The third workshop set forth two objectives: (1) discuss implications to doctrine, concepts of operations (CONOPS), and command and control (C2) of recent acceleration of battlespace operations, arising from wide-scale digitization, large-scale sensing, and faster technologies (e.g., hypersonics); and (2) consider ways to adapt to fundamental changes in the time constants of conflict.
Workshop Series chair and Workshop Three chair Ms. Deborah Westphal, chairman of the board, Toffler Associates, explained that the final workshop would focus on alternative models for and novel insights on how this era of advanced technology affects the ability to make decisions on various timelines. Dr. Richard Hallion, senior adviser, Science and Technology Policy Institute, said that several of the workshop speakers alluded to the “knowledge sphere” that the U.S. Air Force (USAF) is building around its enterprise, which contains several nodes with their own sub-spheres. If those are connected to the national security enterprise, the resulting structure could have many vulnerabilities. It only takes one individual who fails to follow Communications Security (COMSEC) to create an opportunity for exploitation that could remain undetected and/or inactivated for several years. He noted that these sub-spheres not only create substantial concerns for cybersecurity but also have their own internal frictions that decrease overall speed (see Appendix F for further discussion).
Gen. Gregory “Speedy” Martin (USAF, ret.), GS Martin Consulting, Inc., suggested that the workshop participants discuss the 11 technology areas highlighted in the National Defense Strategy. Because these technology areas can become centers of gravity that impede one another downrange, he commented that the USAF would benefit from the development of a mechanism to understand the use of simple but fast acting technological capabilities. Although the Air Force Warfighting Integration Capability (AFWIC) was designed to coordinate new technologies, he wondered if the National Academies of Sciences, Engineering, and Medicine’s Air Force Studies Board could offer further assistance to AFWIC. In the absence of an organizing construct, he continued, such efforts could consume substantial resources without results. Lt. Gen. Ted Bowlds (USAF, ret.), chief technology officer, IAI North America, emphasized the repeated commentary from workshop speakers on the direct relationship between speed and senior leadership support (whether it is related to acquisition or operations). However, he remarked that if each technology area has its own senior leader who is trying to move a technology forward without communication and collaboration, both disconnection and damage could result.
COMMENTARY FROM LT. GEN. HINOTE
Lt. Gen. S. Clinton Hinote, Deputy Chief of Staff for Strategy, Integration and Requirements, USAF Headquarters, discussed Project Convergence, a U.S. Army experiment held recently at the Yuma Proving Grounds. The experiment focused on a method to unite the joint forces (e.g., via a common C2). Currently, when the time to fight arises, a joint task force is formed with air, land, maritime, special operations, and civil affairs components. The majority of the communication—conducted via telephone—is inefficient. He explained that the Department of Defense (DoD) has begun to realize that the unity required across domains to be able to win is much greater than the unity represented by components and telephones. In addition to organizational changes, technology could be used to improve and speed up both C2 and communication. He described Project Convergence’s data—that recognized the target, described how the target was engaged, and assessed after engaging the target—as particularly impressive. The U.S. Army proposed that it is possible to give more time to decision makers in a complex and uncertain battlespace by increasing the use of machine-to-machine communication, automatic target recognition, and production and prioritization of computer-generated courses of action. Lt. Gen. Hinote wondered, however, if these changes would indeed give back time and whether that time would be used well.
Project Convergence demonstrated that automation could lead to an order of magnitude increase in speed. Lt. Gen. Hinote encouraged a better understanding of the role of automation within a system, how people might have to be trained differently, and whether different skill sets are needed. He depicted the application of artificial intelligence (AI) and machine learning to decision making as an interesting space for thought and discovery. People will have to determine with how much automation they are comfortable. For example, if using AI and machine algorithms (instead of a human) to prioritize courses of action, commander’s intent will be enshrined in the algorithms. The commander would likely have to train his or her own algorithms to ensure that intent is met. The USAF would then have to consider whether contractors should be involved. A model in which the warfighting commanders spend much of their time with their decision-making apparatus, training algorithms to do the right thing in the “fog and friction of war” so that there is trust involved during the fight, is a very different model than what exists for the combatant commander today, in which a majority of the time is spent doing security cooperation and building relationships. This raises a question about how to ensure that the time saved with automation is worth the effort.
Referencing Col. John Boyd’s man–machine observe-orient-decide-act (OODA) loop, Lt. Gen. Hinote described a common belief that making decisions faster would result in a better force. He pointed out Col. Boyd’s emphasis on a decision feedback mechanism to allow for adjustment, a crucial part of decision making. Lt. Gen. Hinote suggested that people consider the value of relinquishing speed to gain feedback and adaptation. He stressed that it can be dangerous to assume that faster is always better; speed may not be the best tactic for the United States to defeat China. The future art of command may be to understand when to move quickly and when to move slowly. If the feedback mechanisms are included at the right level (e.g., operational, geostrategic), he continued, a competition could emerge in which faster is indeed better.
During the question-and-answer session, Gen. Martin expressed concern about whether the USAF is equipped for deception, disruption, and all-domain operations. Lt. Gen. Hinote agreed that the USAF is not yet equipped because all-domain experimentation has not occurred. However, plans are in place for a USAF/U.S. Army experiment on the use of machine learning algorithms to determine best courses of action—for example, shooting with artillery, bombing with an aircraft, creating a cyber effect from an aperture in the air or in space, or shooting a cruise missile from a submarine. He described the new role for modeling and simulation to better understand the enemy. Dr. Hallion observed that speed may have different meanings across the command decision enterprise: At the lowest tactical level, there may be a greater desire for immediate speed and effect in getting something on target; the ability to do that may buy time for decision makers to be more reflective. This involves decreasing the accumulated frictions at every level of the decision-making process. He agreed that using telephones to communicate while the threat moves at warp speed no longer makes sense but added that machine-to-machine and human-to-machine activities create new ethical issues. Lt. Gen. Hinote explained that the USAF has participated in wargames related to increasing the autonomy of machines—automatic target recognition allows humans to do things that they could never do on their own. When considering whether to substitute a machine for a human, it is important to remember that humans make
mistakes. He discussed the difference between having a human “in the loop” (i.e., a human pushes a button to hit a target) and having a human “on the loop” (i.e., a human has the capability to stop an automated process) and observed a movement toward the latter, with a “human overseer” for situations in which the confidence rate of hitting a target is extremely high and authority can be delegated. He noted that the national security community has not resisted this concept as much as he anticipated. Increasing amounts of autonomy will eventually be the norm, he continued, not the exception. Dr. Hallion emphasized that points of interconnectivity will have to be protected so that they cannot be exploited without our knowledge.
ACCELERATING JOINT ALL-DOMAIN OPERATIONS
Col. Doug “Cinco” DeMaio, USAF, deputy director of the LeMay Center for Doctrine Development and Education (ret.), described the close cyclical relationship between time and energy: Producing energy could save time, and saving time could produce energy. Quoting Gen. Charles Brown, USAF Chief of Staff, Col. DeMaio said that the “Air Force must accelerate change to control and exploit the air domain to the standard the nation expects and requires from us. If we don’t change—if we fail to adapt—we risk losing the certainty with which we have defended our national interests for decades.”1
Col. DeMaio explained that the USAF’s scope of operations includes all areas between the ground and the Moon. There are large-scale differences in the importance of time in air, space, and land that have not yet been reconciled, in addition to substantial tradespace (e.g., fast computations in cyber). During World War II, Germany’s CONOPS included wars of movement, commanders, and a vision of operations that the subordinates were empowered to execute. They vied to control air and land, blending those together in operational and tactical units in which the operational commander was overlooking the tactical units that had authority and mission orders. They were connected via a common CONOPS and an effective use of radios. He indicated that people understood the operational art, were well trained, and were encouraged to take risk and innovate in the field. This method was designed around a common network of C2. The air and surface domains were connected by radios, and windows of superiority were created: When the land was bogged down, the air could take over.
