10【原声】Chapter 3 | AI Helps us Reach Our Individual Potential as Learner

Part 1: AI is not Robocop

00:00/14:05 

Hello listeners of Himalaya, and welcome to this episode in our series “AI and Us”. In the past episodes we have looked at how AI works; how machines can and actually do learn from data. In this episode and the following episodes, we take a closer look at what AI can do for us; today we talk about the role of AI for decision-making. So let’s delve right in.

For some people, AI is all about the rise of the robots. Inspired by movies from “Terminator” to “Her”, they envision future as one in which intelligent robots run the world and either are evil and wreak havoc or help hapless humans to survive or are a combination of both. It’s a world full of Robocops and autonomously flying robotaxis. 

Some authors have even suggested that our future will be one of “fully automated luxury communism”, in which all wishes of all humans will be automatically fulfilled by an army of robots that serve us, read our minds, and efficiently take action to bring us the latest Gucci shoes and Louis Vuitton handbags. And, no, they are not joking – they are serious. [Sigh]

Obviously, such dystopias and utopias entertain our minds. It’s fun at times to dream a bit; but sooner or later we have to confront reality – a reality that’s very different, a reality that despite all the rosy or dark prognostication about humanity’s loss of power still requires our decisions, many times every day.

AI’s biggest impact therefore isn’t robots taking over. The real focus of AI is to help humans make better decisions. And it’s already happening, just think about it. When we shop online (or even offline), our smartphone apps make recommendations that many of us are intrigued by. It is said that up to one out of three transactions on Amazon are directly based on a recommendation the machine made. When we commute to work, we use apps to help us find the best way, route us around problem spots, such as a broken-down subway or a congestion on the road. And when we listen to music or watch a video, AI-based suggestions influence what we experience.

The interesting and important point here isn’t that this is already happening, although it’s easy to forget about it.  The point is that we have no problems in our daily lives to use the power of data-driven AI to improve our decisions. Because these services are useful, they provide value, and improve our decisions. To put it bluntly: AI is not about replacing humans with robots, but humans using a technology to make better decisions. AI is about human empowerment, not robot domination.

Part 2: Human Decision Weaknesses

03:19/14:05 

As I explained in an earlier episode, we humans face numerous decisions every single day – small ones and big ones; ones that are consequential and ones that are banal. We cannot escape this predicament – as we humans shape our world, we have to decide. Some people may say that there is an alternative – and that is indecision, to not decide. But even choosing not to decide is of a decision. 

And it is a decision that has consequences. If a young man does not message a girl he has met and ask her on a date, they may never fall in love and begin a relationship. If one does not apply for a job, one will never be selected. In practice one can choose not to decide, but the decision does not magically go away, it will be decided – by somebody else, by circumstances and coincidences, and by defaults. We cannot escape the fact that decisions are to be made, we can only decide who makes them, and on what basis.

Most of us, therefore, have accepted that we face decisions that we have to take. But on what basis should we take our decisions?

In an earlier episode, I explained how when we face decisions, we are tempted to let our emotions take over and rely on our gut feeling, on what often is rosily described as intuition, but often are just unsubstantiated beliefs. Or that we can rely on fact and data. And I suggested that relying on facts, on data is a far better strategy than relying just on beliefs and feelings for most of the decisions we face.

But even relying on facts and data is often challenging for us humans. As research has shown humans aren’t very good at processing and analysing data. It begins we the fact that many of us have difficulties keeping multiple factors in mind. Just consider something as simple as shopping at the fruits market and having to choose between, say, pears, apples, and bananas when for each of these fruits we have collected three pieces of fact, say the price, the ripeness, and the taste. 

Logically, this translates into a 3 by 3 matrix with the three fruit categories making the columns, and the three qualities (price, ripeness, and taste) the rows. When humans are faced with such a 3x3 matrix of facts and have to decide which fruit to pick, our brain is already overwhelmed. So, in practice we never compare three fruits and its qualities like that when we shop. We break our decisions down to simpler ones, like just comparing price or picking whatever fruit looks tastier. But at the heart of it lies a cognitive shortcoming that we humans suffer from.

