2 avril 2020 | International, C4ISR

Panel wants to double federal spending on AI

Aaron Mehta

A congressionally mandated panel of technology experts has issued its first set of recommendations for the government, including doubling the amount of money spent on artificial intelligence outside the defense department and elevating a key Pentagon office to report directly to the Secretary of Defense.

Created by the National Defense Authorization Act in 2018, the National Security Commission on Artificial Intelligence is tasked with reviewing “advances in artificial intelligence, related machine learning developments, and associated technologies,” for the express purpose of addressing “the national and economic security needs of the United States, including economic risk, and any other associated issues.”

The commission issued an initial report in November, at the time pledging to slowly roll out its actual policy recommendations over the course of the next year. Today's report represents the first of those conclusions — 43 of them in fact, tied to legislative language that can easily be inserted by Congress during the fiscal year 2021 budget process.

Bob Work, the former deputy secretary of defense who is the vice-chairman of the commission, said the report is tied into a broader effort to move DoD away from a focus on large platforms.

“What you're seeing is a transformation to a digital enterprise, where everyone is intent on making the DoD more like a software company. Because in the future, algorithmic warfare, relying on AI and AI enabled autonomy, is the thing that will provide us with the greatest military competitive advantage,” he said during a Wednesday call with reporters.

Among the key recommendations:

  • The government should “immediately double non-defense AI R&D funding” to $2 billion for FY21, a quick cash infusion which should work to strengthen academic center and national labs working on AI issues. The funding should “increase agency topline levels, not repurpose funds from within existing agency budgets, and be used by agencies to fund new research and initiatives, not to support re-labeled existing efforts.” Work noted that he recommends this R&D to double again in FY22.
  • The commission leaves open the possibility of recommendations for increasing DoD's AI investments as well, but said it wants to study the issue more before making such a request. In FY21, the department requested roughly $800 million in AI developmental funding and another $1.7 billion in AI enabled autonomy, which Work said is the right ratio going forward. “We're really focused on non-defense R&D in this first quarter, because that's where we felt we were falling further behind,” he said. “We expect DoD AI R&D spending also to increase” going forward.
  • The Director of the Joint Artificial Intelligence Center (JAIC) should report directly to the Secretary of Defense, and should continue to be led by a three-star officer or someone with “significant operational experience.” The first head of the JAIC, Lt. Gen. Jack Shanahan, is retiring this summer; currently the JAIC falls under the office of the Chief Information Officer, who in turn reporters to the secretary. Work said the commission views the move as necessary in order to make sure leadership in the department is “driving" investment in AI, given all the competing budgetary requirements.
  • The DoD and the Office of the Director of National Intelligence (ODNI) should establish a steering committee on emerging technology, tri-chaired by the Deputy Secretary of Defense, the Vice Chairman of the Joint Chiefs of Staff, and the Principal Deputy Director of ODNI, in order to “drive action on emerging technologies that otherwise may not be prioritized” across the national security sphere.
  • Government microelectronics programs related to AI should be expanded in order to “develop novel and resilient sources for producing, integrating, assembling, and testing AI-enabling microelectronics.” In addition, the commission calls for articulating a “national for microelectronics and associated infrastructure.”
  • Funding for DARPA's microelectronics program should be increased to $500 million. The commission also recommends the establishment of a $20 million pilot microelectronics program to be run by the Intelligence Advanced Research Projects Activity (IARPA), focused on AI hardware.
  • The establishment of a new office, tentatively called the National Security Point of Contact for AI, and encourage allied government to do the same in order to strengthen coordination at an international level. The first goal for that office would be to develop an assessment of allied AI research and applications, starting with the Five Eyes nations and then expanding to NATO.

One issue identified early by the commission is the question of ethical AI. The commission recommends mandatory training on the limits of artificial intelligence in the AI workforce, which should include discussions around ethical issues. The group also calls for the Secretary of Homeland Security and the director of the Federal Bureau of Investigation to “share their ethical and responsible AI training programs with state, local, tribal, and territorial law enforcement officials,” and track which jurisdictions take advantage of those programs over a five year period.

