十年之后，在就业市场中人工智能会扮演什么样的角色？编者按：我们最先从科幻电影中认识到何为人工智能，慢慢的它们开始出现在我们的生活之中。那你有没有想过，十年之后人工智能会在就业市场中扮演什么样的角色，如果它关系到你的工作，你还会欢迎它吗？近日Sentient Potential and TechEmergence 的创始人 Daniel Faggella就这个问题谈了他的看法。
从 Google 收购 DeepMind、Boston Dynamics 等等开始，到逐步增加对人工智能的资本投资关注，再到最近 Elon Musk 和 Bill Gates 都表示了对未来超级人工智能安全性的担忧，这些都表明人工智能已然从幕后走向了台前，成为大众关注的焦点。
虽然对于未来我们无法预测，但是许多资深的计算机科学研究者认为未来 5 到 10年 内的的人工智能的影响趋势还是可以把握的。
那么自动化到底将怎样影响人类工作的本质和需求？因此我就人工智能未来十年如何影响就业市场这个问题请教了业内 6 位资深的人工智能博士，询问他们对于未来十年里人工智能将如何影响就业市场的观点。虽然他们对不同行业的影响持有不同的观点，但有一点认识是共同的：扩大或加强使用现有的算法已成必然趋势。
涉及面细分、重复性高或者数据评估类的工作最容易被自动化取代。佐治亚理工大学致力于研究计算机视觉的教授 Irfan Essa 表示，在过去的十年里，计算机视觉得到了显著的发展，很多可以引入人工智能技术的领域长期以来一直处于 “聚集模式”，现在我们终于迎来了转折点。
视觉数据并不是唯一一个可能被人工智能取代的领域。《Rise of Robots》的作者 Martin Ford 认为：在未来的十年里，比起蓝领，将有更多的白领会被人工智能取代。
Daniel Berleant 也同意此说法。他表示当下流动性无疑是一个巨大的技术难题，但是计算机在处理数据上比人类更出色，而人类则更能胜任体力劳动工作，至少从目前上来看是这样的。尽管过去十年里双踏板行走的机器人发展态势良好，但是像搬家具或者是餐馆里端盘子这些需要灵活性的工作暂时还没有可能迅速地被自动化取代。
有些研究人员认为，数据处理领域的自动化代替也可能发生在细分数据评估领域。Andras Kornai 说 IBM 正在医疗领域引入 Watson，我更希望在法律界也能有类似的技术。虽然机器学习在医学诊断上可以给予医生一定的帮助，但是机器学习并不能完全取代医生。
长话短说，如果你的大部分工作内容涉及到电子表格，那么未来很有可能会出现一款软件取代你，它的效率比人工更高，所需成本更低。所以，如果你不想在 2025年 的时候丢了工作的话，你要好好考虑这个发展趋势了，从而改善你目前的工作状态。
拥有斯坦福大学博士学位的 Eyal Amir 主攻人工智能研究，他表示我们现在看到的趋势无非就是把不同的数据碎片汇聚到了一起，而且我们给予了计算机更多的自控权。我们开始相信计算机有处理基本任务的能力，也有我们人类所不具备的能力。
在最近的一次人工智能采访中，Amir 表示他认为这种不断增加的信任是人工智能程序影响力不断上升的副产品，比如 Apple 的 Siri 和 Facebook 的广告算法。未来的贵宾级服务大概没有一个能比得上 “超级 Siri”，它可以随时随地给您带来需要的讯息，随时随地为您处理任务（预定披萨，预约衣物干洗时间等等）。
Kornai 还明确指出在某些特定医疗诊断或者甚至是法律诉讼中也可以使用这些计算机算法，他相信人工智能在这些领域的缓慢稳定影响是不可避免的，甚至还会取代人类专家在诸如 X 光评估或者某些法律研究工作上的地位。
未来的语音识别算法则可能会创造出另一种经济变化。在伊利诺伊大学执教的 Daniel Roth 表示：他可以预见十年后，我们能够用真正自然的方式与计算机交流，我们可以向机器咨询全球问题，物理学家可以让计算机检索相关研究文献。
虽然他们的观点都不一致，但是当谈到自动化与劳动力市场的未来趋势时，所有我采访过的人工智能研究员都提到了自动驾驶汽车。Amir 认为，与其说机器控制了车轮不如说是人们自己放弃了对汽车的控制权，在未来的 10 到 15年 里，我们将看到满大街都是无人驾驶的汽车。
在其他相关产业中，机动车驾驶员就业市场受到的直接影响将最为严重。Kornai 表示，在美国至少 100 万名出租车司机，换句话说就是到时候将有 100 万人失业。除了卡车司机或者出租车司机的直接失业，也可能会导致人们对汽车购买的需求会减少。
正如当前 Uber 面临的被抵制的困境，未来自动驾驶汽车公司也将面临一场恶战。Kornai 和其他几位都表示传统车辆会逐渐地向自动驾驶车辆转变，同时也有望缓解转变带来的剧烈经济转型。
我们甚至还可以预想到一系列法律问题，随着从人到机器的逐步 “信任转变” 而得到解决，虽然直接从 100%的人类驾驶员到完全的自动驾驶的转变不是不可能，但可能性终究太小。然而不管是何种方式，这些站在科技前沿的人工智能研究人士一致认为，下一个十年无人驾驶汽车会走进我们的生活。
The Next 10 Years Of Automation And What It Might Mean For The Job Market
After decades of subtle developments that largely went unnoticed by much of the working world, artificial intelligence (AI) has taken center stage in the last 2-3 years as a “hot” technology.
From Google’s surge of acquisitions (DeepMind, Boston Dynamics, etc.), to increased venture capital attention, to the safety concerns of Elon Musk and Bill Gates about potentially super-intelligent AI, the field is undeniably back in the spotlight.
One of the most pressing concerns for those of us in the working world is the effect of automation on job security — in both blue-collar and white-collar work.
Though more far-out considerations are difficult to predict, many experienced computer science researchers feel reasonably comfortable speaking about AI’s influence in the coming 5-10 years.
