selenium初学者指南_如何进化人工智能生活：初学者指南

selenium初学者指南

‘Evolving artificial intelligent life’ might sound like a grandiose claim from an individual — perhaps even science fiction. I would argue your skepticism speaks more to how we’re conditioned to think about life rather than the merits of the claim itself.

“不断发展的人工智能生活”听起来像是个人的宏伟主张，甚至可能是科幻小说。 我想说的是，您的怀疑更多地说明了我们如何有条件思考生活，而不是主张本身的价值。

Prevailing consensus would have you believe that life in the Universe is vanishingly rare, or somehow exceedingly difficult to create. A simple thought experiment, used to better understand Fermi’s paradox, can shed light on the astronomical odds stacked against the ‘rare life’ hypothesis.

普遍存在的共识会让您相信，宇宙的生命正在消失，或者很难创造。 一个简单的思想实验，用于更好地理解费米悖论 ，可以阐明与“ 稀有生命 ”假说相对应的天文几率。

To wit, life forming chemistries are most certainly not uniquely Carbon and water based in a narrow pressure band in a narrow temperature range between 0 and 100 degrees Celsius. Instead, they may appear anywhere up and down the temperature, radiation and pressure spectra, across a near infinite range of substances (i.e. Methane is liquid at low temperatures, Iron is liquid at high temperatures).

简而言之，生命形成化学肯定不是唯一地在0到100摄氏度之间的窄温度范围内的窄压力带中以碳和水为基础的。 取而代之的是，它们可能会在几乎无限范围的物质中出现在温度，辐射和压力谱的上下任何位置(例如， 甲烷在低温下为液态，铁在高温下为液态 )。

Life forming chemistries are most certainly not uniquely Carbon and water based in a narrow pressure band in a narrow temperature range between 0 and 100 degrees Celsius.

可以肯定，形成生命的化学物质并不是唯一基于碳和水的碳和水，它们是在0至100摄氏度的狭窄温度范围内的狭窄压力带中产生的。

The only commonality these chemistries would need is the ability to store and transmit data, making radiation (light) a potential substrate for life, and possibly even Gravity.

这些化学唯一需要的共同点是能够存储和传输数据，使辐射( 光 )成为生命的潜在底物，甚至可能成为重力。

Perhaps the only fundamental prerequisite for life is the ability to store and transmit data.

生命的唯一基本前提可能是存储和传输数据的能力。

Computers have everything we need. But I digress…

计算机拥有我们需要的一切。 但是我离题了……

I’m not talking about AI (artificial intelligence) in its current form. Neural networks are imitations of what we know about how brains work. They’re intelligently designed.

我不是在谈论当前形式的AI( 人工智能 )。 神经网络是我们对大脑工作原理了解的模仿。 它们经过精心设计。

You don’t design artificial intelligent life. That’s not a natural (or elegant) way to do it. Instead, I’m going to provide initial conditions suitable for intelligent life to evolve. Nature did the exact same thing for us.

您不会设计人工智能生活。 这不是自然的( 或优雅的 )方式。 相反，我将提供适合智能生活发展的初始条件。 大自然为我们做了完全相同的事情。

You don’t design artificial intelligent life. That’s not a natural (or elegant) way to do it.

您不会设计人工智能生活。 这不是自然的( 或优雅的 )方式。

But where to begin?

但是从哪里开始呢？

There are a couple of important concepts underpinning my approach, so let’s start there.

我的方法有两个重要的概念，因此让我们从这里开始。

启发式简介 (Intro to heuristics)

Heuristics are a natural way to tackle wildly complex problems. I don’t mean natural for us, I mean natural, as in Nature — the Universe.

启发式方法是解决极其复杂的问题的自然方法。 我对我们不是自然的，而是自然，就像在《 自然—宇宙》中一样 。

A heuristic is a comparative process not guaranteed to produce a perfect result.

启发式是无法保证产生完美结果的比较过程。

I first began using heuristic algorithms to tackle a famous type of NP-Hard math problem known as the Traveling Salesman Problem. Essentially you need to work out the most efficient way to visit a specified number of locations. Sounds easy, right?

