内容丰富 (Informative)

Let’s take a stroll down memory lane and take a look at the times where AI came out on top at a human-based challenge when faced with a human adversary.

L等人漫步在记忆里，看看AI在面对人类对手时在基于人类的挑战中脱颖而出的时代。

In 1997 the world watched the defeat of world chess champion Garry Kasparov. AI dominance in board games was repeated in 2015 and 2016 where AlphaGO, a DeepMind’s AI system, defeated professional GO player Fan Hui and world champion GO player Lee Sedol.

1997年，全世界目睹了国际象棋冠军加里·卡斯帕罗夫(Garry Kasparov)的惨败。 AI在棋盘游戏中的主导地位在2015年和2016年再次出现，DeepMind的AI系统AlphaGO 击败了职业GO玩家Fan Hui和世界冠军GO玩家Lee Sedol 。

AI dominance over human is not constricted to board games, in 2011 IBM Watson, a computer system capable of answering trivial questions beat Brad Rutter and Ken Jennings in a Jeopardy game.

AI在人类上的主导地位并不局限于棋盘游戏，在2011年的IBM Watson中，能够回答琐碎问题的计算机系统在Jeopardy游戏中击败了Brad Rutter和Ken Jennings 。

We can add one more date to this growing list of AI victories over humans.

我们可以在这个不断增长的人类AI胜利列表中再加上一个日期。

2020年8月20日，AI在一次模拟混战中击败了人类F-16喷气式战斗机 (20th August 2020, AI defeats human F-16 Jet fighter in a simulated dogfight)

The DARPA AlphaDogFight is a series of trials that tests the capabilities of AI algorithms executing dogfight manoeuvres, combat and strategies.

DARPA AlphaDogFight 是一系列试验，测试AI算法执行格斗动作，战斗和策略的能力。

The final trial was held on 18–20th August 2020 between eight teams and the winning team, Heron Systems, AI faced off with a human fighter jet pilot.

最终审判于2020年8月18日至20日在八支团队和获胜团队Heron Systems，AI之间进行，对抗者是一名人类战斗机飞行员。

The purpose of DARPA’s AlphaDogFight is to find methodologies of developing AI systems that can shift the responsibilities of aerial combat and manoeuvring from the human pilot to the AI system. Inevitably building trust in AI within flight systems.

DARPA的AlphaDogFight的目的是找到开发AI系统的方法，这些方法可以将空战和机动的职责从人类飞行员转移到AI系统。 在飞行系统中不可避免地建立对AI的信任。

The human pilot went by the name ‘Banger’ and Herons System’s AI was identified by the name ‘Heron’. The faceoff between Heron and Banger took places in a simulated environment, where both pilots had control of an F-16 fighter jet.

飞行员被冠以“邦格”的名字，苍鹭系统的AI被冠以“赫伦”的名字。 Heron和Banger之间的对峙发生在模拟环境中，两名飞行员都控制了一架F-16战斗机。

This final challenge aimed to test the Heron’s manoeuvring capabilities, combat system and decision making against a human adversary. The challenge lasted five-rounds, and Heron, the AI pilot, defeated Banger in an incontestable manner.

最后的挑战旨在测试苍鹭的机动能力，战斗系统和对抗人类对手的决策。 挑战持续了五轮，AI飞行员Heron以无可争议的方式击败了Banger。

The commentators of the event predicted that the faceoff between Banger and Heron was going to be “a very close fight”.

该事件的评论员预测，班格和苍鹭之间的对峙将是“一场非常近距离的战斗”。

After all the rounds, it was a flawless victory to Heron for all five rounds of the challenge.

在所有回合之后，苍鹭在五轮挑战中都取得了无懈可击的胜利。

According to the commentators, Heron displayed superhuman capabilities of being able to shoot and aim very accurately while performing highly dynamic manoeuvers.

据评论员说，苍鹭表现出超人的能力，能够在进行高度动态的机动时非常精确地射击和瞄准。

“Displayed superhuman capabilities”

“显示的超人能力”

In some rounds, Banger changed tactics and strategy to get the upper hand against Heron. During the last round, the change in tactic by Banger led to a round that highlighted some limitation in Heron’s capabilities. The round still ended with Heron coming out victorious.

