Chemistry - How are poisons discovered? Does someone have to die/be poisoned from it first?
Solution 1:
Alle Dinge sind Gift, und nichts ist ohne Gift, allein die Dosis macht dass ein Ding kein Gift ist (The dose makes the poison) - Paracelsus
Poisons (I'm going to use this as an umbrella term for "toxins" and "venom" as well. Bear in mind though, they are not the same thing) have been known since antiquity.
Back in the good old days, you figured out if something was poisonous or not by eating/touching it (or getting someone else to do it), i.e- "discovering" a poison was simply a matter of chance.
These chance encounters alone, led to the discovery of numerous poisons.
With the advent of Chemistry, the gents in white lab-coats figured out that compounds that bear resemblance to the already well-known poisons, in terms of their functional groups and structures, are also toxic (albeit, to vastly different degrees). Poisons could now be identified a priori (you could tell it would end badly if you were exposed to such a substance, but not how badly).
And no, you can't really quantify a poison's effects without testing it on something. Nor can you tell how much of something would be needed to kill or severely maim.
This is where we bring in the idea of a median lethal dose $\pu{LD_{50}}$ which is, simply put, the minimum amount of a substance required to kill off half of all the animals in a particular test group.
$\pu{LD_{50}}$ values for a particular substance depend on the animal used.
Of course, the only way to get an accurate $\pu{LD_{50}}$ (which itself, is really a "mean/average" value of sorts) for a human is to actually poison someone, which doesn't sound very nice. So you do the next best thing, you measure it for a rat/chimp, couple it with your knowledge of the poison's mechanism of action, and extrapolate the value to something that would kill a person.
Another way to establish degree of toxicity, which doesn't involve killing animals, would be by exposing a cell/tissue culture (and not the entire animal) to the potential poison .
Solution 2:
As you already correctly deduced, the discovery of poisons was in former times quite accidental, but once its potency was discovered, the (mis)use of it was predictable. It must also be said that our ancestors were very careless with poisonous substances. The old houses used wallpapers and paint which were spiked with arsenic, lead and antimony. Copper arsenite or Scheele's Green, Copper (II)-acetate and White lead were used for the resulting luscious colors.
The alchemists and later the chemists did test substances on themselves. While the alchemists were often reckless by trying their concoctions undiluted, the chemists were more careful and used strong dilutions which were carefully increased in potency. Still, taste and smell were methodically used until the first part of the 20th century. Christian Wilhelm Posselt and Karl Ludwig Reimann e.g. tested the extracted nicotine on the tip of the tongue and remarked the burning feeling (Nicotine is by the way very poisonous and is used as insecticide). Carl Wilhelm Scheele was not so lucky, his continued testing of every discovered substance finally overpowered his body and killed him. LSD was experienced by Albert Hofmann despite precautions: He chose a dilution which was considered safe, but LSD was so powerful that he still got a 20 times overdose.
Since the advent of science, the information about the potency of poisons has been recorded independent of the cause: accident, suicide or homicide. Animal tests were used, but they were never really completely reliable. LSD seemed to be noneffective in animal tests, but is one of the most potent human drugs in relation to its quantity. The poison of the funnel-web spider is ineffective against dogs or cats, but could be fatal for humans. In contrast, wonderful tasting chocolate is deadly for dogs and cats. This can be mitigated a bit by using primates as test animals, but they are still different enough that deductions from results must be taken with care.
What we learned is that the effect of most poisons is unreliable. There are many, many recorded instances of humans dying on harmless doses caused by an overreaction and also people surviving incredible overdoses of e.g. cyanide poisoning. So the value of $\pu{LD_{50}}$ is a rule of thumb how poisonous a substance really is. Chemistry is now advanced enough that people have a pretty good idea how poisonous the result is, but there will always be uncertainty about the exact effects.
The only way to get an impression of the exact effects would be (apart from heroic self-tests) human experimentation, and as repugnant as it is, this has been in fact done deliberately. Paracetamol's link to the old Romans mentioned that they not only tested animals, but also slaves. During WWII the infamous Nazi and the Japanese Unit 731 experiments tested poisons on prisoners.
Solution 3:
Sometimes poisons are discovered by chance. At least that is what happened to me. We were researching on products made with malonodinitrile and enones. Since I was interested in the mechanism, I used milder conditions in the reaction so I could isolate the Michael and Knoevenagel intermediates and in one case a much more complicated product (see here).
The Knoevenagel adduct between malonodinitrile and cinnamaldehyde yields a beautiful yellow product. At that time there were no gloves in the lab (and we were very young and a bit careless), so sometimes some mother liquor spilled on your hands when the product was filtered.
The first time I made the product my hands got red. I did not pay attention. The second time, my hands got red and my arms burnt. I started to think that there might be a relation. The third time I made it, I was red from head to toes and all burnt like hell. I had to go to the hospital.
It seems to be an allergic reaction, which, as usual, become worse every time you come in contact with the product that causes it. Not exactly a poison, but maybe in a higher dose it can be fatal. I have not made the product anymore.
So in this particular case, it was discovered by the effect on a human.
A note for the younger ones: everything you synthesize can be a poison, so use gloves, work in the fume cupboard and do not get in contact with it.
Solution 4:
If you want to get an idea if a substance could be a venom prior to animal or human testing, you can use computational chemistry to model possible interactions with human proteins. It is primarily used to study possible cures or better understand how a disease works, but if you can not find any information about the substance elsewhere, it will help you narrow the options and lower the number of experiments needed to be done, especially if you not only want to know if it can harm your body, but also how and what a possible treatment might look like. It is a far shot, but it lowers the costs rapidly.