我首先开始使用启发式算法来解决著名的NP-Hard数学问题，即旅行商问题 。 本质上，您需要找出最有效的方法来访问指定数量的位置。 听起来很简单，对吧？

The difficulty arises in the sheer number of permutations involved. According to this overview of route optimization:

困难在于涉及的排列的绝对数量。 根据此路线优化概述 ：

A simple route optimization problem with a single vehicle and 50 stops has 30 million times more permutations than atoms of Hydrogen in the Sun (10⁵⁷).

一个简单的单车50停靠路线优化问题，其排列比太阳中的氢原子(10 ^)多3000万倍。

It’s easy to confirm this using the bang (!) symbol on your scientific calculator. The number of permutations can be calculated as follows:

使用科学计算器上的bang( ！ )符号可以很容易地确认这一点。 排列数量可以计算如下：

50 x 49 x 48 x 47 x … x 2 x 1 = 50! = 3 x 10⁶⁴

50 x 49 x 48 x 47 x…x 2 x 1 = 50！ = 3 x10⁶⁴

The fastest supercomputer in the world would take many trillions of years to check each possibility, rendering this simple problem (faced by delivery companies and couriers every day) intractable without heuristics.

世界上最快的超级计算机将花费数万亿年的时间来检查每种可能性，从而使这个简单的问题( 每天由送货公司和快递公司面临的问题)变得难以解决，而无需进行启发式分析。

By contrast, a system using heuristics can deliver the following optimal route for a 50 location tour in a matter of seconds:

相比之下，使用启发式系统的系统可以在几秒钟内为50个位置的游览提供以下最佳路线：

Optergon. Optergon提供的具有50个停靠点的车辆的最佳路线。

Heuristic algorithms often find their inspiration in Nature. From the gathering of food by ant colonies to mutations in our DNA, heuristics are an integral part of life. Without heuristics there could be no evolution.

启发式算法通常在自然界中找到灵感。 从蚁群收集食物到我们DNA的突变，启发法都是生活中不可或缺的一部分。 没有试探法，就不可能有进化。

What’s weird, but important to bear in mind as we continue, is that naturally occurring heuristics can emerge entirely new heuristics. The term emerge is also really important. I’m going to cover it in more detail shortly.

奇怪但很重要的一点是，随着我们的继续前进，自然发生的启发式方法可能会出现全新的启发式方法。 出现这个词也很重要。 我将在稍后详细介绍。

Naturally occurring heuristics can emerge entirely new heuristics.

自然发生的试探法可以出现全新的试探法。

I’ll explain.

我会解释。

Our brains use heuristics all the time. We can’t catch a ball without them.

我们的大脑一直都在使用启发式方法。 没有他们，我们就无法接球。

Instead of calculating Newton’s laws to predict the flight path of a ball, we compare the current situation (i.e. a ball flying toward us) against a library of stored memories and estimate where and when the ball will arrive based on that. The more we practice, the more stored memories and experiences we have to base our decision on, the better we become. Since heuristics can’t guarantee a perfect result, even the best of us occasionally fumble.

与其计算牛顿定律来预测球的飞行轨迹，不如将当前情况( 即一个球飞向我们 )与存储的内存库进行比较，并据此估计球何时何地到达。 我们练习的次数越多，决定所依据的记忆和经验就越多，我们就会变得更好。 由于试探法不能保证完美的结果，因此即使是我们中最好的人也会偶尔摸索。

The ‘ball catching’ heuristic was not present when life first evolved. It wasn’t present when life first crawled onto land. It emerged some time after as a result of the genetic heuristic process of evolution.

当生活第一次进化时，就不会出现“ 接球式 ”启发式方法。 当生命第一次爬到土地上时，它并不存在。 它是遗传启发式进化过程的结果。

In other words, one heuristic process (evolution) emerged another (ball catching).

换句话说，一个启发式过程( 演变 )出现了另一个( 接球 )。

That’s important!

