UK's beloved RNGesus machine ERNIE goes quantum in 5th iteration

For computing history nerds, the names Tommy Flowers and Harry Fensom likely conjure up images of the code-breaking Colossus. But after the war, they also had a hand in creating a dearly loved, much-anthropomorphised, millionaire-making machine: ERNIE. ERNIE – Electronic Random Number Indicator Equipment – is the machine used …

[Anonymous Coward]

They called him Ernie, Ernieeeeeeeeeee

And he drove the fastest milk cart in the west.

(Yes, I know we're talking about a different ERNIE, but I could not resist the reference to the namesake).

[phuzz]

Re: They called him Ernie, Ernieeeeeeeeeee

That song was going round my head the entire way through the article.

[AIBailey]

"In this case, photons are sent towards a semi-transparent mirror and one of two events happens – reflection or transmission – which are associated with 0 and 1 bit values.

Surely that's not entirely random, I'm not sure how feasible it would be to create a mirror that would reflect exactly 50% of photons? Otherwise there will be an overall bias towards a 1 or a 0 over time?

[Hans Neeson-Bumpsadese]

Surely that's not entirely random,

Possibly (probably?) not entirely random, because of the reason you postulate, but likely random *enough*. There will be enough natural randomness in the movement of the photons towards the mirror that it won't be possible the seed or artificially influence where the beam lands in order to get a 1 or a 0.

Thinking on it a bit more , would such a theoretical mirror require such design/engineering that it could actually be determine a way to influence whether a specific photon gave you a 1 or a 0? Hmmm...my brain's starting to hurt now.....

[Anonymous Coward]

No, such a thing isn't possible. A photon interacting with this thing has a perfectly deterministic state (wavefunction) after the interaction in fact: it's just that that wavefunction is non-zero on both sides of the mirror. When the wavefunction is 'collapsed' (old-fashioned terminology) by the photon interacting with some macroscopic object (so, incrementing a counter or something) then the photon 'decides' which side of the mirror it's on. That decision is 'truly random' in the sense that it's not predictable in advance, even in principle, if QM is correct (see below).

Before it collapses the wavefunction may be skewed towards one side of the mirror or the other, which would mean the thing is biased one way or the other: this can be dealt with by measuring & correcting for the bias as other people have said.

'Truly random' is a bit subtle. We know, thanks to Bell, that if it were predictable in advance, then, if QM is correct, we can bargain that into being able to send information into the past. If we can do that then we can predict it anyway, just by sending the answer back in time. We can also, for instance, win all games of chance, like the stockmarket, and causality generally falls apart. There is strong experimental evidence that QM is correct (look up 'tests of Bell inequalities'), and assuming causality isn't violated then we can assume it is not predictable in advance. But this is subtly different than 'truly random': perhaps 'God' has a big list of all the future states of things and everything is determined in advance, just not from information within the system. This is called 'superdeterminism' and it's pretty easy to see it makes no practical difference: there still is no access to any information which allows us, within the system, to predict the state after the wavefunction collapses.

[Anonymous Coward]

> Surely that's not entirely random, I'm not sure how feasible it would be to create a mirror that would reflect exactly 50% of photons? Otherwise there will be an overall bias towards a 1 or a 0 over time?

IDQ's website has a non-technical paper describing this. The mirror is within 10% - i.e. gives results between 45% and 55% and then an unbiasing procedure is applied to produce 0s and 1s at exactly 50% each.

[STOP_FORTH]

Removing bias

I haven't looked at IDQ's site but bias removal is a well-established technique.

Group the output from RNG into pairs of bits. Discard any 'dibit' which is '00' or '11'. Assign the value of '0' to '01' and '1' to '10' (or the other way round).

Of course, if your electronic die is really rubbish, you'll have to discard lots of values. Maybe that's why they need such a fast, powerful computer?

[Anonymous Coward]

Re: Removing bias

There a lot of "quantum loveliness" puffery on the IDQ website but I still can't see why it is better than a noisy Zener diode, or a neon discharge tube for that matter.

[HieronymusBloggs]

Re: Removing bias

"I still can't see why it is better than a noisy Zener diode, or a neon discharge tube"

Both of which could also be described as quantum tech.

[Anonymous Coward]

Re: Removing bias

It's better because it's faster. The spurious 'quantum' thing is just marketing bullshit: both processes rely on QM.