Of course, we have found ways to deal with it. But over the past decades, psychologists and behavioural economists have discovered a number of additional mental distortions, or biases, that quite seriously impede human decision making. These biases are pervasive in human behaviour, even if you are not aware of it.

Take for instance, the so-called confirmation bias. It means that humans notice and rely more on facts that confirm their views than those that contradict them. We actively look for confirming information, we favour using it in our analysis, and we even remember facts that confirm our views more likely that facts that don’t. Sometimes it happens to me, too; when I drive with my car and I have to stop at a red light, and then a bit later at another one, I believe that I encounter only red lights that day, because my mind simply forgot the times I encountered a green light and could speed right through. 

It looks like, we all like to be right and hate to be wrong. But a confirmation bias is not only bad, it also has advantages as it eases our cognitive workload. Every day we are inundated with new data and new facts. We can’t study all, so we need filters for what is relevant. And evolution opted for filters that confirm views, because such confirmation is easier to process. It does not lead to cognitive dissonance. But of course, confirmation bias makes us vulnerable to make erroneous decisions.

Importantly, the more successful our decisions have been, the more susceptible we may become to confirmation bias. If a military general has won multiple battles with the same strategy, he may think it’s the right one for all circumstances and just stick to it. But that will not work when the circumstances change, and he will lose the battle. Similarly, very successful managers are susceptible to believing they are always right. They may stop critically questioning their own views, and thus fail to decide well when the conditions have shifted.

Or take the endowment effect: We humans are biased towards keeping an object we already have compared with buying the same object. It’s hard for us to throw something away, even if it is old and we haven’t used it for years – but we would not purchase the same thing if offered, not even for very little money. And I am not referring to an important photo, or a memorable letter, but something as banal as a washed-out t-shirt, a DVD of a bad movie we haven’t watched in ages, or a book we found so boring we stopped reading it. 

The endowment effect, too, can cloud our decision making. Often it may be far better to sell or get rid of stuff that we no longer need and that clutters our lives, but we hold back – and thereby invest, materially and cognitively, in something that may have no real value for us.

There are many further such biases. Like the hindsight bias that lets us see patterns and make sense of things in hindsight: as we say, afterwards we always have it right. Or the anchoring bias, that let’s us compare things, especially prices, based on the first number we hear, the so-called anchor. That’s why many merchants write on the price tags a higher price that is crossed out to indicate that the current lower price is a special offer. We all love to get special offers – even if they are not special at all.

There are many more such biases. I am just providing a few examples here for you to get a sense of them. And all these biases impact human decision making even when we do take data and facts into account, because they influence what data we gather, as well as how we analyse and interpret it.

Part 3: AI to the Rescue

11:13/14:05 

And this is precisely where AI can help us. We can set up AI systems so that the collection of data is less biased, because we can think hard about how to do it right and because we can think about it in the abstract, rather than in the context of a concrete decision in which biases may already influence our behaviour. The machine does not need confirmation of its own views, nor does it gain satisfaction in seeing patterns in hindsight or remembers more recent experiences better than those that happened some time ago (which is another cognitive distortion human thinking is afflicted with).

So if done right, data-driven AI can help us humans to make more fact-based less biased decisions. And that has huge advantageous for all of us.

There is a small caveat however: data-driven AI learns from data. If the biases are already in the data, then the AI system will be biased in such a way, too. That’s not a problem of AI, but of the data it learns from. So suppose a company uses AI to hire IT people, and in the past have mostly hired men. Then if the AI system is trained with past data, it may be biased towards hiring men rather than hiring the best person for the job. That’s a real problem, but as we discuss in a future episode, there are ways to mitigate such problems.

So let’s sum up: Decision making is a huge and important part of human activity. Relying on facts when deciding is often better than following one’s guts. But even fact-based human decision making can suffer from cognitive errors in the form of biases. That’s when AI shines, because it does not suffer from these problems in a similar vein. And its remaining weaknesses – namely that it is trained with data, which itself could be biased – can be addressed. In sum, for all these reasons embracing AI to improve human decision making is exactly the right thing to do.  

Thank you for listening. I hope you enjoyed this episode, and I look forward to seeing you around for the next one, when I will talk about how AI changes mobility, from cars to planes to urban transport and beyond.