爲科技業試圖防範“終結者”點贊

“A robot may not injure a human being or, through inaction, allow a human being to come to harm.” Isaac Asimov’s precept formed the moral underpinning of his futuristic fiction; but 75 years after he first articulated his three laws of robotics, the first and crucial principle is being overtaken by reality.

“機器人不得傷害人類，或目睹人類個體將遭受危險而袖手不管，”艾薩克?阿西莫夫(Isaac Asimov)的戒律奠定了其未來主義小說的道德基礎；但在他首次明確表述“機器人三定律”的75年後，這條至關重要的第一原則正在被現實壓倒。

True, there are as yet no killer androids rampaging across the battlefield. But there are already defensive systems in place that can be programmed to detect and fire at threats — whether incoming missiles or approaching humans. The Pentagon has tested a swarm of miniature drones — raising the possibility that commanders could in future send clouds of skybots into enemy territory equipped to gather intelligence, block radar or — aided by face recognition technology — carry out assassinations. From China to Israel, Russia to Britain, many governments are keen to put rapid advances in artificial intelligence to military use.

沒錯，迄今爲止還沒有“殺手機器人”馳騁在戰場上。但現在已經出現了可用來探查威脅並向目標——無論是飛來的導彈還是靠近的人類——開火的防禦系統。五角大樓(Pentagon)測試了一批迷你無人機——它們帶來了一種可能性，即未來指揮官可派出一羣羣的skybot（空中機器人）進入敵人領土，收集情報、阻斷雷達、或在人臉識別技術的輔助下完成刺殺任務。從中國到以色列、從俄羅斯到英國，很多政府都急於把人工智能方面取得的快速進展應用於軍事用途。

This is a source of alarm to researchers and tech industry executives. Already under fire for the impact that disruptive technologies will have on society, they have no wish to see their commercial innovations adapted to devastating effect. Hence this week’s call from the founders of robotics and AI companies for the UN to take action to prevent an arms race in lethal autonomous weapons systems. In an open letter, they underline their concern that such technology could permit conflict “at a scale greater than ever”, could help repressive regimes quell dissent, or that weapons could be hacked “to behave in undesirable ways”.

對於研究人員和科技業高管來說，這種情況值得擔憂。他們已經因顛覆性技術將對社會產生的影響而飽受抨擊，他們不希望看到自己的商業創新被改造後用於製造毀滅。因此，百餘家機器人和人工智能企業的創始人日前聯合呼籲聯合國採取行動，阻止各國在致命性自主武器系統方面展開軍備競賽。他們在公開信中強調了他們的擔憂，稱此類技術可能使衝突達到“前所未有的規模”、可能幫助專制政權壓制異見者，這些武器還可能因受到黑客攻擊而做出有害的行爲。

Their concerns are well-founded, but attempts to regulate these weapons are fraught with ethical and practical difficulties. Those who support the increasing use of AI in warfare argue that it has the potential to lessen suffering, not only because fewer front line troops would be needed, but because intelligent weapon systems would be better able to minimise civilian casualties. Targeted strikes against militants would obviate the need for indiscriminate bombing of the kind seen in Falluja or, more recently, Mosul. And there would be many less contentious uses for AI — say, driverless convoys on roads vulnerable to ambush.

他們的顧慮是有根據的，但試圖控制這類武器在倫理和實踐方面都存在困難。那些支持在戰爭中更多使用人工智能的人認爲，此類技術有可能減少傷害，不只因爲所需部署的前線部隊減少，也因爲智能武器系統可以更好地減少平民傷亡。如果可以針對作戰人員展開目標明確的打擊行動，也就不必進行無差別的狂轟濫炸，從而可以避免費盧傑(Falluja)或最近摩蘇爾(Mosul)發生的那種慘劇。人工智能還將開發出很多沒那麼具有爭議的用途——比如說，在易受埋伏路段使用無人駕駛車隊。

At present, there is a broad consensus among governments against deploying fully autonomous weapons — systems that can select and engage targets with no meaningful human control. For the US military, this is a moral red line: there must always be a human operator responsible for a decision to kill. For others in the debate, it is a practical consideration — autonomous systems could behave unpredictably or be vulnerable to hacking.

目前，各國政府在反對部署全自主武器——這類武器可在沒有實際人爲控制的情況下選擇目標並向其進攻——方面存在廣泛共識。對於美國軍方而言，有一條道德紅線：殺人的決定必須由人類操作者做出。對於爭論中的其他各方而言，存在一個現實的考量，即自主系統可能做出難以預測的舉動、或容易受到黑客攻擊。

It becomes far harder to draw boundaries between systems with a human “in the loop” — in full control of a single drone, for example — and those where humans are “on the loop”, supervising and setting parameters for a broadly autonomous system. In the latter case — which might apply to anti-aircraft systems now, or to future drone swarms — it is arguable whether human oversight would amount to effective control in the heat of battle.

如今在“人在決策圈內”的系統（例如完全控制一架無人機）和“人在決策圈之上”的系統（人類監督完全自主的系統併爲之設定參數）之間更難劃分界限了。後一種技術可能適用於如今的防空系統或未來的無人機羣，但一個疑問是，當戰鬥進入白熱化階段，人類監督是否會形成有效的控制。

Existing humanitarian law helps to an extent. The obligations to distinguish between combatants and civilians, avoid indiscriminate attacks and weapons that cause unnecessary suffering still apply; and commanders must take responsibility when they deploy robots just as they do for the actions of servicemen and women.

現有的人道主義法則有一定的作用。人們有責任區分作戰人員和平民、避免無差別攻擊以及會造成不必要傷害的武器；當指揮官像派遣士兵一樣部署機器人去執行任務時，他們必須承擔相應的責任。

But the AI industry is right to call for clearer rules, no matter how hard it may be to frame and enforce them. Killer robots may remain the stuff of science fiction, but self-operating weapons are a fast-approaching reality.

但是人工智能行業呼籲制定更明確規則的做法是正確的，無論這類規則多難制定和執行。“殺手機器人”可能仍然只存在於科幻小說中，但自主操作的武器即將成爲現實。