2021年经济学人 一种优化机器人的新奇方法（在线收听）

Science and Technology

科技版块

Unimals

通用动物

Balls, sticks and the Baldwin effect

球形，棍形和鲍德温效应

A novel way to optimise robots

一种优化机器人的新奇方法

It might sound obvious that if you want to improve a robot's software, you should improve its software.

如果你想改进一个机器人的软件，你就应该改进它的软件，听起来像是明摆着的事。

Agrim Gupta of Stanford University, however, begs to differ.

但斯坦福大学的阿格里姆·古普塔却不这么认为。

He thinks you can also improve a robot's software by improving its hardware—that is, by letting the hardware adapt itself to the software’s capabilities.

他认为还可以通过改进硬件来改进机器人的软件，这种对硬件的改进是让硬件去自适应软件的功能。

As they describe in Nature Communications, he and his colleagues have devised a way of testing this idea.

据他和同事在《自然通讯》中的描述，他们设计出了一种方法来测试这一想法。

In doing so, they have brought to robotics the principles of evolution by natural selection.

在此过程中，他们把通过自然选择来实现进化的原理引入了机器人科学领域。

They also cast the spotlight on an evolutionary idea that dates from the 1890s, but which has hitherto proved hard to demonstrate.

一个有关进化的观点也因他们而受到关注，这个观点可以追溯到19世纪90年代，但直到今天都还很难被证明。

There is a wrinkle.

他们想到了一个好主意。

The team's robots, which they dub "unimals", are not things of metal and plastic.

这支团队的机器人——取名“unimal”（通用动物）——不是用金属和塑料制成的。

Rather, they are software entities that interact with a virtual environment in the way that metal-and-plastic devices might interact with a real one.

机器人都是软件实体，可以与虚拟环境交互，就像用金属和塑料制成的设备可以与真实环境交互一样。

Unimals are pretty simple, having spheres for heads and cylinders for limbs.

unimal的模样非常简单，头是球体，胳膊或腿是圆柱体。

The environments through which they roamed were also simple, and came in three varieties: flat arenas, arenas filled with hills, steps and rubble, and ones that had the complexities of the second sort, but with added props like cubes that needed to be moved around.

它们漫游其中的环境也很简单，分为三种：第一种很平坦，第二种布满小山丘、台阶和瓦砾；第三种和第二种一样复杂，但还增加了像立方体这样需要被移动的道具。

To begin with, the unimals were given a variety of randomly assigned shapes, but with identical software running each of them.

首先，研究人员给unimal提供了各种随机分配的形状，但每个形状都有相同的软件运行。

That software was a piece of artificial intelligence called a deep evolutionary reinforcement learning algorithm, or derl.

该软件是一款人工智能软件，叫做深度进化强化学习算法，简称derl。

Newly created unimals started in a virtual boot camp, in which the derl learned enough about the world to face the challenges to come.

新创建的unimal始于一个虚拟的新兵训练营，derl可以对世界有足够多的了解以应对即将到来的挑战。

They were then entered into tournaments.

然后开始进行锦标赛。

In groups of four, Dr Gupta put them through tests of agility, stability and ability to manipulate objects.

古普塔博士将他们分成四人一组，让他们接受敏捷性、稳定性和操纵物体能力的测试。

Each group's winner was allowed to "breed" by spawning a daughter with one mutation (an extra limb for stability, perhaps, or extra rotation in a joint, for flexibility).

每组的优胜者都被允许“繁殖”，产生一个带有突变的女儿（可能是为了稳定，或者是为了关节的额外旋转，以提高灵活性）。

This daughter was substituted for the oldest unimal in the pool, assigned to a new group of four, and the process repeated.

这个女儿取代了其中年龄最大的一个，被分配到一个新的四人组，这个过程不断重复。

Unimals were withdrawn from the fray after ten generations of evolution, and Dr Gupta reckons about 4,000 varieties of them underwent training.

经过十代的进化后结束比赛，古普塔博士估计它们中大约有4000个变种接受了训练。

The team were surprised by the diversity of shapes that evolved.

研究小组对进化过程中形状的多样性感到惊讶。

Some had arms as well as legs.

有些既有胳膊又有腿。

Others had only legs.

其他的只有腿。

There were bipeds, tripeds and quadrupeds.

有两足、三足和四足动物。

Some moved like lizards.

有些像蜥蜴一样移动。

Others resembled an octopus walking on land.

有些像在陆地上行走的章鱼。

Crucially, though, the researchers found that the most successful unimals learned tasks in half the time that their oldest ancestors had taken, and that those which evolved in the toughest arenas were the most successful of all.

然而，至关重要的是，研究人员发现，最成功的unimal学习任务的时间是它们最古老的祖先所用时间的一半，而那些在最艰苦的环境中进化出来的unimal是最成功的。

In this evolution of unimals' morphology to promote the ability to learn, Dr Gupta sees a version of something called the Baldwin effect.

古普塔博士看到了鲍德温效应在竞赛中的体现，在unimal的进化过程中促进了学习能力的提高。

In 1896 James Baldwin, an American psychologist, argued that minds evolve to make optimal use of the morphologies of the bodies they find themselves in.

1896年，美国心理学家詹姆斯·鲍德温提出，人的进化是为了最大限度地利用他们所处的身体形态。

What Dr Gupta has shown, though in software rather than in the real, biological world, is that the obverse can also be true—changes in body morphology can optimise the way minds (or, at least, derls) work.

古普塔博士已经证明，尽管是在软件中而不是在真实的生物世界中，相反的情况也可能会发生的——身体形态的改变可以优化大脑（或者至少是derl）的工作方式。

Even though he held the software constant from generation to generation, it became more efficient at learning as the unimals' bodies evolved.

尽管他一代又一代地保持软件不变，但随着unimal的进化，它在学习方面也更加高效。

Whether that discovery can be turned to account in the way robots are developed remains to be seen.

这一发现是否可以用于机器人的开发，还有待考究。

But it is certainly, in the jargon beloved of some businessfolk, an out-of-the-box idea.

但用一些商人喜爱的行话来说，这肯定是一个创新的想法。