The Exquisite Dullness of 8068

Whenever we need to choose a four-digit combination for a luggage lock, a key safe, or a hotel safe, most of us try to come up with something that would be as difficult to crack as possible.

We approach this seemingly trivial task in surprisingly different ways. Some of us try to make the combination look as random as possible, avoiding obvious patterns or sequences: 2971, say, rather than 9876, 2864, or even 4981. Others deliberately choose a number beginning with 8 or 9, reasoning that a would-be thief might try combinations sequentially, starting at 0000 and working their way towards 9999. Why not put our secret somewhere near the far end of that journey?

And then there are those who don’t bother with any of this and simply choose a number they can remember, whether it’s small, round, peculiar, or entirely unremarkable.

But does it really matter? Is there any point in taking the choice so seriously? Is 9876 inherently more secure than 6789?

From a purely mathematical perspective, the answer is no. If four-digit combinations were chosen completely at random, every possible number would have exactly the same probability of being selected. Whether it’s 0000, 1234, 2971, 3355, or 9998, the probability would be one in 10,000, or 0.01%.

There’s a small problem with this reasoning, however: people aren’t particularly good at being random.

Both parties involved in our little security contest – the person choosing the combination and the person trying to guess it – are human beings. The former tries to invent something that would be difficult to predict, while the latter attempts to predict exactly what the former might have invented.

And human beings have preferences. We like patterns, repetitions, familiar dates, and memorable sequences. A person choosing a PIN may favour 1111 or 1234 simply because these are easy to remember. An attacker, knowing this, would be wise to try such combinations first rather than blindly working through all 10,000 possibilities.

In other words, while all four-digit numbers are mathematically equal, they are not equally popular. And this makes some combinations much easier to guess than others.

Of course, if the attacker has unlimited time and the lock doesn’t impose any restrictions on failed attempts, even an unusual four-digit combination isn’t particularly formidable. There are only 10,000 possibilities, after all.

I once had the dubious pleasure of testing this principle myself after forgetting the combination to my own key safe. It took me about forty minutes of trying different numbers before the little thing finally surrendered. Unfortunately, it happened on a chilly December evening, and I couldn’t feel my fingers for another two hours.

But I digress.

Back in 2012, the people at DataGenetics decided to investigate the matter scientifically. They analysed a collection of 3.4 million leaked four-digit passwords and ranked the combinations by popularity.

The results were hardly surprising at the top of the list. Predictable combinations such as 1234, 1111, and 1122 were among the favourites.

But what about the other end of the table? What about the combinations that practically nobody wanted?

The winner – or rather, the loser – was 8068.

Out of 3.4 million entries, this humble number appeared just 25 times. The closest runners-up in the contest for obscurity were 8093 and 9629.

There was another curious detail. Of the twenty least popular combinations, eleven began with 8 or 9, while only three began with a digit smaller than 6. This hardly supports the theory that people commonly choose higher numbers to make sequential brute-force attacks more laborious.

But let’s return to our champion.

What makes 8068 so special?

Well, absolutely nothing.

And that, I think, is precisely what makes it interesting.

There is nothing particularly appealing about 8068. No obvious sequence, no satisfying mathematical relationship, no memorable association that immediately springs to mind. It doesn’t look clever, elegant, or amusing. It’s just four digits sitting together, minding their own business.

It’s a number so thoroughly unremarkable that our minds have little reason to settle on it.

In a peculiar way, 8068 might be one of the most random-looking numbers from a human perspective. Not because it’s mathematically any more random than its neighbours, but because it lacks the qualities that usually tempt us to choose one number over another.

It’s the forgettable face in a crowd: perfectly ordinary, practically invisible, and all the more remarkable for it.

And here’s where things get particularly amusing.

By identifying 8068 as the least popular four-digit PIN, we’ve inadvertently given it the very quality it was missing.

We’ve made it special.

Once a number becomes famous for being obscure, it is no longer quite so obscure. People who would never have considered 8068 before might now choose it precisely because of its reputation. Its extraordinary dullness becomes an attraction. And the more people learn about it, the more likely it is to lose the distinction that made it famous in the first place.

In other words, by celebrating its glorious anonymity, we have begun to destroy it.

There’s something rather beautiful about that paradox. A number becomes interesting because nobody finds it interesting, and the moment we acknowledge this, the whole thing starts to unravel.

So, the next time you’re setting the combination on your luggage lock, by all means choose something unusual.

Just perhaps not 8068.

We’ve already ruined that one.