[Giovani Tapini]

Even it it wasn't random...

It remains utterly irreproducible which is pretty much the point. Your setup would have to be a duplicate at a subatomic level and at a point in time to even get close. Good luck with that.

[Anonymous Coward]

Parallel Universe

So does this mean somewhere in the Multiverse, one of my alternates hit the jackpot?

Also this might explain some strange inconsistencies on my ATM accounts.

"why the heck did someone spend exactly £87.92 on parts yet i got nothing"...

[tsf]

Re: Parallel Universe

Money disappearing from your account with nothing to show for it could have something to do with the multiverse, if by multiverse alternative you mean significant other.

[Anonymous Coward]

Re: Parallel Universe

Not sure but one of the mysterious transactions looks very similar to 2* memristor chips + postage.

Yet I don't remember ordering them but did look at the page recently.

Also never got my 3 yellow OLED displays which I did order and no notification from the seller.

Wonder if I have a double somewhere with similar interests?

[Neil Barnes]

I bet they haven't fixed the bug

The one where it keeps failing to print 'thousand' after 'twenty-five' and before 'pounds'.

[FlossyThePig]

Does not compute

...420 million prizes, worth £19.1m, handed out since the scheme was launched in 1957...

Considering there are two £1m prizes each month (boo hoo I haven't won it again) I think the total value of prizes handed out is a little bit bigger.

[Anonymous Coward]

Re: Does not compute

I could believe a current average of £19.1m per week or perhaps a total of £19 billion since 1957, but perhaps that is an overestimate. A rich aunt gave me a one pound premium bond on the day I was born in 1957. I've still got it, and its never won anything

[tsf]

Re: Does not compute

Don't loose hope, with a £1 bond the odds say you should still win something within the next 2000 years or so.

[Hans Neeson-Bumpsadese]

Re: Does not compute

And I believe you can cash Premium Bonds in for face value, so you could still get that £1 back as cold hard cash (but not inflation-adjusted, tant pis)

[Anonymous Coward]

Re: Does not compute

" A rich aunt gave me a one pound premium bond on the day I was born "

#MeToo - I was going to take issue with the statement that ERNIE has never failed by pointing out that it has consistently failed on a monthly basis to select my bond for any prize whatsoever!

[Anonymous Coward]

Re: Does not compute

I guess your aunt didn't know the right handshake to get the "good" number bonds. Of course this new technology isn't the product of patriotic British boffins, so for all we know the numbers could be sent right back to the Gnomes of Zurich by Quantum Teleportation for "checking and recalibration" to make sure the right people get the biggies.

I'll get my tinfoil hat...

[Martin an gof]

Re: Does not compute

A rich aunt...

Exactly the same for my wife (though a slightly later date). Her mother, on the other hand, who up until recently had invested near the maximum allowed seems to keep winning stuff and is close to the promised "average return" I think.

M.

[Uncle Slacky]

Re: Does not compute

Same here - my mother has an early £1 bond (begins with "AB...") that's never won anything.

[DuncanLarge]

Re: Does not compute

"A rich aunt gave me a one pound premium bond on the day I was born in 1957. I've still got it, and its never won anything"

You were supposed to buy more.

[Will Godfrey]

Hang on a minute

Are you sure that's the way they get the numbers? If they just keep generating random numbers till they get a valid one, it is theoretically possible to never get one!

On a much smaller scale, when I wanted to design a bingo number caller I created a list of the numbers 1 to 90, then shuffled them, after which I just pulled them out in order.

P.S.

Moving a number out of a slot then back into it is still a valid shuffle - but you try telling people that!

[Snowy]

Re: Hang on a minute

Was thinking that myself and instead if they built a list of the all the valid numbers (in date order) and picked a random number from 1 to however many there is. This should quicker as no need to check if valid one is pick as everyone would be valid.

[Anonymous Coward]

Re: Hang on a minute

They don't in fact need to be in any particular order at all: if you just have a list of them, in any order, and then pick random entries from that list you're fine. I suspect the problem is knowing which bond numbers are still alive, although if that's the problem then I don't know how they know how to prune the dead ones after the numbers are generated anyway.

[maffski]

Re: Hang on a minute

'On a much smaller scale, when I wanted to design a bingo number caller I created a list of the numbers 1 to 90, then shuffled them, after which I just pulled them out in order.'