那很重要！

出现简介 (Intro to emergence)

Emergence is a beautifully elegant phenomenon that seems to effortlessly violate the laws of thermodynamics. One of the Universe’s true gems.

出现是一种优美而优雅的现象，似乎毫不费力地违反了热力学定律。 宇宙的真正宝石之一。

Emergence is a beautifully elegant phenomenon that seems to effortlessly violate the laws of thermodynamics.

出现是一种优美而优雅的现象，似乎毫不费力地违反了热力学定律。

We observe emergence when a system exhibits properties not initially present in any of its constituent parts.

当系统显示出其任何组成部分中最初不存在的属性时，我们会观察到出现。

There are plenty of magnificent examples visible with the naked eye. Look out your window. At the start of the solar system there was no planet Earth. There was no moon. There was no sun. All of these things emerged over time from a cloud of dust and gas acted upon by gravity.

肉眼可见许多宏伟的例子。 看着窗外。 在太阳系开始时，没有行星地球。 没有月亮。 没有阳光。 随着时间的流逝，所有这些东西都是从重力作用下的尘埃和气体云中浮现出来的 。

Even once the nuclear fires of the sun ignited and our planet swept clean its orbit, there was no life here. There were no flowers. No trees. No Kakapos (sadly, we’re killing many of the beautiful things Nature has emerged and the Kakapo is likely soon to be one of them). All of these living things emerged from genetic heuristic processes (i.e. mutation) acted on by adaptation for survival.

即使一旦点燃了太阳的核火焰，并且我们的星球清扫了轨道，这里也没有生命。 没有花。 没有树。 没有卡卡波( 可悲的是，我们正在杀死自然界已经出现的许多美丽事物，卡卡波很 可能很快成为其中的一员 )。 所有这些生物都是从适应生存的遗传启发过程( 即突变 )中产生的。

In short,

简而言之，

Simple heuristic systems can emerge complex life and intelligence.

简单的启发式系统可以出现复杂的生活和智力。

That’s important!

那很重要！

“启发式出现”是发展智力的好方法 (‘Heuristic emergence’ is a fine way to evolve intelligence)

Heuristic emergence, the term I’m using to describe my approach to evolving an AI, is the most powerful natural phenomenon you’ve never heard of. It is responsible for all life in the Universe. It is responsible for the Kakapo. It is responsible for our intelligence.

启发式涌现是我从未听说过的最强大的自然现象，这是我用来描述我发展AI的方法的术语。 它负责宇宙中的所有生命。 它对卡卡波负责。 它负责我们的智力。

Evolution is a specific example of heuristic emergence.

进化是启发式出现的一个具体例子。

The beauty of using heuristic emergence to evolve an artificial intelligence is that it requires no designer or guiding hand. Simply set up the initial conditions required for virtual organisms (VOs) to evolve. Press play.

使用启发式涌现来发展人工智能的好处在于，它不需要设计者或指导者。 只需设置虚拟生物 ( VOs )进化所需的初始条件。 按播放。

Sooner or later (in exactly the same way our intelligence emerged from natural selection) they’ll start exhibiting traits like common sense, problem solving, hunches, and creativity. They wouldn’t be designed by humans and trained using machine learning (ML). They would be their own thing. Independent and self-aware.

迟早( 以与我们从自然选择中获得的智慧完全相同的方式 )，它们将开始表现出常识，解决问题，预感和创造力等特征。 它们不会由人设计，也不会使用机器学习 ( ML )进行培训。 他们将是他们自己的事。 独立和自我意识。

Same as us.

和我们一样

概念证明：使用启发式出现的人工进化 (Proof of concept: artificial evolution using heuristic emergence)

Recently I had the time to take a few concrete steps towards building a heuristically emergent system. I call it Grasslands.

最近，我有时间采取一些具体步骤来构建启发式紧急系统。 我把它叫做草原 。

Grasslands is a virtual environment that supports a single type of VO (in this case, grass) consisting of a single gene that codes for size. I tried to keep the rules of the system as close to what we intrinsically recognize as a natural grassland environment in order to make this demonstration more relatable.