If you’re curious about the other mathematical, statistical, and psychological oddities of four-digit PINs, I wholeheartedly recommend the original DataGenetics article. Among its more startling findings is that just 61 guesses would be enough to cover roughly a third of the passwords in their dataset. So much for our ingenuity!

On Cause-and-Effect Trees

Let’s get back to the Herald ferry accident for a moment. Its investigation was a remarkably curious – and quite unusual at the time – example of an analysis technique that sometimes goes under the name of the cause-and-effect tree.

The cause-and-effect tree is an iterative reasoning exercise intended to explore the cause-and-effect relationships behind a problem. Its primary purpose is to help the investigator resist the temptation to stop at the first plausible explanation and avoid jumping to superficial conclusions by forcing him or her to keep asking the same question (the ‘why’) over and over again, even if the answer has seemingly been found.

Cause-and-effect trees belong to a wider family of causal-analysis tools, ranging from the simple Five Whys game often attributed to Sakichi Toyoda, to highly formalized Fault Tree Analysis. The precise technique matters less here than the habit of thinking they encourage: resisting the temptation to stop, and looking one step further, one chapter earlier, one level deeper. The tree structure also helps us realise that a specific failure may have more than one cause. The captain, for example, remained unaware that the ship was leaving with its bow doors open both because there was no warning light and because the company relied on negative reporting. You begin building your tree by placing the problem at its root, and start growing it upwards by asking ‘why’ at every stop. Every ‘because’ adds another branch at your current stop. Once you have added branches for all the plausible “becauses” you can identify, you move upwards and ask ‘why?’ of each one in turn. The more answers you get for a why, the wider your tree becomes. The more ‘why’s you ask in any one subtree, the taller it grows.

Let’s consider the Herald accident. Leaving aside broader aspects of the vessel’s design, we can end up with the following tree (click to enlarge):

A cause-and-effect tree of the Herald accident investigation

It is very helpful to build this diagram collaboratively, where one subgroup of the team provides their ‘because’ while the other provides counterarguments: “Why didn’t the assistant bosun close the doors? Because he slept through! But he wasn’t the first person to do so. Doesn’t that suggest there may be something else worth investigating?” Team members taking sides can make the exercise considerably more productive as it facilitates the search for less obvious causal explanations. It also helps keep the blame game under control and reminds everyone that humans are, well, humans.

Looking at our simplified tree, we can conclude that many of the deepest actionable causes lead back to management decisions: the company’s failure to create a safe working environment through clear, workable procedures and appropriate safeguards, and its failure to establish a safe and productive work culture.

Note that the tree doesn’t stop on “human trait” or “human weakness” causes. Fatigue, forgetfulness, confirmation bias, complacency and distraction are all part of what makes us human. We cannot eliminate those qualities from people. They are not moral failings in themselves, nor useful places to stop an investigation. They are predictable characteristics of human beings – and therefore risks that a well-designed system should anticipate and mitigate.

Only when we start looking at the problem from that angle can we do what we ultimately want to do – make our systems and our world safer. Humans get tired. Humans make mistakes. Humans underestimate their role in the overall safety of the system. Replace one bosun without changing anything else, and another may make exactly the same mistake – as had already happened on another company ferry. Add redundancy, positive confirmation and independent checks, and you remove important routes by which that mistake can turn into a disaster.

There’s another curious thing about that tree. Nearly every leaf that is not a human trait involves a human decision – but those humans occupy very different positions in the system. This brings us back to the earlier point: the accident was undoubtedly a product of human error. And yet saying that tells us surprisingly little until we ask which human, making what decision, under what conditions, and how far upstream.

And one more thing.

It is vitally important to resist the urge to be dismissive of people when analysing causes. Kindness and curiosity are unexpectedly close relatives. Try your best to be kind. Try to be understanding. Try to postpone blame until you understand what happened. Accountability can come afterwards; explanation has to come first.

Most mistakes are not acts of malice. People make poor decisions, overlook things, and sometimes break rules, but there is often context that helps explain why their actions made sense to them at the time.

You may eventually conclude that the person genuinely behaved negligently. But you arrive there after investigation rather than before it. The greatest value of causal-analysis tools – whether Five Whys, fishbone diagrams or formal fault-tree analysis – is that they cultivate the habit of asking one more question after we think we already know the answer.

Sometimes the next “why?” finds a broken procedure. Sometimes it finds a bad decision. And sometimes it merely finds a tired human being. But it is usually worth asking.