[译文]遇见新战友:未来战士将与智慧机器并肩作战
 

作者 赖安·肯尼 陆军少校

2018118

耿铁钢 译

来源于:www.militarycommunicators.org.

 

人们对于机器学习能力和人工智能系统众说纷纭。对于这些话题的炒作和猜测推动了许多企业家、投资者及领导人的想像与雄心。目前,军队是时候考虑机器学习能力和人工智能系统的发展对作战环境的影响,并决定如何最好地适应和应用这些进步了。

谁获得数字感知的主导地位,谁将在未来战争中占据优势。自战争起源以来,指挥官们就利用对作战环境、作战对手及自已的了解来指导决策。他们依靠可执行的情报来驱动作战。孙子名言:“知己知彼,百战不殆”,今天依然如此。改变的只是获取和解释这种情报的手段。

20世纪90年代中期,收集和储存大量数字信息的能力,为培养智能创造了新条件。从不断增长的数据储存中创造价值的动力,激发了预测分析的新发展。而这些新发展可以从“大数据”中解读出人类思维无法解读的洞察力。

随着智能手机的兴起,捕捉、分析和理解其应用程序的需求随之增加。全球约有50亿台手机,每台手机约有十几个传感器,而来源于手机传感器的数据是一笔巨大的财富。最近,“物联网”兴起且已扩展到汽车、住宅和办公室领域,来源于此的数据提供了更富足的领域,数据分析师可从中获取新的见解。

在人工智能研究领域,即广为人知的机器学习领域,新方法不断地改进用于理解大数据的算法。

驱动炒作

在机器学习领域,训练数据被输入模型且持续拟合曲线,以提供越来越高的精度。这种再拟合允许给定系统根据特定变量较重要的改变来调整其预测。此方法使预测系统能够产生越来越精确的算法,这就是机器以“学习”提高其预测能力。这种系统递归提高其表现的能力,正是人工智能系统背后推动宣传的原因。

分析人士和研究人员仍在激烈争论,人工智能系统何时能够到达这样一个临界点,即能够独立开发出通过机器学习来提高其性能的手段。然而,像谷歌的“深度思维”这样的人工智能系统,展示出的机器学习能力的速度是令人震惊的。这一系统从需要教授各种游戏规则,然后控制它们自我学习这些规则,并递归地将其性能提升到无与伦比的水平。许多人认为谷歌“深度思维”实验的结果既令人振奋又令人担忧。

最近,《哈佛商业评论》报道称,近一半的财富1000强企业在大数据和预测分析方面的投资取得了重要结果,不仅是在改善其利润方面,还包括它们如何开展业务。

美国电话电报公司负责大数据的高级副总裁维克多·尼尔森认为,应用大数据和分析技术辅助进行战略决策有好处。“我们使用大数据技术来分析所有不同的排列来增强这种体验,以更快地解决或改善特定情况。”20164月,他对麦肯锡管理咨询公司说:“我们找出复杂的事物,将之它变得简单可行。”摄取复杂的数据以提供明确的指导,不只是企业领导人想要的,这也是军方指挥官对其参谋人员所要求的。

人工智能的劳动力

许多分析人员和研究人员所关注的核心问题之一,是伴随人工智能系统而兴起相关的未来工作的性质。人类将会如何?他们会把什么工作交付给人工智能系统?亚历克·罗斯在《未来的产业》、马丁•福特在《机器人的兴起:技术和失业的未来》,以及理查德和丹尼尔·萨斯坎德的《职业的未来:技术将如何改变人类专家的工作》认为,大多数知识工作者将被淘汰。

国防部及陆军应考虑,人工智能系统可以取代军事形态内的何种工作,或至少可由他们加强,来提高作战人员的战斗力。

就企业、军队而言,普遍存在的数字设备、无所不在的网络连接,以及大数据要素的相互融合提供了一种手段,可以获得以前无法想象的认知水平。从公共资源、军事及情报部门资产源源不断涌入的数据,将需要能力日益增强的人工智能系统。

许多有前途的人工智能系统已经应用于组织决策的战略及操作层面,这就是他们能够提供这样优势的新证据。军队获取和保持数字认知优势,将获得关于对手行动及其可能意图的深邃的洞察力。军队也将发现,在关于如何最好赢得战争的议题,自己更加高度依赖于人工智能系统的建议。

