Today we are living in a dual world. On the one hand, we are worried that AI will become an intelligent arithmetic in the not too distant and not so near future and set everything on fire. Robots will be as fragile as their creators.
But forget those emotional stories about having to deal with robots too. It is enough to continue to develop artificial intelligence as much as possible, and suddenly we will see that we are solving the failure of a handful of algorithms and silicon components. In this article, we want to remind you why you should be afraid of artificial and unknown minds.
Join Digito as we review the stories of human failure with artificial intelligence.
Yurisco, the ruthless space commander
Lent seemed to be making a foolish attempt. While experienced players flew ships of various sizes, AI had built a large fleet of 96 ships, 75 of which were heavily armored ships with giant rocket launchers and, due to their heavy weight, were virtually incapable of moving. This fleet drove all the attackers off the field, but at a heavy price. AI ships were very easy targets for enemies, but they were very, very large. The paddle typically lost 50 ships, but the fleet was able to easily defeat 20 competitors.
Lent’s ruthless artificial intelligence was called Yorisco. Lent began programming it in 1976 and fed a variety of rules into the system. AI could start testing different variables and eventually find a solution. The same trial and error led to the evolution of the system as much as possible, and artificial intelligence was able to produce successful ideas by combining different methods. Many Traveler variables were a great platform to test Yoresco’s capabilities, and Lent worked on 100 Xerox computers for 10 hours a day for a full month to get the final design.
Yurisco was so effective that Traveler game designers were forced to change the rules for next year’s tournament, and fleet agility became an important element of victory. Yurisco competed again with the same design, but again with the help of a new tactic. This time, the AI was destroying those paralyzed ships so that the agility of the entire fleet would not be lost. Cruelty. cruelly.
A flawless game and an almost flawless player
Much of the research on artificial intelligence has been with the help of classic games such as chess, checkers, poker, and the like. These games have simple rules but can be experienced in a very complex way. Human players are usually just as complex. But when these players go against artificial intelligence, usually tragedy strikes.
One of the simplest classic games is Checkers, and it was also the first game in the world in which AI acquired more skill than a human. Named the Chinook AI, it was developed by a team led by Jonathan Schaefer, a professor at the University of Alberta, to compete with the best checkers player in the world, Marion Tinsley. Tinsley was almost the most perfect checker player. During his 45 years of activity, he had lost only 5 games to human rivals. Tizenli’s opponents usually played conservatively and tried to equalize with him. But Chinook had gone to win and was playing very dangerously.
Chinook and Tinzley first met in 1990, and Tinzley won. The second game was held in 1992 and the prize was the world championship. After 39 games, 33 of which ended in a draw, Tizenley lost just two games and Chinook conceded four. Thus, heroism came to humanity. The third encounter was in 1994. The two had six draws, but then Tizenley felt unwell and withdrew. Chinook was declared the winner by default, but the victory was not so impressive. Pancreatic cancer was diagnosed in Tizenli and he died the following year.
Chinook was never able to defeat him directly, leaving Schiffer disappointed: no one but him could challenge Tinzley. So he devoted the next 12 years of his life to checkers to find the perfect form of play. He succeeded in 2007 and realized that the most perfect checkers game would lead to a draw.
When Gary Kasparov accused the computer of fraud
At least Chinook did not face the problem that Deep Blue experienced after defeating legendary chess player Gary Kasparov in 1997. Kasparov did not perform very well due to the high pressure he experienced in the match event, and during one of the games, he noticed a strange move by artificial intelligence. In fact, AI had the opportunity to cult Kasparov and did not use it. Thus, the Russian player withdrew from the match and accused IBM, the manufacturer of Deep Blue, of fraud.
He sought evidence that artificial intelligence had decided to do so, but IBM was unable to provide the information. Deep Blue used search trees to make decisions, and the information processed was so much more than you could easily deliver. It became a conspiracy theory for a long time, but Deep Blue eventually won three games, making Kasparov the first world champion to leave the game to artificial intelligence.
A bad chatbot that accidentally failed the Turing test
You’ve probably heard of the Turing test. This test was developed by Alan Turing in 1950 and is a way of judging whether artificial intelligence can behave intelligently. Attempts to pass this test have led to the creation of a pile of chatbots designed to impersonate real human beings and deceive people. The first known case was the Eliza robot, which was launched in 1965, and today there is a competition called the Lubner Prize for such chatbots. It is not clear for many whether this test has been passed or not. But the claim of passing the Turing test has been heard many times, and one of them belongs to a robot named Eugene in 2014, who pretends to be a 13-year-old Ukrainian boy.
