Artificial Intelligence sounds freaking amazing: humanoid robots, artificial conscious, self learning systems and understanding the human brain. I won’t lie; these were the things that motivated me to look into Artificial Intelligence. And till a certain extent they still do.
人工智能听起来令人惊讶：类人机器人，人工意识，自我学习系统以及对人脑的理解。 我不会撒谎这些都是促使我研究人工智能的东西。 直到一定程度他们仍然会这样做。
I started out doing Physics and Life Sciences. One thing that caught my attention was the advancements in the field of so called “Artificial Neural Networks”. So I grabbed my laptop and ambitiously started writing my own implementation of these exotic Artificial Neural Networks. However, after a week the project became too time consuming. I halted it till indefinitely and thought to never do anything with Artificial Intelligence ever again… Until I did a few years later.
我开始做物理学和生命科学。 引起我注意的一件事是所谓的“人工神经网络”领域的进步。 因此，我拿起笔记本电脑，并雄心勃勃地开始编写自己的这些奇异的人工神经网络的实现。 但是，一周后，该项目变得非常耗时。 我无限期地暂停它，并认为以后再也不会对人工智能做任何事情了……直到几年之后。
Now I work with Artificial Intelligence on a daily basis. But does Artificial Intelligence meet my expectations?
现在，我每天都在与人工智能合作。 但是人工智能符合我的期望吗？
什么是人工智能？ (What is Artificial Intelligence?)
The term Artificial Intelligence is as hard to define as the term intelligence itself. So I won’t try to make a hard boundary, just a list of things which I think aren’t or are Artificial Intelligence, Machine Learning and — last but most definitely not least — Deep Learning.
术语“人工智能”与术语“智能本身”一样难以定义。 因此，我不会尝试划定硬性界限，只是列出我认为不是，人工智能，机器学习以及(但最后但并非最不重要的)深度学习的东西。
不，这不是人工智能(No, this is not Artificial Intelligence)
Do you know what? I don’t even dare to give an example because every program can be seen as “Intelligent” in their own right. Okay, maybe the code that makes a keyboard button press registered on your screen or something.
你知道吗我什至不敢举一个例子，因为每个程序本身都可以被视为“智能”的。 好的，也许是使键盘按钮按下的代码在屏幕上注册了什么。
是的，这是人工智能 (Yes, this is Artificial Intelligence)
This is one you are probably quite familiar with: Path finding. The algorithms behind this Artificial Intelligence are quite old but elegant in their own right. Dijkstra’s Algorithm and A* are algorithms that use a rule based approach to guaranty the shortest path between two points.
您可能非常熟悉以下内容：路径查找。 人工智能背后的算法虽然很老，但本身就很优雅。 Dijkstra的算法和A *是使用基于规则的方法来保证两点之间最短路径的算法。
这是机器学习(This is Machine Learning)
Artificial Intelligence without Machine Learning always kinda seemed like cheating to me when talking about Artificial Intelligence. It just sounds so boring. However, if it works it works. No need to over engineer a solution! What makes Machine Learning so distinctive from other Artificial Intelligence algorithms is the ability to learn from examples.
在谈论人工智能时，没有机器学习的人工智能总是有点欺骗我。 听起来真无聊。 但是，如果可行，它将起作用。 无需过度设计解决方案！ 使机器学习与其他人工智能算法如此不同的原因是能够从示例中学习。
An algorithm that simply draws a straight line through some data points is already considered Machine Learning. Some daring lines even have curves! This doesn’t really sound that exciting, however it can already do a tremendous job when it comes to automatization of certain processes.
简单地通过某些数据点画一条直线的算法已经被认为是机器学习。 有些大胆的线条甚至有曲线！ 这听起来并不令人兴奋，但是在某些流程的自动化方面它已经可以做得很好。
深度学习，是的 (Deep Learning, o yeah)
There are so many examples of Deep Learning nowadays: OpenAI’s GPT text generator, DeepMinds Star Craft 2 playing AlphaStar, Human face generator StyleGAN by NVidia, Tesla’s self driving cars, and OpenAI’s rubics cube solving robot. Now we are talking! This sounds cool! And one thing they have in common is that they use those exotic Artificial Neural Networks.
如今，深度学习的例子很多： OpenAI的GPT文本生成器，玩AlphaStar的DeepMinds Star Craft 2 ，NVidia的人脸生成器StyleGAN ，特斯拉的无人驾驶汽车以及OpenAI的Ruby立方体解决机器人。 现在我们在说话！ 听起来不错！ 他们的共同点是他们使用了那些奇特的人工神经网络。
人工智能好吗？(Is Artificial Intelligence any good?)
The short answer is: yes. Artificial Intelligence models have prove themselves to be useful over and over again. So it most definitely has its place in this day and age. However, they are all very specialised and seem to be very limited in their capabilities. Great though! But limited.
简短的答案是：是的。 人工智能模型已经证明自己一遍又一遍地有用。 因此，它无疑在当今时代中占有一席之地。 但是，它们都很专业，而且功能似乎很有限。 太好了！ 但有限。
To highlight a great specialised Artificial Intelligence model we can look at one of the projects of Google KI. They developed a model that uses image recognition to give a quicker, cheaper and good analysis of an eye in order to speed up the diagnosis. These kind of applications can seriously reduce the stress on the health care system.