Col. DeMaio noted that the concept of “multi-domain” existed during the first and second World Wars but was first discussed in the USAF Future Operating Concept.2Air Superiority 2030 states the Air Force’s projected force structure “is not capable of fighting and winning against this array of potential adversary capabilities . . . [without] a multi-domain focus on capabilities and capacity.”3 He described the seams through which the adversary (or the USAF) can gain advantage and noted that the capabilities of U.S. allies can be leveraged to develop better joint solutions.
In 2016, Col. DeMaio spoke with the USAF Scientific Advisory Board about countering high-speed weapons. The existing Counterair Doctrine integrates offense—attack operations, suppression of enemy air defenses, fighter escort, fighter sweep—and defense—active and passive air and missile defense.4 He commented that a common structure among air, space, and cyber, including a shared lexicon, would enable speed. He initially proposed the Counterspace Doctrine, which similarly integrates offense—attack operations, suppression of enemy space defenses, space escort, space sweep—and defense—active and passive space and missile defense.5 In November 2016, space policy began to change at the national level; the nation began to accept the risk of declaring its mission to control in space. He described the value of creating a similar structure for cyberspace and the electromagnetic
1 U.S. Air Force, 2020, Accelerate Change or Lose, General Charles Q. Brown, Jr., USAF, Chief of Staff, August, https://www.af.mil/Portals/1/documents/2020SAF/ACOL_booklet_FINAL_13_Nov_1006_WEB.pdf.
2 U.S. Air Force, 2015, Air Force Future Operating Concept: A View of the Air Force in 2035, September, https://www.af.mil/Portals/1/images/airpower/AFFOC.pdf.
3 U.S. Air Force, 2016, Air Superiority 2030 Flight Plan, Enterprise Capability Collaboration Team, May, https://www.af.mil/Portals/1/documents/airpower/Air%20Superiority%202030%20Flight%20Plan.pdf.
4 U.S. Air Force, 2016, Air Force Doctrine Publication 3-01: Counterair Operations, September 6, https://www.doctrine.af.mil/Portals/61/documents/AFDP_3-01/3-01-AFDP-COUNTERAIR.pdf.
5 U.S. Air Force, 2018, Air Force Doctrine Publication 3-14: Counterspace Operations, August 21, https://www.doctrine.af.mil/Portals/61/documents/AFDP_3-14/AFDP-3-14-Counterspace-Ops.pdf.
spectrum (EMS). He emphasized that the USAF Doctrine creates windows of superiority by taking minimal losses while conducting operations (see Figure 5.1) and creating multiple dilemmas.
Multi-domain command and control (MDC2) is advantageous, he continued, as the complexity and speed of operations rapidly increase. Linking command and control allows people to talk to each other intellectually and in the physical sense of operations, thereby increasing speed and ability to operate quickly. The ability to integrate domains to create multiple, complex dilemmas for an adversary is dependent on the ability to design and implement an advanced construct. The USAF understands the C2 systems of the other components (i.e., U.S. Army, U.S. Navy, and U.S. Marine Corps) and its allies well; however, it does not have a C2 system of its own. Thus, the USAF is building its C2 system by combining existing and new structures. An MDC2 enterprise capability collaboration team (ECCT) concept emerged with the following elements: (1) Domains (space, cyber, and air); (2) Integration (offense, defense, and deterrence) via intelligence, command, control, and communications; (3) Scope (seamless, dynamic, and continuous); and (4) Outcomes (high velocity, operationally agile, multiple dilemmas, and unmatched tempo). The MDC2 ECCT plan was to begin with a threat-based approach, move to multi-domain operations, and focus on C2 (i.e., lines of effort in operating concepts, enabling technology, and support structures for wargames). The USAF Doctrine was refined, and new Counterair, Counterspace, and Joint All-Domain Operations (JADO) Doctrines were published.
Col. DeMaio next discussed EMS, which is a large tradespace in which U.S. adversaries are competing. He described EMS as its own domain: The Russians are utilizing EMS to create the information-technological space and the information-psychological space, and the Chinese are utilizing EMS to create a network-electromagnetic space. Information and EMS operations focus on degrading adversary decision making and C2. When EMS, which focuses on time and energy, is added to the USAF Doctrine, a more complete representation of the USAF’s challenges emerges (see Figure 5.2).
Col. DeMaio noted that it took 1 year to write the JADO Doctrine. JADO Convergence moves from a single domain, where one can access, control, and exploit in the air; to the cross domain, where something in the air can be used to exploit something on the ground; to all-domain, where several domains are accessing and controlling and creating windows of superiority by planning, creating dilemmas to exploit, and converging together on a target. The JADO Doctrine includes the concept that different domains have different timelines; a synchronization of disparate planning timelines would converge effects. An upgrade to MDC2, JADO includes the following elements: (1) Domains (space, EMS, cyber, air, maritime, and land); (2) Integration/joint functions (offense, defense, and deterrence; centralized control/planning and mission-type orders; and decentralized execution/conditional authorities) via information (fused and shared), fires (convergence), protection (agile employment), C2 (joint all-domain
command and control), intelligence (man–machine), movement and maneuver (integrated tasking order), and sustainment (maneuver logistics); (3) Scope (cooperation, competition, and conflict); and (4) Outcomes (produce decision quality information, converge in multiple domains, present multiple dilemmas, and create windows of superiority) (see Figure 5.3). He emphasized that the USAF is arguably the best at many portions of C2 and that intellectual capital creates speed.
Col. DeMaio next discussed the possibility of moving from the current structure of operations—in which air, space, and cyber are divided up by combatant command and do not plan together until they meet as a joint task force, which is not agile—to an Aerospace Task Force, in which airmen plan and train together, and tailored forces are provided through a USAF Forces Command. He explained that this model for Aerospace Operations provides a unique solution that is better tailored for the joint force: Operating on the same page intellectually leads to faster movement.
Col. DeMaio noted that the Chinese have successfully and successively pushed the United States farther out in range, creating problems related to time, distance, and access. He suggested flying into orbit and holding the enemy at risk from great range—the “trans-atmospheric” gap between the area in which we operate (below 70,000 feet) and where space begins (at 300,000 feet) can be closed and controlled by exploiting cyber and EMS. He said that the United States has low-cost launch and the ability to get into orbit quickly; however, the USAF lacks an integrated concept of warfighting and procures systems by domain. Instead, he reiterated that the USAF could procure, train, and employ as an Aerospace Force. Doing so could enable faster movement owing to multi-domain machines. He added that the United States is accepting great risk if it continues not to operate in this trans-atmospheric region.
Col. DeMaio described Col. Boyd’s OODA loop, emphasizing the value of the “orient” stage. The United States has to know its adversary and itself, yet knowing does not require acting; sometimes it leads only to deciding and observing. AI and other machine intelligence are best suited to assist the human in the observe and orient stages via data collection. He asserted that trusting in systems and interacting with systems could also increase speed.
