According to Ray Kurzweil the Turing Test will be convincingly passed by 2029, but I don't think he quite predicted the way in which that might occur. This article suggests that internet based bots might be used to trick users into giving away information about themselves, or clicking on links to sites containing malware. The proposed method is a situation where the bot is behaving as a kind of parasite, linking two human chat participants together and then modifying the messages in transit in order to insert dubious content. It's not quite a classic Turing Test, but it's still a scenario in which some amount of AI might result in profits for the bad guys.
More in the tradition of AGI is the notion of applying data mining methods to social network information, on order to reconstruct a coherent picture of users lives. Knowing the daily movements of an individual could be useful to crooks planning a robbery, and even a quick cursory search on Twitter demonstrates what possibilities exist, such as:
"I'm going out"
"Going shopping"
"Leaving the house"
"Won't be back until"
If the messages are also geo-located then this surely provides a goldmine of information which could be used for malevolent purposes, and even a weak AGI or narrow AI using primitive natural language processing would probably be successful here.
It's possible to imagine a near future cyber-crime scenario in which an AGI is both mining social networks for data and then hiring and directing human resources to grab the loot, and even instructing them on where the most profitable places to dispose of the ill-gotten gains are. It would be like a kind of mafia operation, but without any human leadership and where the human components of the system were almost entirely deprecated. Perhaps there's a sci-fi novel in there somewhere...or maybe it's a novel that's already been written.
Subscribe to:
Post Comments (Atom)
3 comments:
Spammers are probably among the most technically advanced bad guys. Generally, it seems that the white hats are ahead of the black hats. The main problem seems to be that some white hats are less white than others. I generally cite Microsoft as an example of how things could go badly.
It might be that such a system already exists, with some of the bot masters themselves being bots. Fortunately for the moment it looks as if the spam problem is contained, with phishing being the main frontier of malevolence. It would seem that phishing isn't really scalable without AI.
It is hard to get reliable figures - but I am inclined to think that spam is the bigger industry:
http://en.wikipedia.org/wiki/Spam_%28electronic%29#Costs
http://en.wikipedia.org/wiki/Phishing#Damage_caused_by_phishing
Post a Comment