05【原声】Chapter 2 | Why it Matters to Know

Part 1:  Decisions, decisions everywhere

00:00/17:14 

Hello listeners of Himalaya and welcome back. In this episode we talk about why AI truly matters, and it may surprise you. As I will explain, AI isn’t so much about robots ruling the world than how we humans can make better decisions.

Every day every human faces decisions: small and large, inconsequential and monumental, important or banal. And humans have different ways to decide. They can follow their beliefs, their gut and their instincts for example. 

That may give them a deeply satisfactory feeling to have decided well, but countless scientific studies have shown that simply following your beliefs will often lead one astray and prompt a bad decision. You may think you did the right thing because in your gut it felt right, but looking at it rationally, you may have made a mistake. It’s human, we all make such mistakes.

But for many centuries, we humans also had an alternative way to make decisions: by collecting and assessing the facts, by gathering and analysing the data. Such rational, empirical and fact-based decision-making produces on average better, more appropriate decisions than relying on your gut. Not just scientists and researchers, but enlightened decision-makers in all walks of life have improved their lives and produced better outcomes relying on such a more rational, fact-based approach.

It isn’t one, however, that always comes easily. For instance, in the recent virus pandemic, US president Trump advocated using a malaria drug. He said that he felt in his guts that he was right. Later, it turned out that there was no evidence whatsoever to back up his claim. The World Health Organization even stopped all trials. 

And in one study of European managers, almost seventy percent of them said that they trust their guts more than facts, even though they acknowledged the importance of data. The result are suboptimal decisions at all levels that lead not only to economic inefficiency, but also to societal and individual suffering at a staggering level. Humanity can and must do better than that. And Big Data and AI are the tools that help us make progress.

For years, the Swedish nd bestselling author Hans Rosling would address a wider variety of audiences, from school kids to CEOs of international companies and even Nobel laureates. He would often begin his speech by asking his audience a series of simple questions and offer for each of them three possible answers. One such question, for example, was how over the past 20 years the number of humans in the world that lived in absolute poverty had changed. 

Had the number doubled, stayed constant, or halved, Rosling asked. Most of his audience responded saying that the number of people in poverty had doubled. But the correct answer, as Rosling explained, was that the number of poor people in the world over the past two decades had shrunk by half, reflecting an amazing achievement and a true sign of progress for humanity. The world is actually better than we think.

But that alone was not what Rosling was after. He wanted to make the point that without knowing the facts and having the data, we tend to make false assumptions about the world, and based on these false assumptions make erroneous decisions. Just consider this case: Not knowing that poverty has been reduced （and where） will make us spend more on alleviation of absolute poverty, but neglect perhaps vaccination, education or economic development. Getting the facts right is such a superior way of deciding!

A hundred years ago, Marie Curie was the first woman to win a Nobel prize in science. She ended up winning it twice in her lifetime, once in physics and once in chemistry. She was a giant, and she took facts seriously. She said: “Nothing in life is to be feared, it is only to be understood. Now is the time to understand more, so that we may fear less.”

This was true then, a century ago, but it is certainly true now that we have such powerful tools to help us decide well with data.

Part 2: Collecting the Data

05:38//17:14 

So, we know that we face a lot of decisions. We can rely on our guts and often decide badly, or we can rely on data and AI and improve our decisions. But for it to work, two conditions need to be in place. The first is that we need to collect sufficient data, and the second is that we actually use the data to drive insights; and we can raise insight generation to a new level by employing AI technologies, such as machine learning and deep neural networks.

So, please come along on this journey, and let’s take a deeper and more detailed look at both of these conditions.

Collecting data is the obvious condition to using data for decision making. But collecting data is surprisingly challenging. Before the digital age and before computers became common, data was collected the analog way. If we wanted to measure the length of a building, for example, we had to have a yardstick or a measuring tape. 

And then we needed to apply these tools the right way to get a reading that was sufficiently accurate. We needed to actually note down and record what we had measured, and to add the unit in which we had made our measurement. Perhaps we even wanted to note down additional information, such as the date and time of our measurement, and who actually took the measurement. All of this requires diligence and takes effort. Lots of effort.

And so, for centuries, we humans have been reluctantly gathering data, cognizant of the amount of work it entails. In fact, we routinely only collected just as much data as we thought was absolutely necessary. And even that was too much for some.

In the early 17th century, the mathematician and physicist Johannes Kepler worked as an assistant to the astronomer Tycho Brahe. Kepler hated gathering data, but Brahe had painstakingly collected lots of data about the movement of the planets. Kepler saw his chance and actually stole the data from Brahe to analyse it and come up with the laws of planetary motion, the basis for anything from astronomy to rocket science.

Stealing data of course is not right, but it also shows how onerous and difficult data collection was in these days. Even seemingly respectable people were willing to break the rules for a shortcut to getting data.