He explained that these ideas lead to the Aerospace Task Force Operational Concept, the vision for which is cis-lunar maneuver warfare between Earth and the Moon (which is where U.S. adversaries could establish bases). Aerospace operations hold adversaries at risk in all domains. The CONOPS for the Aerospace Task Force is wing plus building blocks and combined arms (i.e., air, space, EMS, cyber, and information operations). The method is to control in all domains, with the ability to mass and attack from/to anywhere through multiple domains to create multiple dilemmas using man–machine teaming. The people in this task force would be specialists and operational artists, and the machines would operate across domains.
Col. DeMaio discussed several advanced capabilities that could increase speed: directed energy weapons (to achieve the speed of light and magazine depth), hypersonic machines (to exploit the trans-atmosphere and bridge time and distance), quantum computing (to increase processing speed and power and to enable advanced capabilities), machine intelligence (to speed decision making), access to space (to achieve low cost, reusable launch), electricity (via space-based solar power and microwaves), water (to be taken directly from the atmosphere, be self-contained, and be deployable), fuel (from hydrogen turbines paired with advanced batteries), and three-dimensional [3D] printing (to make what is needed). These capabilities could help achieve the Aerospace Task Force Tactical Concept, which has a vision to be a deployable, self-organizing tactical unit with a concept of highly maneuverable building blocks; a method of control by leveraging advanced capabilities to bridge the gap in time and space; with people who are tactical experts but well versed in the operational arc; and a mix of man–machine teaming.
In closing, Col. DeMaio said that time and energy differentials have expanded exponentially. JADO can be accelerated with common concepts, doctrine, and lexicon; common methodologies and planning; advanced computing and connectivity; machine intelligence and man–machine teaming; agile logistics and energy production; and bridging domains with methods, machines, and weapons. JADO has not yet sufficiently considered the impact of warfighting in space, cyber, and EMS in terms of time and energy, but he underscored that the greatest accelerators of JADO are a culture of innovation and a vision of the future.
During the question-and-answer session, Gen. Martin observed that this was the first briefing of the workshop that provided evidence that the United States could win. He shared the following three comments:
- Space does not work without cyber.
- Cyber, C2, and EMS are enabling functions, not missions; until the USAF organizes itself in a way in which EMS can become a mission area, EMS does not have a domain or the control of the other domains.
- When in low Earth orbit, Earth is not far; it only takes a few hundred miles for an electromagnetic or kinetic capability to hit a target, which opens up new possibilities.
Thus, he suggested careful thought about how EMS could become the most significant multiplier for USAF operations. Dr. Michael Yarymovych, president, Sarasota Space Associates, advised caution about predictions related to energy—the Sun only delivers 1,000 watts per square meter to Earth. He added that a presence is required to have direct access through the speed of light. Dr. Hallion agreed with Col. DeMaio that the trans-atmosphere has grown to be a largely unobserved threat region. Ms. Westphal asked how the momentum for this important work continues beyond Col. DeMaio’s retirement. Col. DeMaio replied that he will take this work into his new position
with the National Guard, specifically focusing on making electricity, fuel, and water more mobile. He added that the LeMay Center has the JADO Doctrine, and he hopes that the efforts continue to be explored. Gen. Martin stressed the value of continuity of effort. He inquired about a relationship with AFWIC and the mechanism by which such a vision could move forward. Col. DeMaio responded that the intellectual capital of the LeMay Center is broken down into doctrine, concepts, and wargaming. This work resides in the area of concepts, a group that works directly with AFWIC. Because AFWIC focuses more on the technology, the LeMay Center could actively push this effort forward.
IMPROVEMENT TO MEET CRITICAL NATIONAL SECURITY NEEDS
Gen. Bruce Carlson (USAF, ret.), Commander, Air Force Materiel Command, and director of the National Reconnaissance Office (NRO), shared a motto from Gen. W.L. Creech that he embraced throughout his career: “Make it happen, make it better.” He explained that trust and relationships are more important than processes and directives when trying to improve operations and to increase speed. He described his experience as the 17th director of the NRO from July 2009 until July 2012. He was recruited by the Obama Administration to fix a “broken outfit”: 6 of 14 major programs were either red or yellow for cost, schedule, and/or performance. The first change that Gen. Carlson made was to reduce the number of direct reports. He then introduced an institutional decision-making process, established a budget-development process, and endeavored to declassify hardware and program materiel. This enabled a series of “firsts” that involved space situational awareness and space protection. He emphasized that even as late as 2010, his vision for the importance of space in warfare was not well received.
NRO planned to launch six satellites into orbit within 8 months. However, no tested problem-solving process existed, it had not been done in 25 years (and even in that time more people and resources were available and less sophisticated satellites were launched on a recurring basis), the single satellite transporter would have to be moved from coast to coast repeatedly, and three of the launches required complex Delta IV Heavy rockets. Gen. Carlson was presented with the challenge to determine which launches to postpone, but instead he committed to achieving the goal of all six launches in 8 months. Before describing how NRO accomplished this feat, he shared the principles by which he led the organization:
- The 20/20/20 Rule—Twenty percent of people hear 20 percent of what a person says 20 percent of the time. People have to be told over and over again until they truly listen and believe.
- The 20/60/20 Rule—When a vision is presented to a group, 20 percent will say they have never done things that way and will not start now; 60 percent are unsure; and 20 percent support the vision. The best strategy is to target the 20 percent that support the vision because they can attempt to convince the 60 percent who are undecided.
- The 50/50/90 Rule—There is a 50 percent chance that 50 percent of things will go well and a 90 percent chance that they will not. No matter how much a person plans, mistakes will always be made. Thus, people should be patient, not allow mistakes to get in the way, and learn to plan and execute in real time.
Gen. Carlson explained that to accomplish a significant task, it is important to lead with a vision. Anything worth doing is difficult, and changing is even more difficult. When people are settled into the status quo, he continued, it becomes very difficult to persuade them to do anything else. Change occurs only with complete dedication. It is not feasible to wait to finish one task before starting to improve elsewhere. He commented that most people are predisposed to be cynical as a way to protect themselves when challenged; it is important not to engage emotionally.
To accomplish the launch goal, Gen. Carlson created an NRO public affairs campaign, identified the key players, and assigned specific tasks and deadlines. He became a hands-on leader, continued messaging, met with his team regularly, and reviewed metrics. He also made resources available to ensure that people understood his commitment (e.g., securing waivers to hire the right colonels for the job). He asserted that this was no longer just an NRO issue—because others wanted to launch satellites, this became a national priority. Thus, it was important to minimize errors so as to avoid a domino effect. He established a single office to manage this cross-agency chal-
lenge. One of the more difficult problems to overcome was the issue of the lone satellite transporter. However, because of a trusted relationship, the NRO System Program Office and contractor agreed to ship early and invented a new process to move satellites from the contractor airlock into the processing facility airlock. The first launch, which was on time, set the tone for the rest of the campaign and silenced the NRO naysayers. Next, a mock-up of the processing facility was created—with enough practice moving these school-bus-sized satellites with less than 2 inches of clearance, NRO ultimately achieved all six launches within 8 months. When Gen. Carlson departed NRO, all programs were green for cost, schedule, and performance. In closing, he said that a leader’s role is to train his/her replacements, to mentor others, and to create a vision.