草原是一种虚拟环境，支持单一类型的VO( 在这种情况下为grass ) ，该类型的VO由编码大小的单一基因组成。 我试图使系统规则与我们固有地认为是天然草地的环境接近，以使该演示更加相关。

• In parts of the environment, water and rainfall are plentiful and in other parts water is scarce.
  在部分环境中，水和雨量充沛，而在其他部分中，水稀缺。
• For grass to survive it must have sufficient water.
  为了使草生存，必须有足够的水。

• Grass can grow, reach sexual maturity, reproduce (with a small chance for mutation) under the right conditions, and ultimately die.

  草可以在适当的条件下生长，达到性成熟，繁殖( 发生突变的机会很小 )并最终死亡。

• Large grass needs more water to survive than small grass but can reproduce faster, if conditions are suitable.
  大草比小草需要更多的水才能生存，但如果条件合适的话，繁殖速度更快。

That’s all there is to it.

这里的所有都是它的。

The simulation runs over a period of time without any interference and displays a snapshot of what the environment looks like every few virtual years.

该模拟运行了一段时间，没有任何干扰，并且每隔几年虚拟环境就会显示一次快照。

The following screenshot, taken from my PC, shows the initial conditions used (obviously they are configurable) and what the environment looks like after 20 years.

下面的屏幕快照来自我的PC，显示了使用的初始条件( 显然，它们是可配置的 )以及20年后的环境。

Grasslands heuristic emergence environment initial conditions.

草原启发式出现环境的初始条件。

From the sand colored map at the bottom corner of the screenshot you might be able to make out the faint outline of a river flowing from the North-West towards the center, before draining into an inland delta (like the Okavango in Botswana). The river and delta systems are fed by higher rainfall in the North-West region (average rainfall amounts are not pictured on the map).

从屏幕截图底角的沙色地图中，您可以确定一条从西北流向中心的河流的模糊轮廓，然后再排入内陆三角洲( 例如博茨瓦纳的Okavango )。 河流和三角洲地区的西北部地区降雨较多 ( 平均降雨量未在地图上显示 )。

Initially, there are two small grass colonies taking hold — one at the top-center of the map and one in the center, near the delta. As per the starting conditions, both of the initial grass colonies have gene expression size 20.

最初，有两个小的草丛殖民地-一个在地图的顶部中心，另一个在三角洲附近的中心。 根据起始条件，两个初始草丛的基因表达大小均为20。

The vast majority of the map does not have sufficient water to support grass of gene size expression 20. Simply put, the seed colonies will have to evolve in order to colonize drier regions of the map.

该图的绝大部分没有足够的水来支持基因大小表达20的草。简单地说，种子殖民地将不得不进化以便定居图的较干燥区域。

After 240 simulated years, the map looks like this (the results differ every time since the system is non-deterministic and there is no way to accurately predict the outcome — other than general trends).

经过240个模拟年后，该图看起来像这样( 由于系统是不确定性的，因此每次结果都不同，除了一般趋势外，没有其他方法可以准确地预测结果 )。

Grasslands heuristic emergence simulation showing adapted VOs.

草原启发式紧急情况模拟显示适应的VO。

Dark green regions indicate large, lush grasses that have colonized the wetter regions in the North-West and along the river and inland delta. The largest grass has evolved to a gene size expression of 111. Significantly larger and faster breeding than the seed colonies.

深绿色区域表示茂密的大草丛，遍布西北地区以及沿河和内陆三角洲的湿润地区。 最大的草已经进化为基因大小表达为111的植物。育种比种子菌落大得多且快。

The smallest grass has evolved to a gene size expression of only 7, making it well suited to dry regions. Note that the smaller grass, depicted on the map as light green tiles, has colonized the entire available area despite the fact that this was not initially possible for the seed grass.

最小的草已经进化成基因大小表达仅为7，使其非常适合干旱地区。 请注意，较小的草(在地图上描绘为浅绿色的瓷砖)已经占领了整个可用区域，尽管事实上这对于种子草是不可能的。

The salient point here is that the system is a proof of concept for heuristic emergence in general, and artificial evolution in particular.

