Dan Finlay 🦊
@danfinlay
The bots will never be this easy to detect again.
8 replies
98 recasts
82 reactions
radmadical
@radmadical
I've been working on a blockchain prototype for about a year now (still very conceptual - real part time) that uses every single feature as a PoH (proof of human) criteria that will constantly inform the PoH score based on long term human like behavior (playing games (each game, each turn each shot is unique, etc), paying for things, commenting, etc) so that the cost of faking a personal account is prohibitively high given the time and complexity of raising that score vs the potential payoff of doing so - I foresee a blockchain that represents identity aside any number of applications that could, bit by bit, verify a real human is behind it through normal usage activity over time - I think that's the way to beat the bots in the long run, make them too expensive and difficult to bother developing - what do you think?
0 reply
0 recast
0 reaction
CryptoShroom
@cryptoshroom
But AI will develop itself and imitate human behavior
2 replies
0 recast
0 reaction
radmadical
@radmadical
The other thing is the variation in applications - if the ecosystem were to explode, a high score would very expensive to produce - you need it playing games, listening to music, buying things online, commenting on social media, storing files on cloud, etc etc etc. Things that represent incidental effort to an actual human - but a serious cost barrier to bots/AI attempting to make fake "people". Hopefully that drastically reduces if not eliminates the number of fake accounts as the cost benefit analysis will rarely warrant the necessary effort...
0 reply
0 recast
0 reaction
radmadical
@radmadical
This is what I expect as well - but no matter what it'll always be an arms race, so when AI or other automation manages to pass the cost barrier, you can find more complex relationships that the AI wasn't designed to account for - that's the great thing about organic vs synthetic data - you can rely on it's authenticity and draw more complex insights by comparing known synthetic vs known organic data - so let's use the game as an example. If score and shot trajectory, etc, are being effectively replicated, what about movement behavior, like user response after opponent shoots - then you can update the engine that produces PoH and identify falsified PoH scores and blacklist them. I'm not sure if this can work forever - but remember, you don't have to make it impossible, just prohibitively expensive to do...
0 reply
0 recast
0 reaction