During the question-and-answer session, Dr. Julie Ryan, chief executive officer, Wyndrose Technical Group, reiterated the importance of trust and relationships. She asked Gen. Carlson how often he had to refresh his knowledge base and broaden his educational perspective during his more than 37-year career. Gen. Carlson replied that he had to learn continuously throughout his career. First, he had to learn in order to compete in Fighter Weapons School. Second, when he became a weapons officer in a wing that had never had a weapons officer, he had to start from scratch. Third, when he went to work at the Pentagon, he was given only 90 days to solve a problem and had to learn to communicate with people at a different level and navigate the bureaucracy. Fourth, as Director of Requirements, Headquarters Tactical Air Command (HQ TAC/DR), he had to learn an entirely new skillset to write the requirements for the F-22. Fifth, when he became the 8th Air Force Commander, he had to learn about bombers, computer networks, and intelligence. Lastly, he knew nothing about rocket flights before arriving to the NRO. He emphasized that the system has a way of rewarding people who are willing and able to change. He added that a knowledge sphere is porous; one cannot operate alone, and a willingness to delegate and take risks is vital.
Ms. Westphal described the previous USAF focus on mission, purpose, and CONOPS that was passed down generationally, but the speed at which technology was implemented was not as fast as desired. With advanced technology, which is primarily in the commercial realm, things move more quickly. She wondered if there will be an inflection point in the future when there could be a danger of losing the focus on warfighting. Gen. Carlson noted that people resist change either because they believe they will lose something or because they are unable to adapt. A new model with an emphasis on empowerment and delegation could be useful in a time of such rapid change, he continued. Delegation farther down the chain is especially important when senior leaders do not approve innovative plans because they are afraid of change or do not understand the speed of technology. He stressed that the USAF would benefit from hiring and empowering young people who understand and appreciate advanced technologies. Even in a mechanical, computer-driven world, the key to success is developing trusted relationships. Dr. Hallion mentioned a reoccurring theme about the role of the human and human relations as a natural friction in time cycle reduction. He agreed with Gen. Carlson that in a transparent, information-sphere world, it is crucial to ease those frictions while paying significant attention to COMSEC issues. He added that a leader can have tremendous challenges in dealing with existing, sometimes paralytic bureaucracies. He suggested that Gen. Carlson’s remarks on leadership be shared in a capstone course at the Federal Executive Institute.
Dr. Rama Chellappa, Bloomberg Distinguished Professor, Departments of Electrical and Computer Engineering and Biomedical Engineering, Johns Hopkins University, commented that SpaceX is demonstrating efficient rocket launches; he wondered if NRO’s efforts could be privatized to increase efficiency. Gen. Carlson thought that NRO’s last launch manifest had 60 percent from United Launch Alliance and 40 percent from SpaceX. The plan included the launch of national security payloads, not just test payloads. He explained that it took time for SpaceX to reach this point because mission assurance tests had to be passed first. Gen. Martin questioned whether Gen. Carlson ever found himself in a situation in which he had to alter his vision based on feedback. Gen. Carlson advised that people always be willing to change and seek input from a small group of trusted individuals.
QUANTUM TECHNOLOGY FOR NATIONAL DEFENSE
Dr. Paul Lopata, principal director for quantum science, Office of the Under Secretary of Defense for Research and Engineering (USD(R&E)), described a suite of cutting-edge technologies that could affect time cycles within the USAF and across DoD (see Figure 5.4). He emphasized the value of building military in-house expertise while pushing science and technology forward. Increased coordination across DoD and other federal agencies has enabled
a better understanding of the benefits of technologies as well as provided impact for civilian, political, and military leaders and feedback for other government organizations. USD(R&E) has several modernization priorities, for which it is working across DoD to develop roadmaps. USD(R&E) does not have a substantial budget to affect change, nor does it have specific authorities to sway DoD agency priorities; however, it has the ability to persuade and to work across the U.S. Army, U.S. Navy, and USAF. He explained that the modernization office has a similar approach to that used by Gen. Carlson: using repetition to portray a message and collaborating with others to expand ideas.
Dr. Lopata highlighted four categories of quantum technology: atomic clocks, quantum sensors, quantum computing, and quantum networking. He noted that quantum physics was discovered in the early part of the 20th century, and it revolutionized the understanding of the atomic world. For many years, quantum physics served as the basis for understanding material and other phenomena, but it was very messy. In the 1980s and 1990s, forward-thinking scientists began to embrace this chaos and turn it in to something useful; people realized that computing, sensing, communication, and networking could be done differently.
He explained the way in which Global Positioning System (GPS) atomic clocks function: Military standard time is broadcast to the GPS satellites, each of which has approximately three atomic clocks on orbit. That provides positioning information to the joint forces around the world as well as time synchronization (~ 10 nanoseconds). Thus, the GPS system has high military impact in terms of force projection. Many of the quantum technologies could improve timing, time transfer, or positioning and navigation for a future scenario in which an adversary denies or degrades the use of GPS. For this atomic clock technology to be used in military systems, however, it would have to be smaller, more robust, more portable, and able to operate in more challenging environments than a temperature-controlled laboratory. Dr. Lopata is working to push this technology forward so that when future systems are designed that require a timing resilience (i.e., the ability to plan for operation when there is no access to GPS), it will be possible to purchase from American suppliers. He mentioned that USD(R&E) is driven by technology, opportunity, and anticipated needs instead of by requirements; this ensures that the work is joint and crosscutting. This also drives the promotion of atomic clocks—to get performance up and scale down (in terms of size, weight, and power).
Quantum sensors, for example, are in some ways inspired by atomic clock technology. Dr. Lopata noted several laboratory demonstrations with magnetometers, gravitational sensors, inertial sensors, and new electromagnetic sensors that can detect oscillating fields. While people who build better atomic clocks are trying to drive down their errors, they discover that a stray magnetic field has the power to change their time. Quantum sensors utilize the fact that these quantum systems start behaving badly when they are subject to unwanted fields, acceleration, or rotation; they design the system to behave maximally badly when subject to these fields, and then that is used as a sensor. Gravitational sensors, gyroscopes, accelerometers, and magnetometers are already being used by the military. He said that the next step is to take laboratory demonstrations of extreme sensitivity and determine whether it is possible to impact military missions (e.g., pre-mission training or intelligence, surveillance, and reconnaissance). The challenge to pushing the technology forward involves bridging the gap between the laboratory and the people who would use the technology or understand the need for the technology. For the past 20 years, he continued, the accumulation of expertise while pushing technology forward has occurred in the services’ basic research offices and laboratories as well as in the Defense Advanced Research Projects Agency (DARPA) and other laboratories across DoD.
He noted that quantum computing often attracts more attention than the other quantum technologies. While quantum computing is in some ways the most revolutionary of the quantum technologies, it is important to remember the value of other quantum technologies. Researchers and investment companies are inspired by the open territory of exploration with quantum computing. Given the number of ways in which high-performance computing is used (from understanding fundamental materials, to designing aircraft, to understanding how to plan for the future), it is possible that there are problems that could be made easier with a quantum computer, which could have high military impact. For example, there are some math problems (e.g., factoring and solving elliptic curves) that become easier with a full-scale quantum computer. However, those mathematical problems are crucial for keeping the nation’s secrets, and cryptography standards are vulnerable with a large-scale quantum computer. He explained that the cryptography systems would have to be updated; the National Institute of Standards and Technology is developing new codes that are not hackable by a quantum computer. He noted that even though the development of a full-scale quantum computer is in the distant future, it is crucial to work toward cryptography modernization, for example, to prevent the systems from becoming vulnerable.
The fourth category of quantum technology, quantum networks, offers the ability to transfer small packets of light from one place to another. Theoretical results show that it is possible to parallelize quantum computers as well as to improve telescopes through multi-aperture telescope interferometry with quantum communication on the back end. Using quantum networks for a specific known military advantage will be a goal for the distant future, however. He mentioned that England’s Government Communications Headquarters published a white paper about the fundamental challenges of quantum technology. Because of these challenges, Dr. Lopata reiterated the importance of having in-house tactical expertise to push the technology forward.