But the truth is that the Turing test was passed long ago, in May 1989, by chance. The winning robot was called mGonz and was developed by Mark Humphreys, a 20-year-old student at the University of Dublin. He placed the robot on a server to receive chat messages when he himself was not available. And one day, someone from Drake University starts crawling on the server and is confronted with the answer: “Give up this mysterious task and speak in full.” What happened next was an hour-and-a-half conversation between the robot and an anonymous student.
The secret of mGonz was that it was programmed to speak like a bitch. Man asks, “Mark?” And Bat replies, “Mark is not here and he left me to deal with idiots like you.” The conversation quickly turns into a romantic relationship, and mGonz repeatedly asks the man when he had the last relationship. He claims that the last time was the night before. But after a long conversation, he gets tired and admits that the story goes back long before the last 24 hours. The robot wins.
The mGonz robot was not that sophisticated, but it did show one thing well: that when you question men’s romantic relationships, they try to hide everything and make it look good.
Unreal bots abuse the stupidity of being a stupid human being
Botprize is a competition that is practically the equivalent of the Turing test for video game AI. The competition is designed for bots that are designed in the heart of the Unreal Tournament game and aims to encourage developers to build artificial intelligence that makes the game experience more exciting with human-like behavior. In this race, we are not going to see super-intelligent AI defeating humans, the bats will eventually be as stupid as we are.
In 2014, we saw two winners in this competition, MirrorBot and UT ^ 2. The two robots were able to achieve a “human performance” score of 52%: a figure that is obtained by dividing the number of times the robot looks like a human by the total number of times it has played. By comparison, the best human player in the Unreal Tournament scored 53%, which means that the two robots had a great achievement.
MirrorBot, which scored by a very small margin, was actually less complex than UT ^ 2. UT ^ 2 used a revolutionary training system to formulate human strategies and to choose the best possible option from them, depending on the circumstances. But MirrorBot was much smarter. This robot abused our social nature, that we display behaviors similar to our own identity. MirrorBot simply looked at human players and imitated them. When the robot was in sight of the players and did not pose a threat of death, it exhibited the same movements as human players. And that was enough for the real players to consider him one like themselves.
Bats that defeat professional video game players
Some AIs are really good at video games. DeepMing’s self-taught AI can defeat players in 29 of Atari’s 49 old games and is getting stronger every day. These games include Space Invaders and Breakout.
Then we have another AI designed to play StarCraft: Blood War. The best examples of this AI game are really good: this robot can defeat human players, but it is not yet powerful enough to push professional level players away. In fact, the patterns behind the robot’s strategies are a bit obvious and exploitable.
AIIIDE StarCraft AI organizers estimate that StarCraft AI will be able to defeat professional players in the next 5 to 10 years. But by then, Blood Wars is getting so old that no player will have a career.
Financial robots that get rich before we read the news
Like us, you may be terrified of greed in the world of international banking right now, but did you know that the stock market is now largely run by artificial intelligence? Algorithmic approach or high frequency exchanges is a phenomenon in which bots monitor market inflation and international news and act quickly to take advantage of even the smallest potential opportunities. In today’s world, being first is everything: prices soar in the stock market as brokers set up offices as close to them as possible to minimize delays.
Because we are talking about an incredible amount of money, the world of algorithmic brokerage remains highly secretive and few people know how, when and where this approach is taken. But experts are confident of at least one example in which a robot made $ 2.4 million in one day. The robot invested in a company called Altera, right after rumors surfaced that Intel was looking to buy the company. The logical conclusion is that the robot has read the news, understood it, and decided to buy stocks. That is exactly a few seconds after the news was published. There is no hope that humans can have such practical speed.
Of course, leaving the market in the hands of the Bats has been a frightening moment. For example, during the sudden fall of the Dow Jones Industrial Average in May 2010 and its recovery, the main culprits were high-frequency, bot-dependent traders. The big picture is still a mystery, but the point was that a robot made $ 4.1 billion in futures and delivered 75,000 contracts in a crazy 20 minutes. After that, prices fell and other high-frequency robots began exchanging the same contracts. By buying and reselling these contracts, like potatoes, we saw the conclusion of 27,000 contracts in just 14 seconds.
In the meantime, the robot that started everything simply started to accelerate the stock selling process. All the market participants were watching this madness with astonished eyes. The market was in turmoil and then revived, but it took months for researchers to finally figure out why. And now imagine that our lives and economics are in the hands of such unconscious robots.