And have you calculated how long it would take you to create a random order list of all valid entries (up to a potential 99,999,999,999)?

Your solution would certainly be more determinate, but is unlikely to be typically faster.

[Electricity_Guy]

Shurely theres some mishtake? Hic!

420 million prizes, worth £19.1m?

Shurely theres some mishtake? Hic!

That's an average of 4.5p for each prize.

[Anonymous Coward]

Re: Shurely theres some mishtake? Hic!

The top prize was not always £1m

[STOP_FORTH]

Fake picture

Fensom is not smoking a pipe and does not have a small slide-rule protruding from his top pocket.

How can he have been a boffin?

[Andy The Hat]

Is CECIL actually a long lost nephew of ERNIE?

Is CECIL even random? How does it work? Is it magic?

I heard Rachel Riley just works out a calculation beforehand, discretely types the number in on a keypad which CECIL displays using its super-technical 3-digit 7-seg display, then Rachel substitutes the contestant's selected cards for those she used before and pretends she got the answer by scribbling her workings from earlier? It's all fake!

Now, to prove it all we need is a decent shot of her keypad ...

[Anonymous Coward]

Re: Is CECIL actually a long lost nephew of ERNIE?

Interesting conspiracy theory, but I think it has the faint whiff of BS about it. The contestant is free to pick a different number of "big ones" from the top row or "little ones" from the other row,....*unless* the contestants themselves are in on this ruse, and are briefed in advance for what combination to ask for. Hmmm....the plot thickens.....

[STOP_FORTH]

Re: Is CECIL actually a long lost nephew of ERNIE?

CECIL can't be totally random, surely? If a contestant gets a load of ones and a 25 (or all ones, if you don't have to pick a big number - I don't know the rules) and CECIL indicates 999 you'd be stuck.

Didn't Turing write something about "computable numbers"? He worked at Bletchley too!

[Anonymous Coward]

Re: Is CECIL actually a long lost nephew of ERNIE?

But you win the round by being closest don't you? So even if it's impossible to hit the number exactly you can still do better than your opponent(s) ...

[John Brown (no body)]

Re: Is CECIL actually a long lost nephew of ERNIE?

When I see CECIL in an IT context, it always reminds me of Computer Education in Schools Instruction Language.

You probably have to be a certain age to get the horror of that!

[Hans Neeson-Bumpsadese]

Re: Is CECIL actually a long lost nephew of ERNIE?

I'd completely forgotten about CESIL....that takes my back to my 'O'-level days (last year of 'O' levels before they were replaced by GCSE, first year of the Computer Studies syllabus).

As I recall, everyone on the course had fairly extensive computer experience, so we basically wrote programs in assembler and then translated back into CESIL.

[Timbo]

Re: Is CECIL actually a long lost nephew of ERNIE?

"I heard Rachel Riley just works out a calculation beforehand, discretely types the number in on a keypad which CECIL displays using its super-technical 3-digit 7-seg display, then Rachel substitutes the contestant's selected cards for those she used before and pretends she got the answer by scribbling her workings from earlier? It's all fake!"

Oh yea of little faith...casting doubt on the hugely talented Ms Riley !!

There have been a number of times when the target number has not been able to be achieved, either by contestants or by Ms Riley herself...and in the latter case, she freely admits it cannot be done.

So, either she is a very good mathematician, or these instances are used to cover up your so-called fake calculations ;-)

In one edition, Ms Riley admitted it was a simple random number generator that did not take into account the numbers actually displayed on the board. Hence why sometimes the solution is very easy and other times, when it is impossible to achieve.

[Aladdin Sane]

For all involved----->

[virtualcolossus]

Virtual ERNIE

If you want to know more about the original ERNIE 1, I have written an online simulation which is free to use and you just need your browser.

It's at https://ernie.virtualcolossus.co.uk and you can learn how ERNIE worked and get it to generate and print some bond numbers (although not ones that mean you'll win anything!). As stated in the page above, it's a phenomenal machine for it's time and was a pleasure to learn about and build over the past year!

[Martin an gof]

Re: Virtual ERNIE

Upvote for boffinry, but you did actually get a mention in the article too :-)

M.

[Primus Secundus Tertius]

Not the real reason

I always assumed the real reason for building ERNIE was to generate random numbers for encrypting government documents.

I have won prizes from ERNIE, but only ever the smallest possible ones. Whereas I twice had a four-number lottery win, before they changed the lottery rules so that in effect they pay out only what they feel like paying.