Dr. Chellappa asked how the Department of Energy’s (DOE’s) investments will impact DoD’s progress in quantum technology. Dr. Lopata responded that the focus has primarily been on quantum computing. Quantum radar was discovered not to be functional, which is considered progress since money will no longer be allocated to analyze that technology. He anticipated discoveries related to more limited-scale quantum computers. He commended DOE and the National Science Foundation (NSF) for increased quantum activities and funding initiatives. It is important that DOE and DoD have complementary foci toward technologies with the most potential impact on the military. He also anticipated that DOE will be working on entangled sensor networks. There is fundamental progress being made on repeaters, sources, and detectors. Dr. Yarymovych asked about future impediments to operationalizing these technologies for the military in terms of the physical plant and infrastructure needed to maintain them (e.g., low temperatures, energy supplies). Dr. Lopata explained that the qubit requires temperatures below 100 millikelvin—which necessitates the purchase of an expensive ($500,000) dilution refrigerator that is wired to send radiofrequency signals through many layers of cold and shielding. Each qubit requires at least one wire going toward it, which typically has radiofrequency control traveling down to it, as well as a digital-to-analog converter and waveform generators. These items are very expensive, and it is difficult to synchronize the signals. While these are impediments to progress, he continued, they are not overwhelming and can be addressed with additional funding. Other fundamental challenges include a limited workforce and a lack of profit. Challenges are
fewer with clocks and sensors, which have more clear paths for military utility. Dr. Hallion noted that understanding military impact versus technology readiness level helps to separate the quantum hype from the quantum reality. He said that the United States is in a quantum computing race, and, historically, the nation has not performed well in technology races. He wondered whether the focus on quantum computing should expand. Dr. Lopata replied that he would not characterize the current moment as a quantum computing race. There are very few groups across the world who would take advantage of the cryptography application. Instead of developing a quantum computer, the defense is to update the codes to a system that is not susceptible to hacking through those means. Thus, it is a race to develop those codes and start deploying them to ensure that military secrets remain protected. He said that the United States is on the right track toward cryptography modernization. The current race is also toward understanding which military applications could leverage quantum computers and how soon they could be used. Chemistry, logistics, and airflow problems could be addressed when these machines are smaller, and a virtuous cycle of people buying the current generation of machine could fund the next generation of machine. Right now, however, only government and company research dollars are being spent. The real driver is application discovery and ensuring that cryptosystems are up-to-speed. He thought that hardware technology companies could play a role and that DoD would continue to invest basic research in this area.
In reference to the impact that quantum technologies could have on time cycles, Dr. Lopata explained that atomic clocks provide the ability to share time faster than GPS, which will provide advantages that the adversaries do not have. As quantum technology is better understood, it will become more clear which missions can be sped up. Although opportunities exist for quantum computing, many unknowns remain. Sensor technology enables detection beyond the range of what the adversaries think is detectable. He commented that employing quantum sensors could provide extra mission time to make decisions and react in ways not possible when waiting for objects to get closer or taking multiple passes by an object before being able to understand it. Although some processes will speed up, others will remain the same and be limited by natural time cycles. One of these is discovery—for example, understanding this new type of computing, how to devise new algorithms, and whether that algorithm has an advantage over what is currently possible is a slow process. This discovery process may speed up over time as people develop expertise, but Dr. Lopata expected it to be relatively slow for some time. He commended companies that have their machines available to users on the cloud at low rates. The Air Force Research Laboratory has partnered with IBM and the Naval Research Laboratory to obtain this access, and scientists have been working to better understand military algorithms, which could eventually increase the speed of discovery. The development of component technology would also benefit from increased speed. Transitions (e.g., from laboratory to prototype that can be tested and eventually manufactured and fielded) also remain on a slow time scale. Transitions can be completed with advanced planning, which includes building bridges as the technology moves to the prototype community and giving that community the opportunity to provide input.
Ms. Westphal observed that in the past, the fielding of technology advancements was often halted or delayed owing to logistical or training issues. Dr. Lopata noted that logistical and training issues are low impact for some of the quantum technologies. For example, a high-end gyroscope is a component of a larger system, and the user of the larger system is likely unaware when a gyroscope is updated. Quantum communications technology, however, has a different set of logistical and training challenges. One advantage of quantum technologies is that they are not the weapons systems themselves; they will either be part of the IT infrastructure or a sensor component. More training would only be beneficial if the sensors had different characteristics. Although logistics and training would not be deployed immediately, he continued, they could be considered for the future. Ms. Westphal referenced the National Academies’ report Owning the Technical Baseline for Acquisition Programs in the U.S. Air Force,6 and she commented that understanding what is inside a system will still be important to the military. Dr. Lopata added that developing an understanding among technologists, engineers, and those examining the mission is both very difficult and very important.
Dr. Brendan Godfrey, visiting senior research scientist, University of Maryland, pointed out that injecting reality into discussions of new technologies and their potential impacts is very useful. He echoed Dr. Lopata’s
6 National Academies of Sciences, Engineering, and Medicine, 2016, Owning the Technical Baseline for Acquisition Programs in the U.S. Air Force, The National Academies Press, Washington, DC, https://doi.org/10.17226/23631.
concern that if applications are not developed quickly enough, programs could be cancelled. He hoped that would not be the fate of quantum technologies—this happened with hypersonics, and the United States is paying a high price for it. Gen. Martin asked Dr. Lopata to comment on the roadmap for the use of quantum technologies in military operations. Dr. Lopata replied that DoD has a tremendous amount of expertise in atomic clocks; it has fielded atomic clocks and is working speedily toward achieving the vision of what will be needed in systems over the next decade. Thus, this technology has the clearest roadmap and clearest demand. There is more work to be done on sensors to move into the prototype phase and collect input from the community of users. Although progress in computing has been made, it is a much longer effort than many people realize. DoD is working closely with academia and industry to develop applications for the future. He encouraged the management of expectations; it is early stages for both quantum computing and quantum networking technologies. Dr. Chellappa wondered if the nation is focusing enough on providing people with the appropriate physics and engineering education to build new things. Dr. Lopata mentioned the Science, Mathematics, and Research for Transformation scholarship program and a National Defense Service Fellowship (which has a requirement to work at a DoD Laboratory) that focus on developing these areas of expertise. He has also been working with the services to identify methods to support students and to develop the workforce. He agreed with Dr. Chellappa about the urgent need to engage more domestic students in this field and in technology for government, industry, and civil society.
INTELLIGENT ROBOTS IN SURGERY AND INFECTIOUS DISEASE CRISES
Dr. Russell Taylor, John C. Malone Professor of Computer Science with joint appointments in radiology, mechanical engineering, and surgery, and director of the Laboratory for Computational Sensing and Robotics (LCSR) at Johns Hopkins University; and director of the NSF Engineering Research Center for Computer-Integrated Surgical Systems and Technology, noted that LCSR’s approximately 250 faculty, staff, scientists, and students work on medical robotics as well as human–machine collaborative systems; modeling, dynamics, navigation, and control; robotics in extreme environments; biorobotics; and perception and cognitive systems. Much of LCSR’s work focuses on a three-way partnership among people, technology, and information. He explained that advances in AI and robotics enable a new generation of highly adaptable systems working in partnership with humans to perform useful tasks in society. These systems can significantly transcend human performance limitations, accelerate responses in emerging crises, and promote quick adaptation. A key challenge is mediating between human intention and action to achieve the benefits of the human–machine partnership, especially as the machines become more intelligent, powerful, and complicated.
