未登录,请登录后再发表信息
最新评论 (0)
播放视频

A.I. 不断发展的前方是什么

A.I. is Progressing Faster Than You Think!

你现在收看的是冷流科技
You’re watching ColdFusion TV.
冷流科技
[Music playing]
大家好 欢迎再次收看ColdFusion视频栏目
Hi, welcome to another ColdFusion video.
去年年末
At the end of last year,
我制作了一个视频 介绍了
I made a video about five new technologies
五个大部分人都没能预料到的科技成果
which most people didn’t expect to exist.
本期节目
The episode included an AI that could look at a static image
创造出它认为接下来几秒会发生的动态视频
and dreamed up a video of what it thought would happen in the next few seconds.
是的
Yes, it created a video from a still image.
我还介绍了另一个人工智能
I went on to describe another AI that could
它能够仅仅听音频
guess what’s happening in the scene
就推断出发生的场景
just by listening to the audio.
反复实验证明它能比人类做得更好
It was consistently better than humans at doing this.
在同一个视频里 我还提到另外一个人工智能
And in the same video I talked about yet another AI
只需要通过观察情景剧的训练
that could predict human behavior
就能预测人的行为
with training only from watching sit-coms.
如果你还没看过这个视频 就可惜了
If you haven’t seen this video yet, you’re missing out.
视频描述里第一个链接就是那个视频了
The link to the video will be the first one in the description.
今天我们要看看另一种很有趣的人工智能
Today, we will be looking at another interesting AI,
想象一下 你打出一个描述场景的句子
imagine typing a descriptive sentence of the scene
然后人工智能就可以
and having an artificial intelligence
仅仅根据你的描述 创造出一个
generating a convincing photo-realistic image
像照片一样逼真的照片
just from your text input.
这样的人工智能只要刚提出来 我们今天来看看
This is just being created and that’s what we will take a look at today.
然后 我会分享一下我大胆的想法
After this, I’ll share some of my wild thoughts
关于人工智能的看法
on artificial intelligence.
我们直入主题吧
Let’s get straight into it.
在进入人工智能的世界之前
Before we dive straight into the AI,
我们需要了解一些背景知识
we need to build up a bit of context.
我们接下来要认识的人工智能叫做神经网络
The type of AI that we are about to look at is called the neural-network.
它大致是一个模拟人脑的计算系统
That’s basically a computing system that’s modeled after the human brain.
它由一群处理节点构成神经
There’re processing nodes that act as neurons and
神经元层就像大脑的部分一样运作
the neuron layers behave as segments of the brain.
这个概念并非最近产生的 80年代就已经提出
This concept of an idea is nothing new and has been around since the 1980s.
但直到五年前才真正实现
But it’s only become feasible in the last five years
这得益于拥有成百芯片的图像处理器价格降低 易于获得
due to GPUs with hundreds of cores being cheap and accessible.
除此之外
In addition to this,
普通网络各种公开的资源工具
the available open source tools from usual networks
使得创造神经网络更为简单
are making it easier to create,
从而推动领域的飞速进步
pushing progress parabolically faster.
利用神经网络的人工智能包括谷歌的阿尔法Go
So some examples of neural networks include Google’s AlphaGo
以及这样的人工智能
and also things like this,
它能够用文本描述图片信息
an AI that’s capable of describing images in text form.
另一个很有趣的神经网络是WaveNet
Another interesting example of neural networks is WaveNet
它能生成类似讲话和音乐的原始语音
and AI that can generate raw sound such as speech and music
下面有几个例子
Here are some examples,
第一个例子 WaveNet通过卷积的方式生成语音
this first example is speech generated by WaveNet with a twist
它并没有被指定说什么内容
WaveNet wasn’t specified what to say,
所以它必须产生自己的语言
so it had to make up its own language,
请留意呼吸声和自然的唇音
notice the breathing and natural lip-sounds.
[WaveNet语音]
[WaveNet Speech]
第二个片段是WaveNet生成的音乐
This next clip is the music that WaveNet generated
同样 没有任何指令告诉它放什么音乐 也没有乐谱
again, it wasn’t told what to play, there was no musical score,
它即兴创作
it just played whatever it wanted.
为了能生成音乐 WaveNet通过“听”
For this piece, WaveNet was trained
古典音乐片段来训练
by listening to classical music pieces.
