Would you say that a "good" tool necessarily needs to be deterministic?
(Edit: Exceptions of course for things like a random number generator)
Discussion
Would you say that a "good" tool necessarily needs to be deterministic?
(Edit: Exceptions of course for things like a random number generator)
Casino Owner: Yes definitely!
@tante No, because of various statistical methods (Monte Carlo has been mentioned), and my double slit experiment isn't deterministic either ;)
@tante
Do predictable, reproducible and auditable results mean deterministic?
@tante so following from your edit, the poll asks whether tools need to be deterministic except when they need not to be? My boltzmann machine is ruled out? 🤔
@tante Thinking about it for a few minutes, this quickly becomes very philosophical. Some believe the whole existence is deterministic.
But nevertheless we have structures that seem to us humans non deterministic. Given that, I would say a 'good' technology should not only be deterministic it should also seem deterministic to us humans. Otherwise its magic and not technology.
@tante I answered “yes”, because for me a tool does one particular job. And I need it to be dependable and correct. That said, there are probably niche use cases beyond random numbers that warrant non-deterministic results. For example, I use a transcription tool which is (because it is an LLM use case) non-deterministic, but the result is easily verifiable.
@tante random number generators spring to mind...
@tante There’s not a lot of software that is deterministic, but I’ve never met a good one that isn’t.
@tante Paint brushes are not very deterministic, I think they are good.
But maybe the question builds upon a abstract random/deterministic dichotomy that makes not sense for many technologies.
@tante No, because RNGs can be a "good" piece of technology. Of course, those are deterministic if you fix the seed, but let's ignore that for now.
@tante trick question! We don't live in a universe that allows deterministic processes.
@tante This deeply depends on the purpose. Sometimes I want a stochastic simulation. Monte Carlo methods exist, after all, as do other forms of statistically legitimate sampling. But if I don't have a _valid statistical reason_ which has been _checked by a statistician_, then no. Deterministic methods only. We do not play slots in engineering.
@tante Yes, unless I want something like a random number generator.
@tante at the same time
"Any sufficiently advanced technology is indistinguishable from magic."
If it fulfills a deterministic task.