial sector. Projected downturns in DoD R&D expenditures will impair technology transfer efforts in AI and many other fields and, as a result, impact U.S. competitiveness.
Among AI companies, the most common impact of declining defense funding will be registered as declines in sales to Defense (and to prime contractors). Many AI vendors noted that defense sales have already declined, some sharply. Other companies reported a negative impact on their research activities, and on jobs. Some firms also reported the impact of Defense cuts would be minimal, but added that there could be increased competition as firms shift from defense to commercial sectors.
In 1993 the United States AI market was about twice the size of the rest of the world combined. However, experts agree that this huge lead is diminishing as other countries appear to be incorporating AI systems at a faster rate than is occurring in the United States. Technology application appears to be a key factor in the competitive world market. In this respect, other nations, with stronger collaborative efforts between government, industry, and academia, are doing a better job in applying AI technology than the United States.
| Summary of Company and University
Views of U.S. International Standing in Artificial Intelligence |
||||
|---|---|---|---|---|
| U.S. Standing | Companies | % of Total |
Universities |
% of Total |
| U.S. leading | 3 |
8.8% |
2 |
9.1% |
| U.S. leading, but losing ground in some areas | 4 |
11.8% |
1 |
4.5% |
| U.S. leading, but losing ground in most areas | 7 |
20.6% |
3 |
13.1% |
| U.S. lead eroding, behind in some areas | 17 |
50.0% |
13 |
59.1% |
| U.S. lead eroding, behind in more and more areas | 3 |
8.8% |
3 |
13.6% |
| Total Responses | 34 |
100.0% |
22 |
100.0% |
Universities responding to the OIRA survey noted that the United States is beginning to fall or has already fallen behind international competitors, particularly Japan, in applications of AI. This applies to consumer products, the ability to integrate expert systems with conventional systems, the development of very-large knowledge bases, and business-sector investment in knowledge-based technology. Also, the United States is behind in the design of intelligent computers which can "reason" over large volumes of data. In addition, the United States lacks database, image, and software standardization, which is having a mixed effect; it encourages versatility but discourages large cooperative projects.
Universities also stated that the United States is behind Japanese production of fuzzy logic hardware and software because of Japan's industrial policy, investment environment, and pragmatic acceptance of unconventional technology. Japan has an "implementation based" research community. Japan took the lead in applying fuzzy systems theory to develop intelligent control systems, and is advancing rapidly with industrial and commercial applications. American institutions need to continue taking the lead in theory, but at the same time develop new ways for turning theory into applications. According to the university respondents, the United States is also beginning to fall behind Germany is some areas of knowledge representation and reasoning.
U.S. AI firms responding to the OIRA survey had similar comments on overall U.S. competitiveness. One company stated that, "In the United States money is spent on research that sits on the shelf." Another company said, "We are falling behind because our government does not support the transfer of R&D technology applications. This is especially serious with decreasing defense spending." Some firms noted that the United States is falling behind in the use of AI by industry. Although we still lead in many manufacturing sectors, the industry's reluctance to use new technology products will allow international competitors to catch up with and overtake the United States.
Various companies reported that the United States is losing ground or is now behind in certain aspects of robotics, neural networks, expert systems, fuzzy logic, and machine learning. In machine learning, most of the key people in inductive logic programming and first order concepts of learning are in Europe and Australia, not in the United States.
Overall, U.S. universities and firms are in general agreement that the United States is: 1) behind Japan in nearly all aspects of fuzzy logic; 2) losing ground in generally all categories of AI research; 3) behind Europe in establishing consortia; and 4) significantly behind the rest of the world in commercializing AI technologies.
The Department of Commerce's Office of Foreign Availability (OFA) in the Bureau of Export Administration provided a section for this assessment on AI activities in foreign countries. The OFA report documented the partnerships between government, industry, and academia in Europe and Japan that provide the mechanisms for successfully commercializing new AI technology. For example, the Europeans instituted the European Strategic Program for Research and Development in Information Technology (ESPRIT) under which European Common Market nations collectively sponsor research partnerships in AI and related technology. In Japan, MITI recently established the 10-year Real World Computing (RWC) program that will focus on optical computing, massively parallel processing, and neural systems.
1. AI is an emerging technology of strategic importance to the military and of increasing importance to the international competitiveness of U.S. corporations.
2. The U.S. leads the world in nearly all aspects of AI technology, largely due to over 30 years of patronage by the Department of Defense.
3. AI systems, large and small, often result in a ten-fold or more productivity increase. These increases are possible because knowledge (i.e., the ability to take a specific action to achieve a goal on given information) is the most underutilized asset in any organization. The best AI systems save companies (and the government) millions of dollars a year.
4. Cutting the Defense budget has resulted in a smaller share of the research dollar going to basic research for AI. It has also resulted in cuts in total research that will not be made up by increases from other Federal agencies. University AI labs and private think tanks will be primarily impacted by these cuts.
5. The commercial sector will not support adequate basic AI research nor form AI consortia without Federal involvement. Declining R&D spending will have a negative effect on AI technology developments and, over the longer term, U.S. competitiveness.
6. The slow rate of AI commercialization appears to be a weakness of America's AI Industry. Japan and Western Europe, with stronger collaborative efforts between government, industry, and academia appear to be commercializing and incorporating AI systems at a faster rate then is occurring in the U.S.. This is diminishing the U.S. lead in market share.
7. Corporate computing is shifting from centrally controlled mainframe based to widely distributed multi-platform based. Knowledge automation of widely scattered organization information is growing in importance and is stimulating the market for AI technology.
8. Statistical tracking of AI is inadequate to develop informed policy options.