Missing from the report: any mention of the Pentagon's Directive 3000.09, a 2012 order laying out the rules about how AI can be used on the battlefield. Last year C4ISRNet revealed that there was an ongoing debate among AI leaders, including Work, on whether that directive was still relevant.

While not reflected in the recommendations, Eric Schmidt, the former Google executive who chairs the commission, noted that his team is starting to look at how AI can help with the ongoing COVID-19 coronavirus outbreak, saying "“We're in an extraordinary time... we're all looking forward to working hard to help anyway that we can.”

The full report can be read here.

https://www.c4isrnet.com/artificial-intelligence/2020/04/01/panel-wants-to-double-federal-spending-on-ai/

Sur le même sujet

  • Six questions with France’s Air and Space Force chief

    15 juin 2023 | International, Aérospatial

    Six questions with France’s Air and Space Force chief

    Gen. Mille discussed the French military’s efforts toward an all-Rafale fighter fleet, among other capability goals for the force.

  • ExCobalt Cyber Gang Targets Russian Sectors with New GoRed Backdoor

    23 juin 2024 | International, Sécurité

    ExCobalt Cyber Gang Targets Russian Sectors with New GoRed Backdoor

    ExCobalt targets Russian firms with GoRed backdoor. New cyber threat uses supply chain attacks and advanced tools for espionage across sectors.