这里的重点是，该系统通常是启发式出现 (特别是人为进化) 的概念证明 。

Grasslands is a proof of concept for heuristic emergence in general, and artificial evolution in particular.

草原通常是启发式出现概念的证明，尤其是人工进化。

There is no code, no directive, no program, no function, or anything else, that directs the grass to adapt to drier regions of the environment. It evolves because random mutations in its DNA, over time, lead to traits that allow it to survive and reproduce in a drier environment— passing on those traits to its offspring.

没有任何代码，指令，程序，功能或任何其他命令可以指导草适应环境更干燥的区域。 它之所以进化，是因为随着时间的流逝，DNA中的随机突变会导致一些特性，使其能够在干燥的环境中生存和繁殖-将这些特性传递给后代。

Same as us.

和我们一样

走向进化革命 (Onward to the evolution revolution)

Machines possess many aspects of intelligence. They are good calculators. They have memory. They can communicate. AIs can learn specific intelligent behaviors, such as recognizing faces or objects.

机器拥有许多方面的智能。 他们是很好的计算器。 他们有记忆。 他们可以沟通。 人工智能可以学习特定的智能行为，例如识别面部或物体。

They do a lot of things better than us. Yet we don’t really consider them smarter than us. Why? Computers are presently devoid of heuristics, like common sense, empathy, and creativity.

他们比我们做得更好。 但是我们并不认为它们比我们更聪明。 为什么？ 目前，计算机没有常识，同理心和创造力之类的启发式方法。

Computers are presently devoid of heuristics, like common sense, empathy, and creativity.

目前，计算机没有常识，同理心和创造力之类的启发式方法。

Yet heuristics are precisely those aspects of intelligence that are best emerged using… heuristics. Nature has already shown us how to do this by evolving us. I simply intend to follow the road-map already laid out.

然而，启发式方法恰恰是使用……启发式方法最能体现情报的那些方面。 大自然已经向我们展示了如何通过进化来做到这一点。 我只是打算遵循已经制定的路线图。

The next step is to evolve a virtual organism possessing a heuristic behavior, such as creativity or common sense.

下一步是发展一种具有启发性行为(例如创造力或常识)的虚拟生物。

Right now there are a few challenges (namely, finding the time). On the technical front I need to find a way to tie the emergence of new genes to a specific characteristic. These characteristics, working in synergy, will ultimately emerge the new heuristic.

现在有一些挑战( 即找到时间 )。 在技​​术方面，我需要找到一种方法将新基因的出现与特定特征联系起来。 这些协同作用的特征最终将出现新的启发式方法。

Remember, I can’t simply create a gene expression for creativity, like I did with the size gene in Grasslands. Size is a rudimentary trait. Creativity is a complex behavior. Simply saying a gene codes for creativity won’t make it so.

请记住，我不能像创造《 草原》中的size基因那样简单地创造基因表达以创造力 。 大小是最基本的特征。 创造力是一种复杂的行为。 仅仅说一个基因编码创造力就不会成功。

I have some ideas on how to do that and I’ll keep you posted on my progress (or die under suspicious circumstances — in which case someone really needs to destroy my PC and cloud accounts). In the meantime, I’d like to leave you with the following convoluted thought.

我对如何执行操作有一些想法，我会告诉您我的进度( 或者在可疑情况下死亡-在这种情况下，确实需要销毁我的PC和云帐户 )。 同时，我想让您摆脱以下困惑的想法。

Nature used heuristic emergence to evolve life imbued with heuristics that allow it to use heuristic emergence to emerge life imbued with heuristics.

大自然使用启发式出现来进化充满启发式的生活，从而使自然界能够使用启发式出现来展现出充满启发式的生活。

Seems Nature has a kind of recursive poetic symmetry.

似乎大自然具有一种递归的诗意对称性。

翻译自: https://towardsdatascience.com/how-to-evolve-artificial-intelligent-life-a-beginners-guide-2fdd1336222c

selenium初学者指南