The Rise of Digital Doctors數字醫生的興起

羅賓·費倫

As AI technology becomes more sophisticated， we can expect them to be used more often in the world of human medicine and healthcare. But is it possible to create medical AIs that rapidly outperform doctors in certain tasks？ Find out all the ways AI is helping the healthcare world.

AI is designed to mimic the human brain in decision making and learning， so with the computing power to learn tasks in days or even hours， it is possible to create medical AIs that rapidly outperform doctors in certain tasks.

Most of the AI systems working in medicine employ smart algorithms， with the machine and deep learning techniques， and are supplemented by speech recognition and computer or machine vision to make their decisions.

It will be some time before researchers can develop artificial general intelligence systems capable of abstracting knowledge and developing their own experiences to share with other AIs. But firms like Microsoft， Google， Apple， IBM and Facebook are gearing themselves up1 to deliver the most advanced AI personalized healthcare possible for patients around the world.

Data plays a hugely important role in helping AI systems learn about human medicine. AI systems are trained on large data sets gathered from real-life cases. Providing detailed patient information in volume is a crucial factor for their success.

One of the most important areas for influencing global health is in the field of epidemiology. Predicting disease outbreaks can save millions of lives by having resources ready should the worst happen. Startup AIME2 has successfully combined public health data with machine learning and AI to create a prediction engine capable of anticipating epidemics months in advance with great accuracy.

Another field where medical AIs are making rapid advances is in diagnostics. Doctors base a lot of decisions on information from X-ray， CT and MRI images. Speeding up diagnoses from patient scans can rapidly improve patient care and outcomes.

Computer vision AIs use pattern recognition to work through these images with incredible speed and accuracy. They have been able to outperform junior doctors and even senior specialists in some tests.

Cardiologist Rima Arnaout developed an AI that beat human experts at correctly interpreting echocardiograms by 92 percent to 79 percent. She said that despite the result there is no prospect of AI replacing human doctors any time soon. “As cardiologists， we read the images and then go see the patient，” she said. “So were both reading images and practicing medicine. I dont think that the second piece will be taken over so quickly.”

The results are obviously impressive， but being aware of the hype around AI in medicine is just as important for both physicians and patients. The Institute of Electrical and Electronics Engineers （IEEE） has a handy visualization tool to show where smart algorithms and humans were better at detecting health problems.

People are sharing more and more of their health data through apps on mobile and wearable devices. Now virtual and voice assistants using natural language processing and AI are being prepped to provide healthcare on-demand. Amazons Alexa3 has partnered with the UKs National Health Service （NHS） to provide users with health advice， but it could also be used to tell if you are having a heart attack.

Governments in many countries face the prospect of ageing populations. This will likely see the expansion of AI services， including robotic helpers. Robots designed specifically to interact with people could help solve the problems of isolation and loneliness that affect many older people.

Georgia Institute of Technology built an experimental robot called PR2 that taught itself how to put a gown onto humans in just one day. Those skills could be readily adapted for people in hospitals and care homes around the world.

We already have mobile robotic telepresence （MRT） systems available to provide support to the ill and elderly. This class of social robots are effectively remote-controlled video screens on wheels that allow medics， carers， or relatives to interact with people in their own homes.

Pet androids like Aibo the robot dog， or Paro （a baby seal） provide companionship and learn from their interactions about each owners preferences.

More humanoid robots include the ‘emotional robot BUDDY， which is able to provide social interaction， be a personal assistant， play multimedia and games， and look after the elderly， according to its makers. Mabu is another wide-eyed humanoid robot that uses AI and a recipe of best practices from human doctors to help monitor heart failure patients.

Robot AIs can also be put to work in hospitals to help doctors and nurses spend more time with their patients. Moxi is a robot assistant that helps staff by completing general tasks such as delivering lab samples， collecting laundry or gathering medical supplies.

Even in the surgical suite， there is support from AI robotic surgery systems that reduce variations4 between surgeons which affect patient recovery. Dr. John Birkmeyer， a chief clinical officer of Sound Physicians5 said， “we know that a surgeons skill， particularly with new or difficult procedures， varies widely， with huge implications for patient outcomes and cost. AI can both reduce that variation， and help all surgeons improve—even the best ones.”

The acceptance of AI in medicine will continue to gather pace in the future as it becomes more widespread. Its promise to enhance patient care by reducing errors in diagnosis， improving the ability to predict disease， and providing assistance to busy clinicians is also the promise of keeping humans at the centre of healthcare.