[Cynic_999]

That explains it

My premium bonds could never be chosen by Ernie, because they are not random numbers, and Ernie only chooses random numbers.

[Snowy]

Random or is it

[quote]ERNIE's activities are also checked before and after each draw by the Government Actuary's Department (GAD), which performs four tests to check the machine is truly random:

Frequency – whether every possible character in each position of the Bond number appears as often as it should

Serial – this looks at how many times one digit follows another, e.g. how often does a 3 directly follow a 7

Poker – this looks at the number of times consecutively generated groups of characters contain: four identical characters, three of a kind, two pairs, one pair, all different

Correlation – this looks for correlation between characters in two different Bond positions over a series of Bond numbers

[/quote]

If something is truly random then all of the above do not have to be true. Roll a dice 6 million times does not mean your going to get each number 1 million times. You can not give rules to what makes random random as it is random. 1 2 3 4 5 6 is just as random as 1 1 1 1 1 or 2 4 2 6 1.

[Will Godfrey]

Re: Random or is it

They obviously think that random numbers are too important to be left to chance.

[Primus Secundus Tertius]

Re: Random or is it

"Dice" is the plural of "die". If you are rolling "a dice", you must be doing some clever quantum experiment with a virtual d****.

Leaving that aside, if you repeat the 6 million experiment a reasonable number of occasions, then each number 1 to 6 will turn up roughly a million times per occasion. The variation about a million will be described by the binary distribution, with a mean value of 1,000,000 and a standard deviation of 373. It would be one occasion in a thousand that the result differs from 1,000,000 by more than three standard deviations.

Similar statistical calculations can be made for the other scenarios. The testing of ERNIE would have been to check that these statistical estimates were duly observed.

[Anonymous Coward]

Re: Random or is it

That's true. But you can do the statistics on events like that and they are stupidly rare. In particular if a long enough sequence of purportedly-random numbers fails a randomness test, it is enormously more likely in practice that they're failing it because they're not random than not.

[Doctor Syntax]

"The prize money comes from the interest the government earns on the Bonds"

Shouldn't that be "pays". The bonds are loans to HMG. The recipient of a loan usually pays interest on it and the lender earns that interest.

[Anonymous Coward]

I think that the prize money is the interest the government pays on the bond (ie on the loan we make tothe government).

The only way the article makes sense is if you think of the premium bond bit of the government as beinga separate organisation which takes in money from us and then lends it, at interest, to other bits of the government.

(This comment reads as if I'm some kind of anti-premium-bond person: I'm not.)

[Allan George Dyer]

Did anyone else notice?

If you pause the videos of the ERNIE 1 demo machine, sometimes more than one tube in the ring is lit; on some occasions 3 adjacent tubes are active.

Is this an artefact caused by a slow shutter speed, or does it actually stem from the machine's operation?

[molletts]

Re: Did anyone else notice?

I would think it's mainly a result of the slow (relative to the machine's operation) shutter speed causing several "counts" to be displayed during the time the "shutter" is "open" (i.e. the sensor is gathering photons prior to being read out and reset - its integration time) but there could also be a (small) element of afterglow from the ionised gas in the tubes - they don't quite extinguish instantaneously when the current stops (nor, indeed, is the switching speed of the circuits driving them infinitely fast).

[Anguilla]

NS&I is a persistent PITA

I'm pissed off with NS&I as they NEVER advise changes of their "rules" anywhere subscribers can INSTANTLY see them.

Like today, in Hong Kong, I was in receipt of my returned cheque for Atlantic Pesos (£) 720 which has infringed somehow on their CHANGED form for subscribing my miserable UK pensions (for several months) into the hands of my Granddaughter.

It was several years ago - when I had collected all the info that THEY could provide - that I decided that my pittance would be better served in passing on to my granddaughter - that they - and especially Jane.Platt decided to "evict" my own £50,000 holding a few days from the next draw on the specious argument that I had complained too much at being buggered around by them in my attempts to get my initial deposit secured for my granddaughter.

[peterw52]

Then what happens

The thing that's always worried me is what happens post ERNIE. There must be systems taking the output and checking for valid numbers and matching these to actual names and addresses. I assume these will be normal IT systems with all the usual opportunities for errors. It's somewhere in this system that I think my numbers are being lost!