Dr. Taylor described two use cases of human–machine partnerships. The first considered the role of autonomy and human–machine partnerships in surgical robot systems. A human–robot partnership occurs when a computer makes some or all of the decisions. He emphasized that there are two basic questions about any systems in these partnerships: (1) Can a person unambiguously tell the robot what it is supposed to do? (2) Can a person ensure that the computer that is controlling the robot can cause the robot to do what it is supposed to do? Early surgical robots appeared to have a high level of autonomy. Surgical computer-aided machines take medical images and patient diagnosis records to plan an intervention and use AI, image processing, visualization, simulation, surgical plan optimization, registration, and plan execution that help the physician determine how the intervention will work. When all of that information is taken into the operating room and related to the patient on the table, the technology can be used to help do what is planned and verify. This paradigm is still used today. He said that the earliest surgical robots were radio-surgical systems, which are devices that showed X-rays through the patient in many different directions (e.g., to be able to kill a tumor without doing too much damage elsewhere). In the case of external beam radiation therapy systems, one takes a computed tomography (CT) scan of the patient, plans an intervention, and plans the radiation pattern. There is a “human-in-the-loop” planning process to determine the machine settings. This is a difficult mathematical problem, he continued, and most of the current research revolves around how to use experiences from previous patients to better set radiation levels. The ultimate goal is to be able to predict how radiation will affect particular patients and optimize for the expected outcomes. Another early example of this technology is the use of orthopedic surgery robots for joint reconstruction, in which the human chooses the implant and determines where it should be placed, and a
robot locates a patient’s bone and acts like a machine tool to prepare the bone to receive an implant perfectly. It is possible to adapt to patient changes and motion, and this experience can be used to help the physician plan the best possible outcome for the patient.
Dr. Taylor indicated that the current dominant paradigm for interactive robotic surgery is remote control. For laparoscopic cryoablation of kidney tumors, for instance, the goal is to manipulate a tissue for which the robot has rules but not a reliable model. This is essentially an attempt to have a robot replace a surgical assistant who would retract a piece of tissue while the surgeon operates. Another example is end-to-end small bowel anastomosis, for which a smart-tool autonomous robot designed for suturing tasks is used. Some dexterity, sensing, and mechanical challenges remain, but the emphasis in the current research is how to make the robot understand the surgeon’s intent and appreciate the scene well enough to provide assistance without exhaustive programming.
The emerging paradigm for interactive robotic surgery is shared autonomy and assistant modes, which Dr. Taylor compared to power steering in a vehicle. In this case, both the surgeon and the robot hold the surgical tool; the robot feels the surgeon pull on the tool and knows how to move to adapt to that force. The robot is doing the motion without tremors, as might exist for a human, which is especially important for microsurgery. Another example of shared control is found in robot-assisted confocal endoscopic imaging for retinal surgery. While a confocal microscope provides exquisite cellular-level views, if careful distance is not maintained, either the patient could end up blind or the surgeon will not be able to visualize anything. This technology is similar to a mechanical autofocus: The human is driving the robot, which has a 5-micron precision. The robot is able to maintain exactly the right distance to keep the image in focus and get high-resolution surveillance of a tumor, for instance. Another example is a “smart” sinus endoscopy assistant that enters the nose and records video. From this video, it is possible to reconstruct a 3D model of the view from the endoscope of the patient’s anatomy, which can be related to the CT scan and used to provide safety barriers, offer guidance to an anatomic target, help find a tumor, or see the carotid artery directly behind the location of drilling. Dr. Taylor commented that there are several military analogies for this technology, which mediates between human intention and action to be able to use information. Research issues for the future of human intention interpretation include having better ways for humans to interface with computers, modeling the task environment, creating better task specification protocols, verifying and monitoring protocols, and developing shared autonomy. Robust and reliable technology (via system design and assurance, and information and system security) enable assured action, which is particularly important for military applications; modeling of task and environment; unusual situation recognition (via experience tracking and learning); safe error recovery; and shared control.
The second case study that Dr. Taylor shared related to the use of robotic systems in infectious disease crises. He explained that robots can help alleviate many of the problems presented by COVID-19 because they are programmable, adaptive, increasingly intelligent, and capable of autonomy and cooperation. However, there are serious challenges to rapid deployment in a crisis situation, in terms of capabilities, systems engineering, and application-specific issues. Many COVID-19 patients are placed in isolation in the intensive care unit (ICU), and critical care personnel have to enter the contaminated area several times each day to make simple adjustments to equipment such as ventilators or infusion pumps. Each visit requires the use of a new set of personal protective equipment (PPE), a process that creates a substantial drain on manpower and available PPE. Dr. Taylor described the solution to this problem as a low-cost teleoperated robot system with a video camera and a custom finger. It has a simple user interface with a wireless connection to a tablet. Sophisticated computer vision allows the robot to attach to the ventilator screen and make small changes by pushing buttons, while a human controls the robot on an iPad located outside of the ICU room. Future possibilities include intelligent ICU assistants, such as a multifunctional mobile robot with video cameras, robot arms, and a variety of tools and sensors. Future technologies could include any combination of teleoperation, shared, and/or autonomous control. Sample tasks include operating ventilators and infusion pumps, feeding and/or delivering meals, biohazard trash and sharps collecting, cleaning and disinfecting, assisting caregivers during procedures, surveillance and patient communication, and patient handling. The basic hardware platform already exists, but issues of cost and reliability remain. He advocated for advances in control and intelligence—for example, in human–machine cooperation, sensing interpretation and modeling of tasks and environment, integration with hospital information infrastructure, and autonomy. There are also deployability issues to consider, including standards, regulatory concerns, and workforce training.
Dr. Taylor remarked that humans have shown considerable ingenuity in using available resources to meet unexpected emergencies. The programmability and adaptability of robotic systems could be exploited to facilitate even more effective responses to emergent needs. Advances in AI and human–machine cooperation are increasing this basic capability. However, he stated that there are some key issues to address through research in terms of technical capabilities, systems, mobilization and infrastructure, and application and domain-specific problems: Technical capability issues relate to autonomy and intelligent systems (i.e., explainable machine learning, sensor/information fusion, situational awareness, and safe actions in uncertain situations), human–machine communication (i.e., speech, natural language, haptics; augmented reality; trust between robot and human on shared tasks; and programmability by non-expert users), and hardware and physical capabilities (i.e., sensors, actuators, and mechanisms). Systems issues relate to system engineering methods (i.e., system and interface design specification; design verification and test; fault detection, fault tolerance, and recovery; risk assessment, projection, and mitigation; pre-certified subsystems; and impact of AI and machine learning), testing and verification methods (i.e., complex systems in unstructured environments, formal and rigorous test methods, non-expert users, normal and unexpected corner cases, and adequacy of training data for machine learning components), and agile and low-cost manufacturing. Mobilization and infrastructure issues relate to an installed base (i.e., enough systems in the field so that repurposing is feasible; economically justified systems; systems designed with interfaces that permit rapid adaptation to new needs; regulatory and reimbursement incentives; and a database of pre-certified designs and application modules) and training (for engineers, end users, and other stakeholders; pre-trained individuals; and better methods for just-in-time training).