隨著人工智能技術日趨成熟，其在人類醫學和醫療保健領域的應用會越來越廣泛。但有可能開發出在某些工作上迅速超越人類醫生的醫學人工智能技術嗎？我們來看看人工智能在醫療保健領域都有哪些應用。

人工智能旨在模擬人腦進行決策和學習，因為計算機有超強學習能力，幾天甚至幾個小時就能完成某項學習任務，因此開發出在某些工作上迅速超越人類醫生的醫學人工智能是有可能的。

大多數醫學領域的人工智能系統采用智能算法，利用機器學習和深度學習技術，輔以語音識別和計算機或機器視覺來做決策。

研制出能夠提煉所學知識和積累經驗與其他人工智能進行分享的通用人工智能系統尚需時日，但微軟、谷歌、蘋果、國際商用機器公司（IBM）和臉書等公司正在準備為全世界的患者提供最先進的人工智能個性化醫療服務。

在幫助人工智能系統掌握人類醫學知識的過程中，數據扮演著極其重要的角色。人工智能系統通過來源于真實病例的大型數據集學習。提供大量而詳實的病患數據是其成功的關鍵因素。

流行病學是影響全球健康最重要的醫學領域之一。預測疾病的暴發，從而在最壞的情況發生時有所準備，可以拯救千百萬人的生命。初創企業AIME成功地將公共衛生數據與機器學習和人工智能結合起來，開發出的預測工具能夠提前數月對流行病進行精準預測。

醫學人工智能發展迅速的另一個領域是診斷學。醫生做的很多決定都基于X光、CT和核磁共振成像的檢驗結果，加快醫學影像的診斷速度可以迅速提高病患護理質量和改善治療效果。

計算機視覺人工智能利用模式識別技術處理影像，速度之快、準確度之高都達到驚人的程度。在一些測試中，其表現已超過了初級醫生甚至資深?？漆t生。

心臟病學家里馬·阿爾瑙特開發了一種人工智能技術，解讀超聲心動圖的準確率為92%，比人類專家（79%）更勝一籌。但她認為，即便如此，短期內人工智能也無法取代人類醫生。“作為心臟病專家，我們看過影像后還要進行面診?！彼f，“我們既解讀影像又問診和觸診。我認為這第二部分的工作不會這么快就被人工智能取代?！?/p>

科研成果令人矚目，但意識到人工智能在醫學領域被過度宣傳對醫生和患者同樣重要。電氣與電子工程師協會有便于操作的可視化工具，可演示智能算法與人類在探查疾病方面各有哪些優勢。

人們越來越多地通過手機應用和可穿戴設備分享自己的健康數據。目前正在研制可運用自然語言處理和人工智能技術的虛擬語音助手，根據需要提供醫療保健服務。亞馬遜公司的Alexa人工智能助手與英國國家醫療服務體系合作，為用戶提供健康咨詢，也可以用來判斷你是否心臟病發作。

很多國家的政府即將面對人口老齡化問題，這可能促使包括機器人助手在內的人工智能服務的發展。許多老年人易產生孤獨感，專門設計用來與人互動的機器人有助于克服這些問題。

佐治亞理工學院制造了一款叫做PR2的實驗機器人，一天內就自學掌握了如何給人穿外罩，此技能隨時可以推廣到全世界的醫院和看護中心。

現在已經有了移動機器人遠程呈現系統，能夠為病人和老年人提供幫助。此類社交機器人實際上是帶輪子的遠程控制視頻屏幕，醫生、護工或親屬可通過它們在自己家里與需要照顧的人互動。

機器狗Aibo或小海豹Paro這樣的寵物機器人可以提供陪伴，并且能在互動中了解自己主人的喜好。

還有更多的類人機器人，如“有情感的”機器人Buddy，據制造者介紹，它懂得社交互動，可以做個人助理，會使用多媒體和玩游戲，還能照顧老人。Mabu是另外一款大眼類人機器人，它利用人工智能和人類醫生提供的最佳治療方案來幫助監護心臟衰竭患者。

人工智能機器人還可以在醫院幫忙，讓醫生和護士有更多的時間照料病人。Moxi是一款機器人助手，它可以幫工作人員完成諸如運送實驗室樣品、取走臟衣服或收集醫療耗材等日常工作。

即使在外科手術室里也能用到人工智能自動手術系統，這類系統縮小了不同外科醫生的技術差異，這些差異會影響患者康復效果。Sound Physicians的首席臨床官約翰·伯克邁耶醫生說：“我們知道，外科醫生間的技術差異是很大的，尤其是在做新的或復雜的手術時，對醫療效果和治療費用都有很大影響。而人工智能既可以縮小這類差異，又可以幫助所有外科醫生——甚至是最好的外科醫生——提升技術?！?/p>

將來，隨著人工智能越來越普及，其在醫療領域的應用也會繼續加快步伐。有望通過減少誤診、提高疾病預測能力和為忙碌的臨床醫生提供幫助來提高病患護理水平，實際上也是在醫療保健工作中做到以人為本。

（譯者為“《英語世界》杯”翻譯大賽獲獎選手）