That's not relevant, though - except in the most trivial way of the problem of induction[0]. Technically, we can't be
sure the sun will rise tomorrow, or that gravity will still work the same way 5 minutes from now. Neither can we be
sure the market will not collapse next week. But this way lies madness - reasoning about the future is impossible. Fortunately, experience consistently demonstrates that if we have a good model of something, reality tends to stick to it, so we can use it to predict things.
With nihilism out of the way, how is million people independently trading a thing for $10 once meaningfully different than a single buyer/seller pair trading a thing for $10 a million times?
A price of an item isn't a random phenomenon - it's just a reflection of what buyer and seller believe other people would pay for that item. This belief is based mostly on knowledge of what other people actually paid for it (or a similar widget) in the past. This process is mostly convergent[1]. Prices change at the rate of information flow, and tend toward some equilibrium. The process may be quite unpredictable, but the expectations are bounded.
As a result, markets automatically price everything relative to everything else, in a way that mostly makes sense. This is a very useful property - it's a bottom-up, implicit, somewhat fair way of solving resource allocation problem in society.
Generating fake trades like this? It's injecting bad data into the market. It's poisoning the mechanism of price determination - which, to produce reasonably fair determinations, needs aggregate trades to average great many distinct buyer/seller negotiations. Amplifying a single datum with fake trades ultimately makes the market worse at efficient resource allocation, and thus less useful to society. It's pissing in the pool from which everyone drinks.
--
[0] - https://en.wikipedia.org/wiki/Problem_of_induction.
[1] - Or perhaps "consilient", which is a nice word I just discovered - https://en.wikipedia.org/wiki/Consilience.