在某些回合中，Banger改变了战术和策略，以击败Heron。 在上一回合中，Banger改变了战术，这一回合突显了苍鹭的能力上的某些局限性。 在这一回合中，苍鹭取得了胜利。

In all five rounds, Heron didn’t lose any health, meaning that no single shot from Banger hit Heron’s fighter jet.

在所有的五轮比赛中，苍鹭都没有失去任何健康，这意味着没有来自Banger的单发击中苍鹭的战斗机。

把它带回地球 (Bringing It Back Down To Earth)

Despite the overwhelming victory by Heron, it was noted that the AI system had ‘perfect state information’, this means that the AI systems had uninterrupted quantitative and visual data of the simulated environment.

尽管Heron取得了压倒性的胜利，但人们注意到AI系统具有“完美的状态信息”，这意味着AI系统具有不间断的模拟环境定量和可视数据。

In a real-world scenario having complete information on the environment and other variables is not common.

在现实世界中，具有完整的环境信息和其他变量的信息并不常见。

The unpredictable nature of reality has an impact on how serious military forces take information and seemingly ‘good’ performance from simulated environments.

现实的不可预测性影响着认真的军事力量如何从模拟环境中获取信息和看似“良好”的表现。

Simply kept we are not going to be seeing any AI systems flying an F-16 fighter jet outside of a simulation anytime soon.

简单地说，我们不会很快看到任何AI系统在模拟之外飞行F-16战斗机的情况。

The advancements and lessons learnt from the DARPA AlphaDogFight trials will set a precedent for future developments that can lead to more reliable AI systems that can work hand in hand with combat pilots outside of simulated environments.

从DARPA AlphaDogFight试验获得的进步和经验教训将为未来的发展树立先例，从而可以开发出更可靠的AI系统，从而可以与模拟环境之外的战斗飞行员携手共进。

获奖队伍 (Winning Team)

The final trial held on the 18th — 20th August 2020 involved eight teams. Across three days, the teams’ AI systems each faced other AI algorithms and then after faced each other in a round-robin event.

2020年8月18日至20日举行的最终审判涉及八支队伍。 在三天的时间里，这些团队的AI系统分别面对其他AI算法，然后在循环赛中彼此面对。

The final day had the top four teams face off in an elimination match, in which the top winning team faced the human adversary.

最后一天的比赛中，前四名球队在淘汰赛中对决，其中获胜的球队面对了对手。

The teams involved in the final AlphaDogTrial event:

参与最终AlphaDogTrial活动的团队：

• Aurora Flight Sciences

  极光飞行科学

• EpiSys Science

  EpiSys科学

• Georgia Tech Research Institute

  乔治亚理工学院

• Lockheed Martin

  洛克希德·马丁

• Perspecta Labs

  Perspecta实验室

• PhysicsAI

  物理人工智能

• SoarTech

  腾飞科技

• and the winning team Heron Systems

  和获胜的团队Heron Systems

Heron Systems AI agent is a deep reinforcement learning-based AI that has gained expertise of over 30 years of flight time experience through constant training within a simulated environment.

Heron Systems AI代理是基于深度强化学习的AI，通过在模拟环境中进行不断的培训，获得了超过30年的飞行时间经验的专业知识。

The Heron team attributed their success to two main factors: scale and diversity.

苍鹭团队将其成功归因于两个主要因素：规模和多样性。

The scale component of their attributed success refers to the amount of flight time in years the winning AI agent has been able to accumulate. The diversity component speaks to the number of AI agents that Heron Systems developed to face off against each other; each agent with their own unique neural network architecture, reward systems and build.

他们的成功归功于规模，是指获胜的AI代理能够累积的飞行时间(以年为单位)。 多样性的组成部分说明了Heron Systems开发为相互对抗的AI代理的数量。 每个代理都有自己独特的神经网络架构，奖励系统和构建。

According to a spokesperson from Heron System, the next steps for the team is to take the AI systems and incorporate them within drones that can be manned in real-life environments.

Heron System的一位发言人表示，该团队的下一步是采用AI系统并将其整合到可以在现实环境中操纵的无人机中。

希望您觉得这篇文章有用。 (I hope you found the article useful.)

To connect with me or find more content similar to this article, do the following:

要与我联系或查找更多类似于本文的内容，请执行以下操作：

1. Subscribe to my Email List for weekly newsletters

   订阅我的电子邮件列表以获取每周新闻

2. Follow me on Medium

   跟我来中

3. Connect and reach me on LinkedIn

   在LinkedIn上联系并联系我