In closing, Dr. Taylor reiterated that a new generation of very highly adaptable systems is emerging, and they will be working in partnership with people more often. Machine capabilities and data processing and mechanical sensory motor capabilities could be coupled with human judgment. In order to address all of the key issues, he continued, it is important to maintain strong partnerships among the academic community, the user community, engineering developers, researchers, and users (whether military, space, or civilian).
Gen. Martin asked if AI will help determine more efficient and effective ways to achieve results without sacrificing the technical standards that have been developed for humans. Dr. Taylor pointed out that when designing complex systems for the future, it is possible to design the maintenance process and the maintenance equipment simultaneously and with an understanding of robot capabilities that could transcend human limitations (e.g., robots that can fit where a human hand or wrench could not to make a repair). It will be important to ensure that these robots are controllable by a maintenance technician instead of by a master programmer. If the capabilities of the robot are reliable, they can be characterized, mapped back into the design, and mandated as part of the manufacturing process. Gen. Martin added that human-led processes currently drive military effects, with the aid of tools. With the advent of all-domain operations, a machine might be more effective in providing specificity of results. Dr. Taylor noted that AI has an important role to play in this “decision control” as well as in systems design and theory. However, this scenario more closely resembles a decision-support system that recognizes the tradeoffs of prompt action and performs appropriately in new situations. He agreed that machines operating in partnership with one another and with the humans who are making decisions are desirable. Dr. Chellappa asked how human–AI collaboration is different from human–AI interaction. Dr. Taylor replied that interaction, which is machine- or human-oriented, is part of collaboration, which is more task-oriented. Ms. Westphal concluded the session with an observation about the significant overlap between the issues in the medical world and those of the USAF.
THE PACE OF CHANGE IN ARTIFICIAL INTELLIGENCE: TEMPORAL COMPETITION
Dr. Matt Turek, program manager, DARPA Information Innovation Office, discussed Col. Boyd’s OODA loop, which was inspired by air combat maneuvering and developed in the 1960s (see Figure 5.5). It has been said that time is the dominant parameter of the OODA loop—the pilot who navigates the OODA loop in the shortest time prevails, because his opponent is caught responding to situations that have already changed.7 The “observe” stage of the OODA loop focuses on perception, and the “orient” stage deals with reasoning and building a mental
7 H. Hillaker, 1997, “Tribute to John R. Boyd,” Code One, July, https://www.codeonemagazine.com/f16_article.html?item_id=156.
model. Observing and orienting create the ability to “decide” and, in some cases, “act.” He noted that the OODA loop has additional cycles embedded within it. To predict an adversary’s behavior, one needs the initial position, the range, and the target angular velocity. This problem space is constrained by physics, making this a very different problem than those found in open-ended domains such as information warfare. In a head-to-head dogfight, two OODA loops interact. He explained that AI has an important role in this problem space, enabling quicker perception, reaction, maneuver, and accuracy, as evidenced in DARPA’s Alpha Dogfight Trials. Other elements exist beyond the operational domain of the OODA loop. Focusing on DoD’s technical endeavors, Dr. Turek guided participants to think about the broader OODA loop in terms of the research enterprise and the development and transition of technology. There is temporal competition in this space, particularly around AI. Using the analogy of the dogfight, he indicated that the United States is not only in competition with one adversary in the AI space but rather with several adversaries. This competition is multi-way and highly competitive, with each participant running its own OODA loop as well as additional loops around research, development, and transition.
Dr. Turek mentioned that the attendance at a recent Computer Vision Foundation/Institute of Electrical and Electronics Engineers Computer Vision and Pattern Recognition Conference skyrocketed—the 2006 conference had more than 1,000 attendees, and the 2019 conference had more than 9,000 attendees. The publication of a paper on AlexNet in 2012 triggered the deep learning revolution, and an explosion of deep learning applications has since dominated the computer vision field. While the majority of conference attendees are from the United States, many attendees are from China and other countries. There has also been exponential growth in papers being submitted: In 2019, 56 percent of submissions came from Asia (39 percent of those were from China) and only 27 percent came from North America (25 percent of those were from the United States). Commenting on the “Best Paper, Honorable Mention” from 2019, A Style-based Generator Architecture for Generative Adversarial Networks, Dr. Turek explained that generative adversarial networks (GANs) produce fully synthetic faces. All of the researchers involved with this research paper were based outside of the United States. Within only 2 weeks, the paper went from a research publication to use in the real world: Employing that research is possible by simply refreshing one’s browser for a new synthetic face. While this technology may not seem threatening, it could be used to generate fake social media profiles. He noted that the ability to move from publication to real-world use at that speed within DoD would be remarkable.
Dr. Turek turned to a discussion of the evolution of manipulation technology. Image synthesis and video manipulation have both improved in performance each year owing to the entertainment industry as well as to general advances in machine learning. In principle, GANs can be used for several types of machine learning,
but the standard application has been the generation of faces. Video manipulation is another technology from the computer and machine learning literature that was released as open source—the tool chain was quickly made available, used, and updated. While there are plenty of positive uses for such technologies, the pace of advancement continues rapidly and more negative uses are emerging. As the level of controllability has increased (e.g., from a GAN that will produce a random face to video editing technology that can change mouth movements), he continued, more opportunities now exist for adversaries, thus accelerating temporal competition.
He said that continued advancements in synthetic media generation have begun to move into a multimodal space—for example, fake rental home ads, résumés, and dating profiles. In the past, an online profile with a photograph tended to provide more credibility; however, synthetic photographs are now easy to create. DARPA’s MediFor has developed several techniques that can be used to detect fake images. Dr. Turek described imposing structure on the adversarial space with MediFor in two dimensions: To what resources does an adversary have access (e.g., money, compute, time)? What skills does the adversary have (e.g., technical capabilities and talent pool)? He noted that nation-states have “Hollywood-level” capabilities for media manipulation, but near-peer adversaries may have even more exquisite techniques. Even low-resource groups can retrain a GAN, if they have the right person; with additional resources, they may be able to create their own GAN algorithm. Technology also makes it easier and faster for unskilled individuals to create compelling media manipulations. MediFor has built a range of techniques to cover the adversarial landscape. An important question for MediFor remains, in terms of building adequate defenses: What is the distribution of these sorts of media manipulations in the wild? MediFor’s follow-on program, SemaFor, dedicates a technical area to ensuring access to the latest technology in media manipulation as well as expertise in threats. He has a team that is the generative model for the manipulations that might take place in the wild in order to develop an understanding and build defenses.
Text is another area in which rapid change is occurring, he continued. The T5 text-to-text transfer transformer is a language model with 11 billion parameters. State of the art in natural language models has exploded over the past 18 months, and these types of large language models can take on multiple tasks (e.g., translating, assessing grammar and semantic similarity, and creating a summary). Compute and training data are used to train these models. GPT-3, a model owned by OpenAI, has been licensed relatively exclusively to Microsoft. Dr. Turek explained that the amount of compute necessary to develop state-of-the-art AI algorithms is doubling every 3.4 months. The models are increasing in size, as well as improving in performance, though not necessarily at the rate of growth. One estimate of the cost to train the GPT-3 was $4.6 million, which is relatively inexpensive, for 355 years of compute. OpenAI is providing some research access to GPT-3—for example, a graduate student created a fake blog that successfully fooled a large pool of people.
Dr. Turek stated that solutions can be derived by trading resources for time. For example, compute can be traded (e.g., more compute leads to more parallelization; better compute comes with new hardware designs and new paradigms). Money could be used to buy compute, support research that might change the compute or the algorithms, or hire more skilled individuals—having more skilled individuals is similar to having a better algorithm, and better algorithms can dominate better compute, he continued. More AI assistance as well as increased trust could also potentially allow for faster performance. In closing, Dr. Turek emphasized that time is a critical element of competition that spans the entire enterprise. Time relates to the ability to deploy a new fighter aircraft that can adapt to an adversary more quickly (via short OODA loops across the entire enterprise).