With so much potentially unfounded speculation about how automation might influence the nature and demand for human work, I decided to ask six artificial intelligence PhDs about their informed perspectives on how AI might impact the job market in the coming decade. Their answers didn’t share much commonality in terms of industry, but they did share a common thread: The expanded or strengthened use of existing algorithms.
One wide swath of jobs that may be most easily automated are likely to be jobs that involve narrow and repetitive manipulation or assessment of data. Irfan Essa at Georgia Tech focuses his research on machine vision, a domain that has developed markedly in the last 10 years. “Many fields were AI could be applied have been in ‘aggregation mode’ for quite some time, and now we’re finally getting to a point of sense-making,” says Essa.
While identifying human faces, or categorizing web images (identifying animals, landmarks, objects) was once the arduous job of human beings, many of these tasks can now be automated by trained neural networks (Google’s Peter Norvig explains this process rather well).
Visual data is far from being the only area of narrowly focused intelligence that might be under siege. Martin Ford (author of the well-received book Rise of the Robots) mentions that in the coming 10 years, we’re likely to see more automated job displacement in white-collar jobs rather than blue-collar.
There is ongoing debate as to whether or not technological advancements inherently create more job market opportunities than they destroy.
Daniel Berleant agrees, stating the current difficulties of “mobility is undeniably a rather difficult technical problem, and computers are more likely to manipulate data better than humans than they are to take over most manual labor jobs, at least for the time being.” Despite the impressive developments in bipedal robots in the last 10 years, people with dexterous physical jobs such as moving furniture or carrying plates in a busy restaurant aren’t likely to be automated out of a job anytime soon (though stationary assembly jobs are under siege now as much as ever, with devices like Rethink Robotics’ Baxter).
Some researchers believe that the same might be said of narrow data assessment, not just data manipulation. Andras Kornai states, “IBM is moving Watson into the medical field — I expect the same thing to happen in the legal area.” Though it may be possible that machine learning will aid in the detection of cancer or other maladies in medical imaging, these technologies don’t seem likely to put doctors out of a job.
Long story short, if a large portion of your time at work involves tinkering with spreadsheets, there is likely to be software that will perform your job faster and cheaper than human labor. Marc Andreessen put this in intelligible terms in his “software eating the world” WSJ interview, and it’s worth understanding if you plan on being employed in 2025.
