医生八成工作将由科技代劳

系统将从蹒跚学步的婴儿逐渐成长、成熟，达到高效

    不要指望电脑一夜之间就会成为顶尖的诊断系统。它们一开始可能只是小步创新，或是看起来还很笨拙，没有到大显身手的时候。

    想象一下，使用AliveCor*推出的iPhone程序，每天只需不到1美元就可以做一次心电图。与许多病人一年只在医院做两次心电图相比，这类设备获取的信息量明显要多得多，而且费用也便宜得多。可能你做500次“自动诊断式”心电图，都比在医院里做一次还要便宜。今天，大多数心脏病都是在患者突然发病后才得到确诊。但是有了能够鉴定异常情况和能够预测发病的自动学习软件，人们就能获得预防性的心脏治疗。我们可以在突然发病或中风前，就发现大多数的潜在心脏病。而且，与发作后的治疗相比，付出的医疗成本也少得微不足道。不过我们还需要几十年的数据积累，才能实现这个目标。

    皮肤病的诊断可以交给CellScope公司生产的低成本的iPhone配件来完成，它可以扫描皮肤上的痣、皮疹，以及耳部感染，将来可能还可以扫描喉咙和视网膜。通过计算机算法，可以对图像进行处理，从而做出最接近的诊断。Eyenetra公司生产的一款设备，可以对人们的眼睛进行验光，帮助人们配眼镜，省去了许多费用和奔波的麻烦。Adamant公司目前正在研制一款芯片，能检测人们呼吸中的几百种气体，因此可用来检测甚至确诊几种不同的肺癌。它们都比现在的大型CT强得多，因为后者折腾一番之后，只会告诉你，你长了一个瘤。Ginger.io的产品可以监测人们发电子邮件、微博、短信和打电话的频率，通过人们的社交活动，研究人们的行为变化，它可以比一个精神病专家更好地反映人们的精神状态。

    这些创新在一开始可能显得无关紧要，但日积月累，它们会日益壮大，形成一场革命。届时，拿今天的科技与2020年的科技相比，就好像拿1986年砖头一样的大哥大与今天的iPhone相比一样！

人的因素仍将存在

    有些质疑自动化医疗的人士指出，医学不仅仅是输入症状、输出诊断那么简单，医学是建立在医生和病人之间互动关系上的学问。人类能比机器提供更好的病床护理，更好地回答病人的问题。这当然是实情，不过一般来说，我们不一定非得拿到医学学位才能做到这一点。许多护士、护理师、社工以及其他收费更低的非专业医生也能做到这一点，甚至有不少人比医生做得更好，而且他们会花更多的时间来提供个性化、有爱心的护理。笔者并不是在这里鼓吹让医生下岗，而是说，我们应该通过先进的机器学习和人工智能技术，打造强有力的后台传感器技术和诊断技术，让它们来处理人力所不能及的庞大信息量。

Systems will start as clumsy toddlers and develop to maturity and efficiency

    Don't expect ace diagnosis systems overnight. They may start as seemingly minor point innovations or as clumsy-sounding systems not ready for prime time.

    Imagine using the AliveCor* iPhone case to take an ECG every day for less than $1/test. This device and others like it would capture a lot more information than the typical heart patient's semiannual ECG check at the doctor's office (it would also cost a lot less). What if you could send 500 "auto-diagnosed" ECGs to your doctor for less than it costs to get one ECG done in the hospital? Today, most heart disease is identified only after patients have heart attacks. But imagine having preventative cardiac care, enabled by machine-learning software that identifies abnormalities and predicts episodes. We could discover most heart disease before a heart attack or stroke and address it at a fraction of the cost of care that would be needed following such a trauma. But we need a decades-worth of data to be really good at it.

    Dermatology appointments could be handled by CellScope*, which produces low-cost iPhone attachments for imaging skin moles, rashes, ear infections, and (in the future) your retina or throat. Those images could be processed by algorithms to detect patterns that warrant closer inspection. A device like the Eyenetra* could give you an eye test and fit you for glasses at little cost or hassle. Adamant* is attempting to produce a chip that can identify hundreds of gases in your breath, which could be used to detect and even identify different types of lung cancer, all for far less than a big CT scanner that'll just tell you that you have a nodule. Ginger.io* monitors your rate of emailing, tweeting, texting, and calling to gauge your social activity. By watching for changes in your behavior, it can tell how you're doing far better than a psychiatrist.

    These point innovations will seem immaterial at first, but, when there are enough of them, they will integrate and start to feel like a revolution. The technologies of 2020 will be as different from today's systems as the car floor-mounted, multi-pound cell phones with bulky handset cords of 1986 are from today's iPhones!

The human element will survive

    Some critics of more automated healthcare argue that medicine isn't just about inputting symptoms and receiving a diagnosis; it's about building relationships between providers and patients. Providing good bedside manner and answering certain questions can often be handled better by a person than a machine, but you generally don't need a medical degree to do that. Nurses, nurse practitioners, social workers, and other less expensive, non-MD caregivers could do this just as well as doctors (if not better) and spend more time providing personal, compassionate care. I'm not advocating the removal of the human front-end here. I'm arguing that we should build robust back-end sensor technology and diagnostics through sophisticated machine learning and artificial intelligence operating on data in greater volumes than humans can handle.