BitFlips-Unexpected Connections 1
By Jacob Franco
2003, Brussels, Schaerbeek. Election season. Everyone has been waiting for the National Election result. This is the first time the voting’s being done electronically. Besides the usual computer glitches, everything is going fine. Officials of the ministry suddenly detect an issue—they notice that one of the candidates of a communist party, Maria Vindovogel, has an unnaturally high number of votes from a specific polling station. The number of votes she has received is a mathematical impossibility. She has received more votes than the count of the electorate. The officials suspect tampering, fraud or a hardware glitch. They get all the ballots back and recount(magnetic plastic cards), and after the recounting, every contestant has the exact same number of votes, except for the candidates who have the abnormally high number of votes. The difference between her original number of votes and her second count is exactly 4096. Not a random number.
As far as IT goes, not a random number at all.
There are no coincidences. There are only clues—clues to a cosmic culprit. To explain what exactly happened in Brussels that election season, we’ll need to go back to when the polling was opened.
The morning of the election, the polling opened and apparently, someone hastily slid their voting card in and registered the first vote for Maria Vindovogel. At that exact moment, somewhere deep inside the computer, a binary changed its form, a zero turned to one—an anomaly which translated itself across the vote count and the computer’s system. Computers count using powers of 2. The binary system works the same way as the decimal, which is a system we’re all used to; however, in this case, the base is 2. As a result, every digit has a higher positional value than the digit to its right. In other words, the first place is 2^0 the one to its left is 2^1 and so on. And the value of the digit is only taken into consideration if the digit is one; and if it’s a zero in its place, that specific power of two isn’t counted.
Coming back to the voting error: 4096 is a power of 2.
Our culprit number is the 13th bit of memory in a computer. Which means, at some point during the voting process, some alien force manipulated the signal and flipped what should have been a 0 into a 1. This was a binary flip. The logical next step was figuring out what manipulated that bit. The first suspect was a possible software glitch, but that search was in vain. Next, they suspected a hardware issue; they got the actual machine used in the voting process checked and tested. Nothing wrong with that either.
Zoey Genou, was an MP in Brussels, half the age of the average age of Brussels’ parliamentarians. She argued for the motion that people had to be able to trust the electoral process and the legitimacy of that election. Maligned and shunned, she was determined to find an explanation for the anomaly. She approached David Gload (member of a group on ethics of electronic voting), sent him a report on the suspected binary flip during the election in question. His initial search led him to blame software bugs. Upon searching further, he claimed that a “cosmic ray could flip a bit of memory”. What he meant was, that if a cosmic ray hit the computer at the right place and at the right time, it could flip a 0 to 1. He came across an IBM paper from the 70s, talking about random bit flipping and issues with computer memory. NASA had noticed these issues with bit flips in some of its satellites. IBM had conducted a similar experiment by testing computers at higher altitudes and found out that these bit flips occurred more frequently there. The only explanation that they could come up with, was that cosmic rays, i.e., charged particles, can interact with the electric circuits of computers. Now, the question in the Belgian Parliament changed: if the bit flip caused an error of a much smaller size, would we be able to notice? How often do bit flips actually occur? Pretty frequently, actually. Internet routers get shut off by bit flips. Google’s indexing systems were down the same way for 5 months.
Custom manufacturing plants had their designs altered, and their consequent products were largely anomalies for many production cycles.
Airplanes are continuously bombarded by these space particles. A Quantas AirBus in 2008 nosedived for about a thousand feet post a bit flip in the autopilot system. Bit flips occur often, but the probability of it being of any consequence is very low.
Our story now shifts to Barbara Walton, a resident of Boiling Springs North Carolina. She lives with her husband Barbara is going to South Carolina with her husband to visit their son (4.5 hour drive). The passage is rather rural, and they are travelling in a 2008 Prius. Her husband is driving.
Everything is fine for the first three and a half hours. Then Barbara notices that her husband is speeding. Speeding abnormally. Alarmed, she asks him what’s wrong. He says, “cant slow down, brakes aren’t working”.
They’re speeding at 65 MPH. They’re on a 2 lane highway, through a huge wetland area, moslty swamp. Whenever they try to apply brakes, a signal malfunctions, and as soon as he takes his foot off, the car gets back to speeding. This goes on for over 15 minutes. Finally, he applies the brakes so hard that it wears out his rotor and the car finally stops. It turned out that Toyota drivers all over the world were facing the same issue: multiple incidents of cars speeding even though the brakes were being applied.It was around this same time that cars started being completely computerized. Earlier, you could pop the hood and check what’s happening. Now, you generally cannot, and not just in the engine. You are inundated by car computers all around. When you try to unlock your vehicle, you press the key; immediately, a message is sent from the computer inside the car key to the computer in the car telling the computer at the locks to open the locks (a minimum 3 computers in conversation are required for this menial task). However, that is not all. When you press your foot down on the gas pedal, it’s now an electrical signal for the engine. The reason for this shift was better fuel economy and on the whole a better driving experience, but now you just entrust your life to software. As this shift in technology wasn’t common knowledge, when the whole news broke, the companies just blamed the drivers as most of the victims were on the older side. These incidents started floating in quite frequently until a similar 911 call was received from the driver of a Lexus Toyota. Only this time, the casualty count was up to four. The driver was a 45 year old, California Highway patrol officer. His specialty: vehicle inspection. This completely broke the driver error narrative. This was when the inquiries and the recalls started. Toyota settled a 1.6 billion dollar lawsuit, without admitting any blame. The death count was 89, injury cases 500.
When the experts were given full access to the software and the hardware of the malfunctioning vehicles, multiple issues were found. The key to all of these were bit flips. If you had a bit flip that tasked the controls with the throttle position, it could cause it to stick in some open position. This was reproduced in a Toyota Car in multiple experiments. The jury found Toyota liable for the deaths, but the only theory presented was this: computers could kill people.
Toyota quickly settled all of these cases. This also led to the freedom of a certain Minnesota man, Koua Fong Lee, who already spent 2.5 years in prison and had been charged with vehicular homicide of 3 people. He was driving his Toyota and had faced a similar malfunction.
The problem was fixed by using multiple copies of the same circuits which copied the outputs where at least 2 circuits had to agree to take an action. The same solution was used in Schaerbeek.
Sometimes, these doubling and tripling of circuits turn out to be too expensive. However, as technology progresses and the chips get smaller and smaller, bit flips are becoming more and more common, as the energy threshold for a bit flip is now considerably lower. A cursory temporal analysis of the problem will shed light on the fact that the in 70s, one bit was represented by one million electrons; today, however it takes only a thousand. Does this mean that as technology progresses, we find ourselves increasingly vulnerable to the dangers of this cosmic kind?