However, the influence of AI in the coming decade may imply an expansion beyond the “narrow” focuses that it’s best known for (i.e., analyzing images, beating silly humans at chess, etc.), and some of the AI experts I’ve interviewed seem to think that people are becoming comfortable handing over that control.
Eyal Amir is a Stanford PhD and Associate Professor at The University of Illinois at Urbana-Champaign focused on AI research. “More generally what you see as a trend is for different pieces of data coming together, and that we give the computers a little bit more autonomy,” says Amir. “We start trusting the ability of the computer to do basic tasks and to have knowledge that we don’t have.”
In a recent AI-focused interview, Amir states that he sees this increased degree of trust as a byproduct of the increased effectiveness of AI programs, such as Apple’s Siri and Facebook’s advertising algorithms (which infer data about individuals’ preferences, vocation, gender and more — based on cues and clues from Facebook’s myriad data points). The concierge services of the future may simply be no match for a souped-up Siri who can instantly bring you information and perform tasks for you (order pizza, order pick-up for dry cleaning, etc.).
Other algorithms in use today include those used to judge the credit scores of consumers and businesses. Andras Kornai, a Stanford PhD and professor at the Budapest Institute of Technology with experience in designing credit algorithms, states, “It is no longer a local friendly banker who makes these decisions around credit, and that trend isn’t likely to slow down.” It’s likely that other efficient algorithmic use isn’t going to slow down either, and because there wasn’t much backlash in AI taking over loan and insurance decisions, it seems quite likely that it’ll handle more complex financial issues in the coming decade.
Kornai also refers explicitly to the use of algorithms in specific medical diagnostics, or even in legal proceedings, and believes that slow and steady traction in these domains is somewhat inevitable, and may invariably box out human expertise from tasks such as x-ray assessments or certain kinds of legal research.
Nearly all the researchers I’ve spoken to about automation and the job market have brought up the topic of self-driving cars.
Speech-recognition algorithms of tomorrow may create their own economic shakeups. Daniel Roth received his PhD from Harvard in 1995. He now teaches at University of Illinois and has been working in the domain of natural language processing for nearly 20 years: “In ten years, I can see us being able to communicate with computers in a truly natural way…. I will be able to consult a machine in really thinking through a world problem… a physician will be able to consult a computer to navigate research articles.”
Roth mentions that many millions of medical research articles will be published in the coming decade, and that having a machine that can understand natural commands to sift through this massive swath of information would be of extreme value (i.e., “Find me all the articles published within the last three years in any language that study the impact of air pollution on osteoporosis in men.”). The same natural language algorithms might comb legal files or compliance documents, potentially shaving hours of tedious work from a professional’s day, but also potentially leaving some entry-level positions (such as paralegals) out of a job.
Though the AI researchers I spoke with didn’t tend to converge on similar industries when it came to making predictions, nearly all the researchers I’ve spoken to about automation and the job market have brought up the topic of self-driving cars. To Amir’s point — there seem to be few more visceral ways of “giving up control” than letting the machine take the wheel, and 10-15 years seems to be enough time for many AI experts to suspect that we’ll see consumers buying cars that drive them, not the other way around.
Berleant mentions there has been a steady progression to automatic transmissions, anti-lock breaks, automatic locks and cars that can park themselves. He states, “I believe it’s reasonable to suppose that such completely autonomous cars will be commonplace in ten years.” If even one-tenth of the cars on the road in 10 years are self-driving, the impact on the economy as a whole could be relatively drastic.
Among other sectors, the immediate impact on the job market for motor vehicle operation would be hit the hardest. “There are a million cab drivers in the United States alone — that might be a million people without a job” says Kornai. In addition to direct unemployment for folks in truck driving or taxi driving positions, there also could be a drastic decrease in demand for car ownership if cars can be ubiquitously accessed for transportation with the push of a button on an app.
Car manufacturers might be fighting over a much smaller market of individuals who still wish for a car of their own — or they would battle over who’s autonomous fleets are employed in the most cities. Manufacturing demand for vehicles seems destined to decline sharply under these circumstances.
One of the most pressing concerns for those of us in the working world is the effect of automation on job security.
