Remix.run Logo
▲ wpietri a day ago

That's actually a great example of something where we don't and shouldn't trust computers. E.g.: https://www.kqed.org/news/12051862/sf-teachers-are-yet-again...

We can and should trust computers for the straightforward, boring parts. But there's a huge gulf between the people who think, "how hard could payroll be?" and actually making a system that handles all the interacting complexity of real-world problems.

▲wvenable 20 hours ago | parent | next [-]

All the problems with that payroll system are caused by humans, not computers.

The odds that computer is making a mistake in a calculation is near zero (although not impossible). Instead it's doing exactly what it is told to do.

▲wpietri 19 hours ago | parent | next [-]

All the problems with computers are caused by humans, not computers.

▲aorloff 20 hours ago | parent | prev [-]

I guess you never heard of the pentium bug ?

Edit : and more generally - math has infinite resolution, whereas computers have a finite resolution, and cannot be more precise than that

▲fluidcruft 2 hours ago | parent | next [-]

Humans doing payroll by hand are not using infinite resolution.

And to the extent they are there are always big math, rational and decimal math systems.

▲zamadatix 18 hours ago | parent | prev | next [-]

Computers have no more or less a finite resolution problem than humans - it just seems to be far larger than what is practically needed in both cases. I.e. problems can be encoded symbically or numerically and the maximum size of each (including the numerical precision of the latter) is bound by physical constraints. But just because computers use fixed sized numerical types does not mean they are unable to handle working with exact symbolic values or objects we use in math.

Regarding the Pentium bug, I think one has to agree when it stopped being up to the human introduction of the bug. I.e. sure, many think we have designed faulty CPUs before. Some feel computers have designed faulty computers, others feel it still comes back to the human definitions. At some point though, most seem to call human mistakes human mistakes rather than where they are derived from prior to humans. I just don’t think we’ve all reached agreement of when that boundary definition is is for mistakes from the computer.

▲hardbass 6 hours ago | parent | prev | next [-]

>computers have a finite resolution, and cannot be more precise than that

Yes and neither can us. We have never written infinite digits down of anything. We work in finite sets of fields.

The field of Q(pi) perfectly represents various expressions of pi for example.

▲aorloff 3 hours ago | parent [-]

You can just write Pi, which is precise, in math.

▲hardbass 3 hours ago | parent [-]

Yes pi, 4 atan(1), the ramanujan formula, a spigot program are all valid representatives for the same entity.

▲noosphr 16 hours ago | parent | prev | next [-]

The numbers you can talk about in mathematics are just as countable as the numbers you can talk about in computer science: https://en.wikipedia.org/wiki/Definable_real_number

This should be pretty clear unless you've never used arbitrary precision arithmetic.

▲aorloff an hour ago | parent [-]

Computer science is part of math. We are talking about math versus computing

▲wvenable 18 hours ago | parent | prev [-]

> I guess you never heard of the pentium bug ?

I was going to put that exact example in brackets after "although not impossible" but decided it was unnecessary.

▲fluidcruft 2 hours ago | parent | prev [-]

That a payroll system isn't implemented in Lean though