当多数军队领导考虑到人工智能系统的应用时,他们想像中的战场散落着“终结者”一般的数字机器尸体和自主行动的无人系统。尽管未来军队许多有人系统也许会被完全替代,但近期这一幕不太可能发生。最有可能的是有人系统及无人系统共存,二者协同完成工作。这种人工智能系统与人类决策者相互结对,被称为人机协作。

人机协作的概念,关注于人类和机器最好能力的协作关系。在战术层次,这意味着增加“技术到任务”的分析,作为“部队到任务”整体分析的一部分。战斗指挥官想要从何处购买低风险的人工智能无人系统,以及他们如何利用人工智能系统来增强战斗能力?

3个主要领域似乎是军队研究的很好的起点。

1.提高理解力

人工智能,由大数据喂养和预测分析驱动,已经证明了他们在商业领域的价值。例如,与人工智能助手配合的医生,能够更好地诊断症状和直接治疗。人工智能系统已经通过了医学考试,超过了诊断肺炎的放射学家,并进入了蓬勃发展的医疗器械市场。在这些进步的基础之上,提高医护兵的战斗力只是军队可以追求的现成能力的一个例子,但军队还有更多能做的。

2.增强决策

许多人工智能系统已经证明能够处理基本的决策。例如,自动驾驶汽车已证明其能够独立自主的通过复杂环境。一旦苹果公司的“Siri”智能语音助手、亚马逊公司的“Alexa”等初始人工智能系统发展成下一代的类人顾问,用于解决合作问题及人工智能系统增强的决策的新算法将会出现。在战斗中,这意味着未来指挥员也许会拥有实时的人工智能参谋人员,用于分析可能的结局及提供建议。但要实现此种可能,陆军现在必须投资,开始收集人工智能系统所需要的数据,以通过机器学习逐渐提高其性能。

3、外包的战术

人工智能系统的进步,导致机器人的发展。机器人可以完成大多数人无法完成或仅仅是不想去做的枯燥的、讨厌的、危险的工作。简单的、遥控机器人早已被爆炸物和处理单位采用,用于探测危险装置。陆军航空兵单位最近寻求编入更多的无人系统。另一方面,无人车正在进入步兵旅战斗队中,同时努力为徒步步兵提供支持。例如,美国陆军研究实验室正在研究步行机器人,它像人类一样灵巧,能够在变化着的环境中操纵物体和实施机动。

人工智能系统控制的无人系统配合,作为先进侦察兵或受保护的百夫长很有诱惑力。然而,围绕这些系统的使用,仍然存在伦理和法律争议。

自主系统,当被一系列严格条件所程控时,能够脱离人为干预而发现、锁定和攻击目标。政策制定者面临的持续挑战,是运用此类系统的交战规则。他们何时会在没有得到人类决策者同意的情况下夺走他人的生命呢?

传感器、通信和控制的进步,会使得运用自主系统的复杂性得到日益增长。因而,陆军必须为自主系统的使用制定清晰的指导方针,并确保其政策能跟上这些技术的发展。

其他的担忧

伴随机器人与人机结对的兴起,除了这些伦理问题外还有其他的一些关注点(担忧)。对于先进通信系统的过度依赖,被认定为是关键的脆弱点。如果未来系统继续依靠通信系统初期基本功能,这将成为更大的风险。网络安全及这些系统的关注点被对手降级或操纵,将使许多人对采用人工智能系统持审慎态度。

同样,作为散布在训练环境和战场上的普遍存在的数字认知,更多的个人数据也将被以数字形式捕获和存储,这将导致对隐私担忧的增长。最后,陆军不能忽略公众这一争论,即未来人工智能系统是否会背叛他们的人类创造者。目前此种恶毒的机器并不存在,但一些著名的思想家担心这些机器将会出现。

政府及军队在此观点上意见一致,即他们不能忽视大量的研究及掌故所指出的,由于大数据、预测分析的拓展以及人工智能系统的兴起所引起的作战环境的巨大变化。陆军领导人必须洞悉这些变化,并在战略、战役及战术层面开发新的方式,以确保军队继续保持在今天所有拥有的决定性优势。

尽管人工智能可能被夸大,但仍然有部分是客观真实的。

 

 

 以下为英文原文:

 

ASSOCIATION OF THE UNITED STATES ARMY

 

MEET YOUR NEW BATTLE BUDDY: SOLDIERS WILL PAIR WITH SMART MACHINES TO FIGHT FUTURE BATTLES

 

 

MAJ. RYAN KENNY

Thursday, January 18, 2018

 

The buzz surrounding machine learning and artificially intelligent systems has become deafening. The hype and speculation about these topics are driving the imaginations and ambitions of many entrepreneurs, investors and leaders. The time has come for the Army to consider the impact these developments have had on its operating environment and determine how best to adapt and apply these advancements.

