Post without Account — your post will be reviewed, and if appropriate, posted under Anonymous. You can also use this link to report any problems registering or logging in.

The Rise of AI

  • 1 Replies
  • 438 Views
*

Offline Phil

  • Global Moderator
  • Hero Member
  • *****
  • 553
    • View Profile
The Rise of AI
« December 28, 2018, 01:50:18 PM »
This topic will be (at least, starting out as) a collection of links to interesting articles on AI. Discussion is welcome. If the volume becomes sufficient, it might be split up into several subtopics under a child board.

Here's the first: AlphaZero, the unstoppable chess bot. Machine learning by playing millions of games against itself and earlier machines, once given the rules, to learn the best techniques.

*

Offline Phil

  • Global Moderator
  • Hero Member
  • *****
  • 553
    • View Profile
Re: The Rise of AI
« Reply #1: February 03, 2019, 11:11:21 AM »
An interesting article on CAPTCHAs: https://www.theverge.com/2019/2/1/18205610/google-captcha-ai-robot-human-difficult-artificial-intelligence . It states that AI is expected to improve to the point that it will solve any CAPTCHA puzzle much better than humans can. It's just about there, already. The emphasis will have to shift from how perfectly the "user" can solve a problem to watching how very human imperfections and randomness in the interaction betray who is human. Also, rather than relying on a one-time hard-shell defense against bots, we will have to watch users in their interactions with a site and see if they're doing bot-like things. Big Brother, anyone? The article points out that Third World CAPTCHA farms use people to sign up for forums and blogs, etc., which then can be handed over to bots to do the spamming. This would require monitoring of the user interactions beyond just the signup, such as an occasional CAPTCHA challenge from time to time. If most spammers crap on your forum just once (or use your tell-a-friend function for one mass mailing) and then never come back, that may be more annoying than useful.

The comments are rather interesting too. Several people pointed out that the reCAPTCHA emphasis on traffic lights and street signs and vehicle recognition suggests that we are being used to train Google's self-driving cars -- for free.

In general, any kind of CAPTCHA is a hard-shell defense against bots signing up. Once they get through, and they will, they can run amok unless you have further defenses to detect and stop them. This can include how the user interacts with the page (irregularities in timing, etc.), random rearrangement of page layouts to stop bots expecting a certain page layout, occasional challenges, speed of moving through pages (clicking links), subject matter entered (lots of links?), and so on. Just be careful not to make it difficult for humans with visual, hearing, or dexterity problems. And needless to say, AI will improve to make a bot "look" more and more like a human in its interactions.