That is very different today. Sensors in the many digital devices that we use every day capture a huge amount of information with very low effort on our part, and at low cost. Your typical smartphone has more fifteen sensors that measure and record data routinely time. A typical car in 2020 has over a hundred sensors built into it, with high end cars containing 250 sensors and more. 

When we multiply these with the number of smartphones and cars that are in use, you quickly end up with tens of billions of sensors. And that is just for smartphones and cars. But sensors are everywhere – from electrical toothbrushes to commercial aircraft. And they all collect data, without much fuss, seemingly seamlessly, and at an average level of quality that surely exceeds many of the measurements done by humans in earlier days.

Moreover, tech companies around the world have invested many hundreds of millions of dollars each year to make measurements about something in one unit translatable and thus comparable into measurements of another unit. It’s easy for feet into meters, and Fahrenheit into Centigrade, but it can get quite complex quickly, for instance to translate Watts of an old incandescent light bulb into something equivalent for new LED lights.

The shift in ease and cost of collecting data has prompted a shift in data collecting behaviour. In the old days, at best we collected just as much data as we needed. But today, we collect all the data we can, because it is so easy and cheap, even if we do not yet know how to use it.

Part 3: The Calculation Machines

10:40/17:14 

So data is being collected. But there is a second condition that needs to be in place, for us to make data-driven decisions. And that is the task to analyse and use all the data we have collected to inform our decision making.

That, too, has been very challenging in the old days. Mathematics in general, and statistics in particular were not that advanced in previous centuries. We lacked important methods of managing and examining data. But most importantly, we lacked the tools to store and process data with ease and at low cost.

Just consider: In the 19th century, a US naval officer named Matthew Maury discovered a huge trunk load of ship’s logbooks in the basement of the Navy building he worked at. The logbooks contained a treasure of data about winds and waves, and Maury realized that if he could analyse the data, he could help sail ships to identify prevailing winds and thus faster paths across the Oceans. He convinced the US Navy to invest in the project, but it took a hundred human calculators, tabulating numbers day in, day out for ten years before he had the valuable results in hand. That’s a very long time.

And Maury’s example isn’t the only one. Every decade since its earliest days, the US surveys every person living in the United States as part of the census. The data is used for long term planning and much look for by public authorities. But in the late 19th century, the amount of data gathered had exceeded the human resources available to examine it. It took a full seven years for the census data to be tabulated. By that time, the data was already hopelessly outdated, and a new census was being prepared.

The challenge to calculate the data of the US census gave rise to a peculiar new tool: the tabulating machine. Its inventor provided the US authorities with a powerful tool that calculated census data in mere weeks rather than years. About half a century and many innovations later the tool had morphed into the general-purpose computer. And from these early beginnings, computers exceeded human capacity of tabulating numbers.

To put it bluntly: In the past, we not only collected but also analysed as little data as we could, because it was so costly and difficult. Only with computers, there can be Big Data and more generally substantial progress in fact-based and data-driven decision making.

Part 4: Going Beyond Calculations – the Birth of AI

13:55/17:14 

But there is a final and most crucial aspect. As computer technology progressed, more basic mathematical operations could be completed per second, and more instructions could be executed. In the 1940s, the first big digital computer, ENIAC could complete 5,000 additions in a second. Today, a single processing chip can exceed a hundred billion additions a second. That’s an astonishing achievement. But at its core, it’s an increase in speed and scale. 

Fundamentally, even today’s computers follow the instructions that we provide. And so, as they analyse the gigantic piles of data that we are collecting, these powerful computers are still only looking for the things that we told them to examine. The enormous power of our digital devices is constrained by our human ability to order our devices to march in the right direction. What if we examine data by looking at it the wrong way? 

What if the data that we have collected holds the key to amazing new insights into how the world works – how we can improve our lives – but we fail to find the way to unlock it, not because our machines can’t calculate the data, but because we humans fail to see the patterns in the data that matter, because we command the machines to do the wrong thing? What if, for example, a marketing department of a cake manufacturer analyses data about what flavour people prefer, when in reality the most important question is the right size of the cake? Then the data will offer the correct answer to the wrong question.

This, dear listeners, is where Big Data meets AI, where our machines begin to go beyond doing exactly and predictably down to every instruction, every addition what we have told them to do, where we have told them to look. With AI, machines go beyond us humans in seeing the patterns that are hidden in data – patterns that we would not discover manually, because we would not even know exactly where to look. And that, dear listeners, is what I am going to tell you about in the next episode.

In the meantime, please keep in mind what we talked about today: that data is key to better decision-making; but it requires the ability to collect and analyse data at speed, scale, and low cost. Sensors, computers and networks – the tools of the digital age – have enabled us to do that. But to truly enter the data-driven age of AI, we need to go beyond digital tools. We need to focus on the data, and on how a novel way of thinking can get machines to see patterns we humans can’t.