The incumbents to driverless cars are likely to fight just as fiercely as those currently railing against Uber, and Kornai and others foresee a reasonably gradual shift to autonomous vehicles, and this may cushion the shock of a drastic economic shift.
We might see a way around these legal concerns with a gradual “trust transition” from man to machine, rather than an overt jump from 100 percent human driver to 0 percent human driver. Either way, a lot of very smart AI folks seem to think that the next decade is the one when driverless will kick in.
Like many double-edged effects of technological change and automation, driverless cars may have tremendous upsides, as well. “There’s so much release of human potential if you don’t have to be behind the wheel for an hour per day or more,” says Berleant. This isn’t to say that truck drivers are all going to become tremendously efficient with all the freed up time they have in their hands-free commute to their next job, but it’s a potential example of the silver lining of automation and the job market.
There is (and for the foreseeable future, will continue to be) ongoing debate as to whether or not technological advancements inherently create more job market opportunities than they destroy. The most ignorant arguments are black-and-white, and it’s clear from interviewing subject-matter experts that there is no consensus on the future outcomes, economically or technologically.
What does seem clear is that there are important current automation and AI trends with existing algorithms and technologies that are likely to only have a greater job-market influence in the coming decade, and they are worth keeping an eye on. Maybe machine vision can help us with that.
联合创始人Charles Lee和Ben Cheung均是前VMWare的员工，两人为了创建一个不一样的初创公司需要根据各自日程表来安排时间见面来讨论，然而两人在时间安排上很难达成一致，在这之后他们决定创建Genee。
该公司已经获得145万美元的首轮融资，参投方有Uj Ventures、Streamline Ventures以及Garnett Ventures。这笔资金将会用来扩大Genee的理解能力、整合其他的信息平台（比如说Facebook和Slack）并且推出其他附加的功能，例如用户可以进行预订服务。
Meet Genee, Your Artificially Intelligent Personal Assistant
Genee, an artificially intelligent scheduling app, is launching today into public beta so that anyone can have a personal assistant.
On the heels of Zirtual’s collapse, Genee hopes to take humans out of the equation entirely with its end-to-end scheduling helper that plugs into any existing calendar app and email provider.
Former colleagues at VMWare, co-founders Charles Lee and Ben Cheung decided to build Genee after struggling to coordinate meetings around their own schedules to chat about a different startup idea.
“The reason why executives have assistants is that once they tell them to schedule a meeting, the assistant takes over completely,” says Cheung. “The tools available now are pulling to automate that, but they don’t solve the problem end to end: they’re not taking away the responsibility.”
Anyone can use Genee after allowing access to existing calendar and email apps. You simply copy Genee on your emails, just as you would a personal assistant, and Genee takes over. If you need to move a meeting back by 15 minutes or reschedule, there’s a one-click option that prompts Genee to notify the other parties.
Recently, the team integrated Genee into Twitter so that users can get a meeting on the books without having to revert back to email at all. Of course, the whole process works more smoothly if both parties have allowed Genee to access their schedules, but it’s not necessary.
While many applications of natural language processing leave a lot to be desired, Cheung says the key with Genee is focusing on a very small segment of natural language to make sure the system actually feels like a human assistant.
“We focus the natural language on a single context, scheduling meetings, which makes it easier,” says Cheung. “We have a sophisticated algorithm for scheduling the preferred time, and we’ve trained the computer a lot about the common sense part too.”
If you tell Genee you want to schedule “drinks after work at some point this week,” for instance, she has to know that you probably mean after 5 p.m., and that if you’re completely booked this week, planning drinks for next week wouldn’t be the end of the world.
“We’re not setting out to build a human assistant replacement, it’s a product to address the 99 percent of the population that doesn’t have it,” says Lee. “Our target user is not going to be the C-level executive who already has an assistant.”
Over the past year, 10,000 private beta users have helped the team tweak the system and teach Genee common slang and nuances in scheduling language.
The company has raised $1.45 million in seed funding from Uj Ventures, Streamline Ventures, and Garnett Ventures, which it will use to expand the Genee’s understanding, integrate with additional messaging platforms (such as Facebook and Slack), and roll out additional features, such as booking reservations for users.