这很棒 如果想再进一步呢?
So this is great, but what if you want to take things further?
如果结合两个神经网络 让它们相互竞赛
What if we combine two neural networks together and make them compete against each other
它们不就可以不需要人类干预自己训练提高了吗
so that they can train and improve themselves without human intervention.
这就是STackGAN这个人工智能正在做的
That’s what our featured AI called StackGAN is doing.
它的其中一个神经网络产生图像
It uses one neural network to generate images
同样系统里的另一个神经网络
and another neural network within the same system
来判断图像是真实的还是假的
to decide if the images generated are real or fake.
结果是 产生图像的网络通过判断网络的反馈
What ends up happening is that the generative neural network improves itself generating images
不断提高自己产出的图像的逼真度
based on the feedback given by the deciding worknet
而同时 判断网络
and in the same stride, the deciding network
也能更好的辨别图像的真伪
gets better at distinguishing what’s real and fake.
这就产生了一个不需要人类干预的持续反馈循环
This creates a feedback loop of continuous improvement without human intervention.
最终的结果是创造出了低分辨率的合成图像
The end process of this is the creation of a low resolution-synthesized image.
在这之后 就进入第二阶段
After this, a second stage takes place.
在第二阶段
In the second stage,
人工智能被要求消除原来图像中的所有瑕疵
the AI is told to clean up any defects in the original picture,
结果也令人瞠目结舌
and the results are nothing short of stunning.
你看到的是人工智能通过文字描述创造出的图像
Yes, what you’re looking at are images created just by a written text description.
这种AI双元网络被称为生成对抗网络
This type of dual network of AI is called the Generative Adversarial Network,
我在之前的视频中介绍的
and it’s the same type of system that
能够通过静态图像生成动态视频的AI
allows the AI mentioned in my previous video
也是靠这个系统
to dream up those videos when shown as a still image.
这种人工智能实际上是新产物
This form of artificial intelligence is actually brand new
2014年才发明出来
and was only invented in 2014
我已经开始觉得
and already I’m beginning to think that
这是最强大的机器学习方法
this is one of the most powerful methods for machine learning.
这款文本图像AI紧跟着卡耐基梅隆大学
The text image AI comes hot on the heels of an artificial intelligence
的人工智能后出现
from Carnegie Mellon University
在德州扑克比赛中打败了世界最厉害的选手
beating the world’s best players of Texas Hold’em poker.
我记得在我介绍深层思维的视频里 有人评论
I remember some people commenting on my deep mind video,
跟我说 实际上的情况是
telling me that this exact event was something that
人工智能坎坷前行了一段时间
artificial intelligence has been struggling with for some time.
波士顿的Gamellon公司发布了另一种人工智能
There’s also an AI that’s been released by Gamellon, a Boston company.
这款人工智能可以根据经验和概率
This one can rewrite its own code based on
自动修改代码 而不是基于“冷冰冰”的变量
experience and probabilities rather than hard variables.
开发者表示 这可以让编写代码沉闷的过程完全让人工智能自己完成。
It’s creators say that it could make the tedious part of coding of AI completely automatic.
为了让你了解人工智能领域发展到什么地步
To give you an idea how far some areas of AI progressing
我截取了一个片段 杰夫·迪恩是谷歌研究团队资深研究员
Here is a clip from Jeff Dean, a senior fellow at Google’s research group,
这是他在TED上的演讲
giving a presentation at a TED Talk,
就拿计算机视觉这个例子来说
To take an example, the field of computer vision,
每年都有这样一个比赛
every year there is a contest where
各种机器人要通过观察图片
teams compete to see who can give the right categories
然后在上千个种类中选出属于该图片的种类
out of a thousand different categories when giving an image.
2011年 还没有神经实验室之前
And in 2011, before people were using neuron labs
获胜的机器队伍的错误率是26%
the winning team got an error rate of 26%
这结果不是很好 毕竟人类在这项任务上
which doesn’t sound too good when you think that humans are
错误率仅有5%
5% on this task.
但是 发展得很快 在短短五年时间