During the question-and-answer session, Ms. Westphal referred to earlier assertions about the many trusted relationships that enable MDC2. However, if the capability exists to manipulate faces and text, and operations are not occurring in-person, she wondered how to know whether decision makers are actually making the decisions. Dr. Turek replied that having control over the communications chain and the perception chain is important. The commercial space is working on ways to secure the imaging chain. It is also important to think about how to authenticate data holdings on an ongoing basis. Algorithms can help, he noted, but there is no easy way to address this complicated problem. Ms. Westphal questioned whether the kill chain should be reassessed. Dr. Turek stressed that several changes may be necessary, such as increasing the speed of the kill chain while confirming that it is as reliable and secure as possible (i.e., certifying the integrity of the data within it, the software, and the systems). He added that several issues remain to ensure trust and assurance around information, particularly in the acquisition process.
Dr. William Powers, retired vice president of research, Ford Motor Company, asked if game theory is embedded in AI. Dr. Turek said that GANs provide the insight to pose the training of a machine learning algorithm as a competition between two algorithms (between the generator and the discriminator)—moving from an optimization framework to an adversarial framework. He has, however, seen some work that looks at collaboration (instead of competition) between components. Two AI agents, or two machine learning agents, could try to learn how to communicate information in a compact way. However, that may not address the broader need for collaboration with a human partner. Several issues related to the collaboration with human partners, even in explainable AI, could be explored further, he stated. A good understanding of the user’s mental model and perspective is important but difficult to achieve and assess. Because trust and collaboration between humans and AI will be fundamental, Dr. Turek described this as an area worthy of further exploration by the AI community—for example, core machine learning and core AI researchers could collaborate closely with people who understand human factors or cognitive psychology. Several years from now there could be a new AI career path that will incorporate those elements, he added. Ms. Westphal asked Dr. Turek if he was referring to AI warriors, and he suggested that the concept be embedded in the USAF and in DoD. If time is a new domain, it affects everything and has to be addressed holistically across all branches of service—not just in terms of building faster but making entire processes move faster to mirror the pace of change outside of DoD (for the adversaries). He stressed that transformational change enables rapid movement.
Dr. Chellappa asked if it is possible to use a blue GAN and a red GAN to determine which AI is better in terms of awareness of the other’s deception and efforts to counter (i.e., information war and the ability to interpret, counter, and seek truth). Dr. Turek said that there is a much work on real-time strategy games. He added that there is a natural back-and-forth in the information operation space: A new GAN is built, a new detector is built, and then something that overcomes the new detector is built. In response to a question from Dr. Powers about AI and improved communication mechanisms, Dr. Turek described the possibilities of richer communication, better situational awareness, and a more shared operating perspective. More importantly, AI helps most with perception tasks; for example, AI can help process a large volume of information and surface cues.
Ms. Westphal inquired as to the difference between time advantage and temporal advantage. Dr. Turek responded that it is difficult to parse those definitions; for instance, there are times when it may be strategic not to move quickly. It is important to be able to make decisions without getting caught behind the curve and having an adversary enter the OODA loop. Just because an OODA loop can be run faster does not mean that actions should be executed faster, he noted. Dr. Turek reiterated his key takeaway: Nation-states are trying to make the best information warfare generators they can at a timescale that is beyond anything that can be matched with DoD’s current acquisition technology process. He hoped that the United States is not caught off guard by the pace of change in the future. Dr. Godfrey wondered if other countries are moving substantially faster than the United States in terms of technology adoption, and Dr. Turek mentioned a recent article about the need to establish a National Technical Intelligence Organization.
Dr. Godfrey referred to Col. DeMaio’s focus on directed energy weapons and noted that U.S. capability in this area is systematically disintegrating. Gen. Martin added that it is important to target areas within the defense strategy for technology acceleration that are not moving forward as fast as they should. AFWIC and others are tasked with the responsibility of finding the most innovative ways to integrate activities and build strategies to be dominant in the future—the United States could be “decapitated” if it does not find ways to catch up and to counter. Ms. Westphal observed a lack of commercial drive for hypersonics and directed energy. Gen. Martin noted that the commercial world does not seem as interested in quantum technologies as DoD. Dr. Powers disagreed, noting that Google and IBM are investing heavily in quantum technologies, although the technology is still further away from development than we realize. Ms. Westphal wondered if the relationship between the government and the commercial market is changing in such a way that could be detrimental in the future. (If the government is not providing funds, will there be a commercial market?) Dr. Hallion noticed a fundamental change in the relationship between military-supported technologies and commercial-supported technologies. For example, the air supremacy
of the United States in World War II came as a result of the dual-use industrial base from 1919 until 1939. The primary driver for technological innovation was the commercial sector, specifically the development of metal streamlined long-range airplanes with advanced radial engine propulsion systems. On the electronic side, there was a much larger investment by the military in systems such as radar, which drove the initial avionics revolution after World War II. Moving into the turbojet era, the commercial world was cut off at the high transonic level, so the military services went beyond that on their own. Although there are some current technologies (e.g., hypersonics) that do not have much support from the commercial sector, AI is coming largely out of the commercial sector, he continued. AI and cyber are the likely modes of future conflict, so there is motivation to avoid corruption of the intelligence, surveillance, and reconnaissance process as well as strategic and tactical communications, and to preserve the fidelity of whatever the kill chain evolves to in the future.
Dr. Godfrey reiterated Dr. Lopata’s concern about generating a pipeline of new researchers in the area of quantum information and added that other areas are suffering from the same issue. University funding is often transferred from physics and electrical engineering departments to biology programs and medical schools, because that is where the research dollars reside. Ms. Westphal observed that COVID-19 could forever impact the education system. Dr. Ryan explained that when the Defense Intelligence Agency eliminated its scientific and technical intelligence organization, it lost a lot of talent. Gen. Martin highlighted the strength of a nation that has clear objectives, and Ms. Westphal added that China is able to force that level of focus.
Ms. Westphal asked participants to think about whether more workshops or a follow-on study should be convened.8 Dr. Godfrey noted that the workshop would have benefitted from the inclusion of international speakers and experts on the business of other countries so as to highlight alternate perspectives as well as learn more about the competition. Dr. Hallion suggested Phillip Saunders of the National Defense University, who runs an organization that studies China, as a future speaker. He also advised the National Academies to review the concept of a national technical intelligence center. Dr. Godfrey pointed out that the USAF has such an organization, though it has little influence; Lt. Gen. Bowlds said that it was called the Foreign Technology Division, and Dr. Hallion added that it became the National Air and Space Intelligence Center. Dr. Yarymovych referenced the North Atlantic Treaty Organization Science and Technology Organization, which has 6,000 scientists and issues reports every week. Dr. Joseph “Jae” Engelbrecht, president and chief executive officer, Engelbrecht Associates, LLC, was impressed by the workshop’s strategic thinking about time and operations but noted that this practice is only happening in small pockets instead of permeating across the forces. Dr. Hallion pointed out that the USAF has an endemic problem of mistreating its strategic thinkers; when they are not given well-deserved promotions, they leave and work elsewhere. Ms. Westphal agreed and noted that this relates to one of the planning committee’s early discussions about the need for the USAF to start recruiting thinkers in order to make the desired changes.