  • Army Buys 9,000 Mini-Drones, Rethinks Ground Robots

    18 juin 2019 | International, Aérospatial

    Army Buys 9,000 Mini-Drones, Rethinks Ground Robots

    By SYDNEY J. FREEDBERG JR. WASHINGTON: This summer, Army soldiers will deploy to Afghanistan with air support literally in the palm of their hands: the 1.16-ounce Black Hornetmini-drone. New ground robots are entering service too, next year — not to fight but to haul supplies, at least at first — but field tests have convinced the Army to issue these often-cumbersome mechanical mules to specialists and only loan them to frontline troops as needed. By contrast, soldiers are so consistently and unequivocally enthused about the mini-drones that the Army is buying 9,000 systems — each with two drones — over three years to issue to its smallest and historically most vulnerable units, nine-man infantry squads. The mini-drone and larger robots are all part of a wider revolution in the long-suffering infantry, a revolution sparked in large measure by former Defense Secretary Jim Mattis. The Marines, Mattis's old service, have issued upgraded 5.56 mm rifles and are adding a specialized drone operator to every rifle squad. The Army is going much farther, developing new 6.8 mm rifles, high-tech targeting goggles, virtual-reality training, and, of course, robots. Now, none of these unmanned systems is truly autonomous, so they require a human to run them by remote control, which in turn requires a functioning battlefield network that hasn't been shut down by enemy jamming. The FLIR Black Hornet has a lot of automated functions and only flies short missions, so you don't need a soldier babysitting it all the time. Ground robots, however, require much more oversight, because they have to avoid rocks, bogs, tree stumps, and other obstacles that no unmanned air vehicle has to worry about and that artificial-vision software still struggles to spot. The Army is eager to improve the technology so that, instead of one soldier remote-controlling one robot, they can have one soldier overseeing a largely autonomous swarm. But even today's limited autonomy allows for big changes on the battlefield. The palmtop Black Hornet — dubbed Soldier-Borne Sensor (SBS) by the Army — is already in the hands (literally) of a brigade of the elite 82nd Airborne that's about to deploy to Afghanistan. The second unit scheduled to get the mini-drone, starting this fall, is the 1st Security Force Assistance Brigade, which has already served in Afghanistan. For the first time, a squad leader will have the ability to scout ahead by air before exposing human soldiers on the ground. The SBS has sensor options for both night and day, and it can fly about 20 minutes before needing to recharge. But the squad-level mini-drone is just the entry model. Larger units will get larger, more capable, but also more expensive and more maintenance-hungry drones. “Our vision is every echelon has unmanned aerial systems,” said Don Sando, civilian deputy to the commander of the Army's infantry and armor centerat Fort Benning, Ga. “The question is, how many?” While squads get the Black Hornet SBS, platoons will get the slightly larger Short-Range Reconnaissance (SRR) drone, Sando and other Army officials told reporters in a conference call last week. A series of tests this month, September, and January will whittle six SRR competitors down to one that will enter service in April 2020. The winner must weigh three pounds or less, fly for 30 minutes, and be able to “perch and stare,” landing in a vantage point overlooking a target area so it can keep watch without burning through its flight time. Companies will stick with the current RQ-11 Raven, which is still small enough that soldiers launch it by picking it up and throwing it. Battalions currently use the Raven as well, but the Army plans to develop a new Long-Range Reconnaissance drone for them to use. The LRR isn't an official program yet, however. Brigades currently have the RQ-7 Shadow, but that aging system needs a catapult to launch and a runway to land. It will be replaced by the Future Tactical Unmanned Aerial System (FTUAS), which takes off and lands vertically like a mini-helicopter, starting in 2021. Divisions currently use the Grey Eagle, a variant of the venerable Predator, but the Army is experimenting with potential Advanced UAS drones to replace it too. Ground Bots For Some The Army is also fielding ground robots, but these machines are still much clumsier and harder to work with than aerial drones, so they're being issued only to specialist units. That includes what was formerly called the SquadMultipurpose Equipment Transport but is now renamed the Small Multipurpose Equipment Transport, because it's not going to belong to an individual squad. The SMET, by either name, is still a small unmanned ground vehicle, about the size of a golf cart, designed to trundle along with the foot troops, hauling supplies and gear like an old-fashioned pack mule. It must carry 1,000 pounds of supplies and gear over 60 miles in 72 hours, trundling along with foot troops, and provide three kilowatts of power to recharge batteries for night vision goggles, radios, and other electronics. Four competitors remain in contention: the Polaris MRZR, Howe & Howe Grizzly, HDT Wolf, and General Dynamics MUTT. “They're all viable candidates. They all met the standards of range, offloading power, silent watch capability, payload carrying,” Sando said. “Soldier feedback on all of them was very comparable with regard to what missions they're good for — and what conditions they're not appropriate for.” The Army plans to field the winner starting next year to selected training centers, combat brigades, and support companies — but not infantry squads. “We found out in the operational tech demonstration [that] it can best be supported now at the battalion level,” Sando said, which is the lowest echelon of an infantry unit to include technical specialists such as a heavy weapons company. The Army's objective is to incorporate the new technologies without adding personnel to take care of them. The issue with SMET is not just the maintenance the robots require — though that can be a large burden for a squad of nine — but also their limited mobility. “There are places where we ask our soldiers to go where nothing else can go... jungle terrain, steep embankments, water, and dense urban environments,” Sando said. “There are areas soldiers can walk and crawl and climb that we just couldn't put a vehicle of this size with them.” The SMET remains very useful for long marches with heavy loads, the bane of infantry soldiers increasingly overburdened by body armor, ammunition, and electronics. Being able to recharge gear from the robot instead of carrying several days' worth of batteries for every item of equipment is itself a significant reduction in weight. Future SMET variants, Sando said, might carry long-range sensors, communications relays, or even weapons. But when foot troops have to go places you can only go on foot, they need to be able to leave the robot vehicles behind and let someone else take care of them. By contrast, Sando said, “the Soldier-Borne Sensor is smaller than a pack of cigarettes, [so] I can use it when I need it, I can put it back.” https://breakingdefense.com/2019/06/army-buys-9000-mini-drones-for-squads-rethinks-ground-robots-for-2020/

Toutes les nouvelles