Whoever gains digital awareness dominance will have the high ground in future warfare. Since the dawn of battle, commanders have used their understanding of their operating environment, their adversary and themselves to guide their decision-making. They have relied upon actionable intelligence to drive operations. Sun Tzu’s well-known quote, “If you know the enemy and know yourself, you need not fear the result of a hundred battles,” remains true today. What has changed, however, are the means to acquire and interpret this intelligence.

In the mid-1990s, the ability to collect and store massive amounts of digital information created new conditions for cultivating understanding. The drive to produce value out of these growing data stockpiles inspired new developments in predictive analytics that could tease out insights no human mind could ever decipher from this “big data.”

With the rise of the smartphone, the demand to capture, analyze and understand their use increased. The data from the dozen or so sensors found in the approximately 5 billion mobile devices tethered to their users globally has been a bonanza. Most recently, data from the rise of the “internet of things” expanding into cars, homes and offices has given data analysts more fertile fields from which to harvest new insights.

New methods of iteratively improving the algorithms used to make sense of big data have garnered advances in the field of artificial intelligence (AI) research—most notably in the concept of machine learning.

Driving the Hype

In machine learning, training data is fed into a model that continuously fits a curve to provide higher and higher levels of accuracy. This refitting allows a given system to adjust its predictions based on changes in the relative importance of given variables. This method allows a predictive system to produce more and more accurate algorithms—hence the machine is “learning” to improve its predictive performance. This ability for a system to recursively improve its performance is what is driving the hype behind AI systems.

Analysts and researchers still heavily debate when the tipping point for AI systems to independently develop the means to improve their performance through machine learning will occur. However, the rate at which AI systems, such as Google’s DeepMind, are demonstrating such performance is astounding. This system went from having to be taught the rules of various games then dominating them to self-learning these rules and recursively improving its performance to unparalleled levels. Many view the results of Google’s DeepMind experiments as both electrifying and alarming.

Recently, Harvard Business Review reported that nearly half the Fortune 1000 companies are achieving measurable results from investments in big data and predictive analytics—not just in improving their bottom line—but how they do business.

Victor Nilson, senior vice president for big data at AT&T, has argued for the benefits in applying big data and analytics to aid strategic decision-making. “We’ve used big data techniques to analyze all the different permutations to augment that experience to more quickly resolve or enhance a particular situation. We take the complexity out and turn it into something simple and actionable,” he told the McKinsey & Co. management consulting firm in April 2016. Ingesting complex data to produce straightforward guidance is not just what corporate leaders want—it’s what military commanders demand of their staff.

The AI Workforce

For many analysts and researchers, one of the central questions they have concerning the rise of AI systems relates to the future nature of work. What will humans do? What will they outsource to these systems? Alec Ross in The Industries of the Future, Martin Ford in The Rise of the Robots: Technology and a Jobless Future, as well as Richard and Daniel Susskind in The Future of the Professions: How Technology Will Transform the Work of Human Experts, argue that most knowledge workers will become obsolete.

DoD and the Army should consider what work within military formations may also be replaced by AI systems—or at least enhanced by them—to improve the effectiveness of warfighters.

For corporations as well as the military, the combination of ubiquitous digital devices, omnipresent connectivity and big-data opportunities offers the means to gain a level of awareness previously unimagined. Data, flooding in continuously from public sources as well as the military and intelligence community’s assets, will require increasingly capable AI systems.

Much of the promise of AI systems employed at the operational and strategic levels of an organization’s decision-making stems from emerging evidence that they can provide such an advantage. The military that gains and maintains digital-awareness dominance will be rewarded with profound insights regarding an adversary’s actions and likely intentions. It may also find itself relying more heavily on AI advice on how best to win.