But, fastforward, just five years,
机器完成任务的错误率只有3%
we are now at 3% errors,
通过深度学习和更强的计算能力
using deep learning and much more computational power.
机器实际上在这项任务上已经超过了人类
we are actually better than humans on this task.
所以从某个角度来看 机器人已经第一次
So in a way, computers can now see and recognize objects
比人类更能精准地识别物体了
better than us for the first time ever,
这在之前从未发生过
this’s never happened before.
也是在那场演讲中 他举了另一个例子
In the same talk, he goes on to give an example of
指明了眼科医生在诊断某些
how unreliable human ophthalmologists were
眼科疾病上有多么的不准确
for diagnosing certain eyes diseases.
在实验当中 任意两个眼科医师
In the experiments, any two human ophthalmologists
同意彼此诊断的概率只有60%
will only agree with each other’s diagnosis 60% of the time.
更可怕的是 如果你让任意一个眼科医师
And what’s worse, if you give any single ophthalmologist
观察他们几个小时之前诊断的眼科图像
the exact same image that they diagnosed a few hours earlier,
他们同意之前的诊断的概率只有65%
they will only agree with themselves 65% of the time
2017年 在这个领域 已经证明
In 2017, artificial intelligence image recognition
人工智能的图像识别要比专业人员做得好
has been proven to perform better than professional humans in this field.
正如你在短片中看到的一样 人工智能发展迅猛
As you’re seeing this video, things are progressing fast
也许已经让你们当中有些人大吃一惊
and may have taken some of you by surprise.
即便是Alphabet公司的董事 Sergy Brin
Even the president of Alphabet, Sergey Brin,
也对整个人工智能领域的表现惊叹不已
is being taken by surprise by the entire phenomena of AI in general.
…人工智能 但老实说我之前并没有留意过它
AI effort, but, I didn’t pay attention to it at all, to be perfectly honest,
你也知道 我自己是在90年代 经过训练后成为一名计算机科学家的
and you know, myself have been
经过训练后成为一名计算机科学家的
trained as a computer scientist in the 90s,
那时人人都知道人工智能是不可能实现的
everybody knew AI didn’t work.
并不是说 你知道 他们都试过了
It’s not like, you know, people tried it,
神经网络也试过了 都不成功
they tried neural nets, none of them worked out.
是的 这场深度网络的革命
Yeah, this kind of revolution in deep nets
非常的彻底
has been very profound and
尽管我就是行业的人 还是完全震惊了我
definitely surprised me, even though I was like right in there.
现在人工智能处于不可思议的阶段 非常难预测
It’s an incredible time and it’s very hard to forecast.
机器到底能做什么 我们还没有找到极限
You know, what can these things do, we don’t really know the limits.
也许是一切我们能够想到的 也许还有更多
Everything we can imagine and more,
这是很难想通的 真的有无法预测的可能性
it’s a hard thing to think through and has really incredible possibilities.
似乎人们以玩一玩的心态对待人工智能的时代已经过去
It seems like playtime has over in regards to AI
尤其是在深度学习和神经网络技术发展的现在
especially with techniques like deep learning and neural networks
我想这也会使社会充满焦虑 因为AI已经开始取代一些工作了
I think such things will cause social discomfort as they begin to encroach upon jobs.
尽管它有好处也有坏处
There’s definitely both positives and negatives to this though,
但重要的是 要弄明白这意味着什么
but the question is, what does all of this mean
以及如何面对我们认为可能发生的持续的转变
and what are we to do with all of these continual shifts and what we thought was possible.
总得来说 人工智能可以被视作200多年前开始的
In a general sense, artificial intelligence can be seen as a continuation
一直持续不断的工业革命
of the industrial revolution happening over the passed 200 years.
到了现在
Even right now,
人工智能已经可以帮助医疗领域工作
artificial intelligence is helping in the medical field
帮助汽车驾驶和其他危险工作更安全的进行
and making driving and other risky tasks safer.
[画外音]这是一个问题 我觉得
[voiceover]…that is a question and I think…
它拯救了我的
[voiceover] That just saved my…
在以后 许多单调乏味的工作可以由人工智能来处理
In the future, many mundane tasks can be handled by AI,
人们就有更多自由时间来创造
giving more free time for people to do creative things.
但是当然 我也听过这样的说法
But, of course, and I’ve heard this argument before:
并非人人都十分有创造力
not everyone is creative.
生活必须有意义 对于某些人来说
Life needs to have meaning, and for some individuals,
日复一日 朝九晚五的工作就是他们的意义
that meaning comes in the form of their day-to-day, nine-to-five job.
我曾经想过
I’ve been thinking about
制作一个应对人工智能大举进攻相应的金融解决方案
making a video about possible financial solutions to AI disruption.
这包括基于区块链技术的
This includes concepts such as universal basic income
全民基本收入的概念
based on blockchain technology
还想做一期深度剖析研究者们对比的看法
and insights into how some researchers are starting to thinking about such things.
当然你们有些人可能会想到《天网》的情节
Now, of course, that some of you going to be thinking of a Skynet scenario
但目前为止 一切预测还太早
but right now, it’s still far too early to predict
现在预测未来发展成熟的人工智能
Predicting the fully developed AI of the future
就像在60年代试着预测今天的
would be like trying to predict today’s computing industry
计算产业一样困难
from the point of the view of the 1960s.
我个人觉得我们更应该关注的是
Personally I think what we all should actually be worried about
谁控制人工智能
is who controls the AI.
如果未来强大的人工智能
If the strong artificial intelligence of the future
能够向人人开放
is open-source and available to everyone,
这将很有可能是一件好事
there’s a high probability that this will be a good thing.
有一个开放的人工智能被很好利用的小例子:
Here is a small example of an open-source AI being put to good use:
日本一位农民通过人工智能的计算机视觉和机器人技术
A farmer in Japan used AI computer vision and robotics
自动给他的黄瓜筛选分类
to enable his cucumbers to be automatically selected and categorized,
这减少了他几百个小时的工作量
saving him hundreds of hours.
协助他农场工作的人工智能叫做TensorFlow
The artificial intelligence driving his farm is called TensorFlow
它是由谷歌开发的 这项技术向公众开放
and it’s open-source and was created by Google.
也许将来 比起为公司工作
Maybe in the future, instead of working for a company,
更多的人在人工智能的积极辅助下
there will be more common for people to run their own businesses
可以自己发展生意
with their own personal AI acting as a proactivity aid.
这就是向公众开放的人工智能被有效利用的例子
So that’s an example of open-source AI being used in a good way,
但相反的是
But, conversely,
如果一些世界上最精尖的人工智能被少数人所控制
if some of the best AI in the world is held in the hands of a few,
那总的来说 出现好结果的可能性就大大降低了
the probability of a good outcome overall drops significantly.
如果是后一种情况 那么可能是
In the latter case, it might be a case of
得AI者得天下
whoever owns the AI makes the rules.
为了防止人工智能被不正当的利用 防止它变得危险
To stop AI getting into the wrong hands and to stop it becoming dangerous,
已经设立了Future of Life 这样的机构
institutions such as the Future of Life Institute have been set up.
机构拥有上千万美元的资金
It has tens of millions of dollars in funding,
它说明人们已经开始考虑走向未来正确的下一步
and shows that people are already thinking about the next best steps for the future.
本期视频也快结束了 我想把这个问题交给你们
That’s pretty much the end of this video, and I’m gonna hand the ball to you guys.
你怎么看待过去几年里人工智能领域的进步呢
What do you think of the progress being made by AI in the past few years?
请在下方评论区留下你的想法
Leave your thoughts in the comment section below.
总之 感谢收看 这里是GOGO 你正在收看的是ColdFusion
Anyway, thanks for watching guys, this’ve been GOGO, you’ve been watching ColdFusion.
如果你是恰好路过的话 快订阅这个节目
Subscribe to this channel if you just stumble across it,
还有一定要看一看我另一个视频
definitely get to check out my other video
《你没能预计会出现的五个新技术》
in the Five New Technologies That You Wouldn’t Think to Exist Yet.
当然 下次节目我们很快会再见
And of course, I’ll see you again soon for the next video.
祝大家过得愉快!
Cheers guys, have a good one!

发表评论

译制信息
视频概述

AI的发展是必然的。我们要如何保证AI被正确地利用才是重点。

听录译者

海参崴

翻译译者

陈榆木

审核员

译学馆审核员B

视频来源

https://www.youtube.com/watch?v=mQO2PcEW9BY

相关推荐