A wave of AI robots is threatening to ruin HQ Trivia

02-10 00:26

Enlarge / They know...

Aurich / Getty

Cheating in online games is an ever-present problem that infects the likes of shooters, MMOs, and  open-world crime simulators —and just about everything else. Now a rising tide of cheaters seems poised to threaten mobile general knowledge tester HQ Trivia

as it continues to explode in popularity.

The daily trivia game attracts millions of players to battle for real money 12 times a week by answering 12 multiple-choice questions sent via live video stream. In recent days, though, the app makers have been locked in their own battle with sites like HQuack . These bot sites use optical character recognition and Google to try to figure out the answers to the game's questions and feed them to players before the game's ten-second timer is up.

Bots like these are still imperfect—HQuack advertises only "up to 82 percent accuracy," which is often not enough in a game where a single wrong answer leads to elimination. But if and when they work, they have the potential to ruin a game that's becoming a bona-fide phenomenon.

Running the numbers

To see just how much of a problem these question-answering bots can cause, look no further than the 3:45 HQ Trivia game on Tuesday, February 6. Of the 786,883 who started the game, a full 9,046 answered all 12 questions correctly, a number host Scott Rogowsky confirmed was a record. Each of those winners got a grand total of 23¢.

For context, just a week before, in an exceedingly average January 31 game, a playing crowd of 798,796 had been reduced to just 81 winners, who each won $30.86. That's a more than hundredfold increase in the win rate, from 0.01 percent to over 1.1 percent.

It's probably no coincidence that the AI bot at HQuack happened to get every single answer right during Tuesday's show. The week before, the site had barely launched to the public and had yet to hit the mainstream via an Outline article . (There are anecdotal reports that a reusable extra lives bug may have also contributed to the record result.)

You can see the HQuack effect even when the bot misses a question or two. On a Thursday, February 8 game, 11,581 players managed to answer the first 11 questions correctly, just as the HQuack bot had (the bot actually missed the second question about the Olympic logo, but it was so simple that about 98 percent of the playing humans managed to get it right anyway).

For question 12, though, HQuack's algorithm suggested that Karl Marx was born with the same name as a famous ketchup, and over 8,800 of the remaining 11,500 players agreed. Only the 2,000 who knew the correct answer was Henry Kissinger (born Heinz Kessinger) ended up winning the prize. A similar pattern of seemingly HQuack-led answer grouping could be seen in the final few questions of Sunday's Super Bowl halftime edition of HQ Trivia, where the bot and its loyal players eventually fell to a question about the Reimann hypothesis .

  • On Feb. 6, a record 9,000 HQ Trivia players got this final question right to win a share of $2,500...

  • ...there's a strong chance most of those players were looking at a page like this one from HQuack, which gave them the answer through Googling.

  • For the final question of a Feb. 8 HQ game, HQuack suggested that Karl Marx originally shared his name with a famous ketchup...

  • ... a huge majority of the remaining players followed HQuack's advice, missing the correct answer this time.

  • HQuack sometimes misses relatively easy questions. Despite this wrong answer, over 98 percent of players got this right as Question 2 in an HQ game.

  • HQuack's Google-based algorithm can be misled, too. the New York Stock Exchange is actually owned by ICE, though the NYSE acronym likely shows up more often in Google results.

Dedicated HQ players are beginning to notice this growing problem. "Will this even be worth it if they cant [sic] cut down on programs googling?" Redditor Vikemin1 asked somewhat rhetorically on the HQ Trivia subreddit this week. "There's really no reason to keep playing this game until they find a way to take care of the bots," redditor cbooz added by way of answer in another thread.

Where trivia cheaters come from

HQ players have been manually speed-Googling their way to HQ wins for a while , and a numbers of hackers have talked up their own automated methods for finding answers since December at least. But it was the recent launch of HQuack (and similar sites like the members-only HQHelp ) that seems to have upped the prevalence and effectiveness of HQ cheaters.

HQuack creator Jake Mor told Ars the site started as a simple program on his computer, Googling questions and giving confidence intervals to the possible answers based on the number of results. After throwing the program on a website for his friends to use, Mor told Ars that "word of the site kind of just spread."

Mor wouldn't tell Ars how many people are using his site these days, but he told The Outline last week that he was already getting "a thousand unique visitors a week" a few weeks ago, and that he'd been seeing "20 percent day over day growth."

Mor says he doesn't make money off of HQuack and never intends to. "This just started as a side project out of curiosity," he told Ars. "I never expected it to pick up the way it did."

And while he acknowledged that the site "may have played a part" in recent record-setting win rates, he said he doesn't want to destroy the legitimate HQ experience. "If it does start to truly ruin the game, I think I will take it down."

Fighting back?

While HQuack and other automated tools are prohibited by HQ's terms of service, preventing people from using them would seem difficult as a practical matter. After all, how can the app tell the difference between someone who just knows a lot of trivia answers and someone who's being told a lot of trivia answers by a website? Even pattern-matching algorithms could be thrown off if players just go against the bot every so often. (HQ has not responded to a request for comment from Ars.)

HQ's best protection for now might be the bots' imperfection—with the aforementioned 82 percent accuracy, HQuack will be perfect or near-perfect for some 12-question games and then miss four questions in a row on others. Mor says HQuack currently has the most trouble with "double questions" where the answer requires knowing two things at once (One recent example that tripped up HQuack: "If you capitalize it, what Disney character name doubles as a healthcare program?" Answer: Chip).

Changing up the question formats could help too. While a bot can search for the answer to a text question pretty easily, questions based on pictures or audiovisual snippets could prove more resistant to simple Googling. Google also struggles with more abstract questions like "Why did..." and relational questions like "How close is [place] to [place]" according to HQ bot maker Stephen Cognetta .

But Google's question-answering algorithms are getting better every day, and more sophisticated AI approaches could perform even better at the game. In a world where

, a multiple-choice quiz app seems like easy pickings. And with HQ offering real money for winners 12 times a week, the incentive for hackers to continue to try to beat the app will remain high.

原文链接:https://arstechnica.com/gaming/2018/02/a-wave-of-ai-robots-is-threatening-to-ruin-hq-trivia/?utm_source=tuicool&utm_medium=referral
标签: HQ Trivia 谷歌
© 2014 TuiCode, Inc.