When most military leaders consider the application of AI systems, they imagine battlefields littered with Terminator-like digital corpses and unmanned systems acting autonomously. While future armies may replace many manned systems completely in the near-term, this is unlikely. What is more likely is a mixture of manned systems and human coordinated unmanned systems working together. This pairing of AI systems with human decision-makers is called human-machine teaming.

The concept of human-machine teaming centers on a synergistic relationship between the best of human and machine capabilities. At the tactical level, this means adding tech-to-task as part of the broader troop-to-task analysis. Where do combat leaders want to buy down risk and how can they leverage AI unmanned systems to enhance combat effectiveness?

Three primary areas appear as good starting points for the Army to investigate:

1. Enhanced Understanding

AI systems, fed big data and powered through predictive analytics, have demonstrated their value in the commercial world. For instance, doctors, paired with an AI assistant, have shown themselves better able to diagnose symptoms and direct treatments. AI systems have passed medical exams, outperformed radiologists in diagnosing pneumonia, and made their way into a booming medical device market. Building on these advances to enhance combat medics’ effectiveness is just one example of a ready-now capability the Army could pursue, but there are many more.

2. Enhanced Decision-Making

Many AI systems are proving they can handle basic decision-making. For example, self-driving cars have demonstrated they can navigate complex environments autonomously. Once the nascent AI systems such as Apple’s Siri and Amazon’s Alexa evolve into next-generation human-like consultants, new paradigms for cooperative problem solving and AI-enhanced decision-making will emerge. In combat, this means future commanders may have real-time AI staff members analyzing likely outcomes and offering advice. But for this to become a reality, the Army must make investments now to begin collecting the data these AI systems will require to improve their performance iteratively through machine learning.

3. Outsourced Tactics

Advancements in AI systems have led to the development of robots endowed to complete the dull, dirty and dangerous work most cannot do, or just do not want to do. Simple, remote-controlled robots have long been used by explosive ordnance and disposal units to investigate dangerous devices. The Army aviation community has recently sought to incorporate more unmanned systems into their formations. Likewise, unmanned vehicles are making their way to infantry brigade combat teams. Efforts are also underway to support dismounted personnel. For example, the U.S. Army Research Laboratory is researching ways robots can manipulate objects and maneuver through changing environments with human-like dexterity.

The pairing of AI controlled unmanned systems as advanced scouts or protective centurions sounds appealing. However, the ethical and legal challenges surrounding the use of these systems remains contentious.

Autonomous systems, when programmed with a set of strict criteria, can find, fix and attack targets without human intervention. The ongoing challenge for policymakers concerns the rules of engagement for the employment of such systems. When can they take a human life without a human decision-maker’s approval?

Advances in sensing, communication and control will enable ever-increasing sophistication in the employment of autonomous systems. Therefore, the Army will have to establish clear guidelines for their use and ensure its policies keep pace with advances in these technologies.

Other Concerns

With the rise of robots and human-machine teaming come other concerns beyond these ethical questions. The overdependence on advanced communications systems has been identified as a critical vulnerability and will become an ever-greater risk should future systems continue to depend on them for basic functionality. Cybersecurity and concerns of these systems either being degraded or manipulated by an adversary will keep many wary of adopting AI systems.

Also, as ubiquitous digital awareness permeates training environments and battlefields, more personal data will be captured and stored digitally, leading to increased concerns about privacy. Finally, the Army cannot ignore ongoing public debate about whether future AI systems will rebel against their human developers. Malevolent machines do not exist, but some leading thinkers are concerned they will appear.

Governments and militaries alike are at a point where they cannot ignore the abundance of research and anecdotes pointing to broad operating environment changes due to the expansion of big data, predictive analytics and the rise of AI systems. Army leaders must understand these changes and develop new approaches at the strategic, operational and tactical levels to ensure the service maintains the decisive advantages it has today.

While AI may be overhyped, there is something there.

future

Robotics

Maj. Ryan Kenny

Maj. Ryan Kenny is the executive officer of the 30th Signal Battalion, Wheeler Army Airfield, Hawaii. A Signal Corps officer, he has primarily served in the 82nd Airborne Division and the special operations community and has deployed multiple times to Afghanistan. He has created an online forum to foster discussions on emerging technologies at www.militarycommunicators.org.

 

 2018-3-14 10:45:18