为了强调出色的专业化人工智能模型，我们可以看一下Google KI的一个项目。 他们开发了一种使用图像识别的模型，可以对眼睛进行更快，更便宜和更好的分析，从而加快诊断速度。 这些应用程序可以严重减轻医疗保健系统的压力。
人工智能将给我们带来什么？ (What will Artificial Intelligence bring us in the future?)
But we have all been promised to fly with HAL9000 to the moon and what about a proper cleaning robot? Why can’t my Roomba still not remove the dust from my book shelves?
但是我们都被承诺会带着HAL9000飞往月球，而一个合适的清洁机器人呢？ 为什么Roomba仍然不能清除书架上的灰尘？
There has been a quick surge of Artificial Intelligence systems since the last few decades. This is due to the large availability of data by the Internet, relatively cheap compute by the use of graphics cards to run neural networks and of course new models.
自最近几十年来，人工智能系统Swift发展。 这是由于Internet的大量数据可用性，使用图形卡运行神经网络以及相对而言新模型的相对便宜的计算。
硬件将是关键 (Hardware will be key)
I suspect that we will get more ground breaking achievements when compute becomes even cheaper. A good example of this is the latest GPT-3 network by OpenAI. This network has 175 billion trainable parameters which is still just 1‰ (1 per thousand) of the number of synapses in the neocortex. And training this model is estimated to have cost at least $4.6 million.
我怀疑，当计算变得更加便宜时，我们将获得更多突破性的成就。 一个很好的例子就是OpenAI最新的GPT-3网络。 该网络具有1,750亿个可训练参数，仍仅为新皮质中突触数量的1‰(千分之一)。 估计训练这种模型至少要花费460万美元。
Luckily enough, hardware is still getting cheaper and cheaper. This statement is now even more clear than ever with the release of the new NVidia GeForce RTX 30-series. These cards will be absolute beasts when it comes to Artificial Intelligence; enabling researchers to make even better Artificial Intelligence models.
幸运的是，硬件越来越便宜。 现在，随着新的NVidia GeForce RTX 30系列的发布，这一声明比以往任何时候都更加清晰。 这些卡在人工智能方面将是绝对的野兽。 使研究人员可以创建更好的人工智能模型。
Additionally, there is quite some research going on when it comes to using Quantum Computers for training our Artificial Intelligence models. We do have to be patient for this but Quantum Computers running an Artificial Intelligence already sounds buzzworthy!
此外，有关使用量子计算机训练我们的人工智能模型的研究也很多。 我们对此必须耐心等待，但是运行人工智能的昆腾计算机已经听起来很抢眼！
更多模特自由 (More model freedom)
Cheaper compute will result in more freedom when it comes to trying larger experiments. A lot of ideas are simply too expensive to try for the average Joe with a laptop. Of course you can use the cloud like AWS, Lambdalabs, Google Cloud, Azure, Linode and many others to rent a good computer but this will cost you a couple of Euros per hour.
尝试大型实验时，便宜的计算将带来更多的自由。 对于拥有笔记本电脑的普通Joe来说，很多想法太昂贵了。 当然，您可以使用AWS ， Lambdalabs ， Google Cloud ， Azure ， Linode等许多云来租用一台好的计算机，但这每小时的费用为几欧元。
One thing that I think will be amazing will be using many different objective functions during training an Artificial Intelligence. Most models have just one or at most a handful of different objectives. Like:
我认为会令人惊奇的一件事是在训练人工智能期间使用许多不同的目标函数。 大多数模型只有一个或最多几个不同的目标。 喜欢：
- Is this sentence positive? 这句话是肯定的吗？
OR
要么
- Is this an image of a cat?这是猫的形象吗？
OR
要么
- What move should you make in this game?您应该在这场比赛中采取什么行动？
But that is not how we grew up, right? A baby has probably thousands of different objectives a day!
但这不是我们成长的方式，对吗？ 一个婴儿一天可能有数千个不同的目标！
人工智能会满足我们的期望吗？ (Will Artificial Intelligence Meet Our Expectations?)
Artificial Intelligence keeps amazing us. Progress seems to go strong at every aspect. I have high hopes that we keep getting amazed for the coming years. We will probably get those buzzworthy things, but we just have to be patient…
人工智能让我们惊奇。 似乎在各个方面都取得了进展。 我寄予厚望，希望我们在接下来的几年中不断感到惊讶。 我们可能会得到那些嗡嗡声 事情，但我们只需要耐心等待...
我更多 (More of me)
My website
我的网站
My own projects on youtube:
我在youtube上的项目：
Our Machine Learning and Data Science courses:
我们的机器学习和数据科学课程：
Meetups about Machine Learning in The Netherlands:
在荷兰举行的有关机器学习的聚会：
翻译自: https://medium.com/the-innovation/does-artificial-intelligence-keep-its-promises-a9560b2a3ba5
1841
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
