The incumbent industry deals with this problem by ignoring it. They pick one possible answer and act like the other ones don't exist. And as a result, we design suboptimal mines, make suboptimal decisions, often mining unnecessary material.现有的矿业行业处理这个问题的方式是忽视它。他们只选择一个可能的答案，然后假装其他可能性不存在。结果就是，我们设计出的矿山并不理想，做出的决策也并不优化，经常还会开采大量不必要的物料。

We've invented a different way. We collect all the possibilities consistent with the data measured, and we do this by simulating the physical response of each of the arrangement of rocks. We do this 10,000 times faster by training an AI to learn the relevant physics of the rock beneath, in the time it takes the conventional method to test one. That means we collect better data, we make better predictions of where to look next. So if you had a rock body and a rock body that's denser than material around it, you might drill through the middle of it. But if you have all the hundreds of thousands of possible solutions, the best thing you can do is to collect data where you're the most uncertain and rigorously eliminate as many possibilities as possible. This enables us to maximize the information we get for every dollar we spend, and we do this repeatedly so we can quantify our uncertainties.我们发明了一种不同的方法。我们会收集所有与测量数据相符的可能性，并通过模拟各种岩石组合的物理反应来实现这一点。借助人工智能学习地下岩石相关物理特性，我们的速度比传统方法快上 10,000 倍——在传统方法只能测试一个的时间里，我们能完成成千上万次模拟。这意味着我们能收集到更好的数据，进而对下一步的勘探地点做出更好的预测。比如，如果你发现一个岩体，其密度大于周围的物质，你可能会选择直接在它的中间钻探。但如果你手上有成千上万种可能的解决方案，最明智的做法就是在最不确定的地方收集数据，并尽可能严格地排除掉不可能的情况。这让我们能够最大化每一美元投入所获得的信息，并且我们会不断重复这一过程，从而量化我们的不确定性。

Even after we've made an ore body discovery, we still have to contend with this uncertainty. We have to define the size and shape of this ore body. Let me illustrate how difficult this is. So now, 1,000 meters below your feet, you drilled, you sampled the rock and you determined that it has five percent copper. So now you know, you've got your data point and your observation. Now, I ask you to make a prediction of the concentration of copper of the person sitting next to you.即便我们发现了一个矿体，仍然需要面对这种不确定性。我们必须界定这个矿体的大小和形状。让我来说明这有多困难。假设现在你在脚下 1000 米处钻探，取出了岩石样本，并测定其铜含量为 5%。到这里，你得到了一条数据点和一个观测结果。接下来，我让你预测一下：坐在你旁边的人脚下 1000 米处的铜含量是多少？

What would your prediction be and how confident would you be in your prediction? What about across the room? Think of any person across this room and try to predict 1,000 meters below them. What about in the next building or the next city? This is the vast challenge that we face. We've only sampled a tiny fraction of rock, collected several football fields apart from each other, for which we're trying to make predictions of all the rock properties in between.你的预测会是什么？你对这个预测有多少信心？那么房间另一头呢？想象一下房间那头的某个人，试着预测他脚下 1000 米处的铜含量。那隔壁大楼呢？或者下一座城市呢？这就是我们面临的巨大挑战。我们只采集了极少量的岩石样本，而且这些样本之间相隔相当于几个足球场的距离，却要用这些数据去预测其间所有岩石的属性。

This technology has helped us move fast in Zambia, where I come from, to design and develop a mine based on our predictions for which we've only sampled a tiny fraction of rock.这种技术已经帮助我们在我来自的赞比亚快速推进，仅凭极少量的岩石样本和我们的预测，就能够设计并开发出一座矿山。