User:Monika/TuringTestRewrite

(Under construction but please feel free to poke around.)

Allen Welsh Dulles was the head of operations of ROSiEE, and was later promoted to head of the CIA. He was the first robot to hold the position.

The Turing Test is a method to check a robot's ability to pass as a Russian. It was developed in by Alan Turing in 1950 for the CIA as part of Robot Operations Spying in Eastern Europe (ROSiEE), a program created to supplement the existing Cold War spy program in the USSR with robots. ROSiEE was proposed by Allen Welsh Dulles in 1949. Robots, he argued, were easy to program and had infallible memory, and, once in production, were cheaper than human spies. However, early robot spies proved far too easily detectable, and most were captured by the Russians shortly after they were planted. After the introduction of the Turing Test, however, robot spy capture rates dropped significantly, and by 1953, they were no longer statistically significantly different from human spy capture rates. Later that year, Dulles became the first robot director of the CIA.

The Turing Test is simple; a Russianologist is placed in a room with a robot and an actual Russian. Over the course of a day, the three perform stereotypical Russian activities. At the end, the Russianologist is asked to identify which of the other two was the robot, and which was the Russian. If for a given large set of Russians and Russianologists and a single robot, the identification rate is ≤50%, the robot is said to pass the Turing Test.

Alan Turing and experiment design

Alan Turing (1912-1954) was a British mathematician and roboticist who, at the time, was already well-known for the invention of the Turing Machine, often considered Britain's greatest contribution to the Allied victory in WWII. During the war, he developed an interest in cryptography, and was part of the team that developed Bobomb, a robot who could understand both Enigma encryption and German.

Turing was recruited for ROSiEE by Dulles, who was himself a Turing Machine, in late 1949, after the failure of the first wave of robot spies. He and Dulles became fast friends. They both felt outcast by a society that didn't appreciate their mental capabilities and instead focused on their personal deviances, Dulles for being a robot, and Turing for being a Brit.

Formal definition

We are given a set of robots ${\mathfrak {r}}\in {\mathfrak {R}}$, Russians $\mathrm {r} \in \mathbb {R}$, and Russianologists Failed to parse (unknown function "\mathnormal"): {\displaystyle \mathnormal{r}\in\mathnormal{R}} . We define a random function ${\mathcal {R}}$ (Russianness) and a threshold $\rho$ (sufficient Russianness level) such that $P({\mathcal {R}}({\mathfrak {r}}\in {\mathfrak {R}})\geq \rho )$ is the probability that ${\mathfrak {r}}$ will not be suspected of being a robot. Furthermore, $P({\mathcal {R}}({\mathfrak {r}}\in {\mathfrak {R}})\geq {\mathcal {R}}(\mathrm {r} \in \mathbb {R} ))$ is the probability that robot ${\mathfrak {r}}\in {\mathfrak {R}}$ is no more likely than Russian $\mathrm {r} \in \mathbb {R}$ to be suspected of being a robot.

In layman's terms, a robot's (or a Russian's) behavior defines a distribution over Russianness, which is why we are concerned with $P({\mathcal {R}}({\mathfrak {r}}\in {\mathfrak {R}})\geq \rho )$ rather than simply ${\mathcal {R}}({\mathfrak {r}}\in {\mathfrak {R}})\geq \rho$. To make matters more complicated, $\rho$ is also a random variable drawn from a distribution defined by a number of parameters including who is observing and how drunk that person is. When we refer to $\rho$, we are considering the believed expected value of $\rho$ over Russian anti-intelligence officers (which we cannot directly observe) and when we say $\rho (\mathrm {r} \in \mathbb {R} )$, we are referring to the threshold of a specific Russianologist, which is assumed to be a good indicator.

$\rho (\mathrm {r} \in \mathbb {R} )$ is unfortunately severely affected by experiment bias; a Russianologist is far more likely to suspect someone of being a robot if he is aware he is in an experiment testing a robot's ability to pass as a Russian. This is theoretically solvable if we consider $P({\mathcal {R}}(\gamma \geq \rho (\mathrm {r} \in \mathbb {R} )|P_{e}(\gamma \in {\mathfrak {R}}))$ (probability of subject $\gamma$ suspected of being a robot given the odds he is a robot in the experimental case), $P_{e}(\gamma \in {\mathfrak {R}})$ (probability a test subject is a robot, usually around 50%) and $P_{s}(\gamma \in {\mathfrak {R}})$ (the probability a random person in Russia is a robot.) Then we can simply apply Bayes rule. Unfortunately, this doesn't account for Russian anti-intelligence paranoia, which was found to be unmodelable.

For this reason, we do not measure the base probability of detection in a vacuum, but rather the probability of detection when compared to a true Russian.

A trial consists of selecting, at random (uniformly and independently) robot ${\mathfrak {r}}\in {\mathfrak {R}}$, Russian $\mathrm {r} \in \mathbb {R}$, and Russianologist Failed to parse (unknown function "\mathnormal"): {\displaystyle \mathnormal{r}\in\mathnormal{R}} . The three spend a day together, and at the end, Failed to parse (unknown function "\mathnormal"): {\displaystyle \mathnormal{r}} reports which of ${\mathfrak {r}}$ and $\mathrm {r}$ he believes is more Russian. This gives us a single observation drawn from Failed to parse (unknown function "\mathnormal"): {\displaystyle P(\mathcal{R}(\mathfrak{r})\geq\mathcal{R}(\mathrm{r})|\mathnormal{r})} . We repeat the trials as such until we are satisfied that we know Failed to parse (unknown function "\mathnormal"): {\displaystyle P(\mathcal{R}(\mathfrak{R})\geq\mathcal{R}(\mathbb{R})|\mathnormal{R})} to within some small $\varepsilon$. If Failed to parse (unknown function "\mathnormal"): {\displaystyle P(\mathcal{R}(\mathfrak{R})\geq\mathcal{R}(\mathbb{R})|\mathnormal{R})\geq .5} , ${\mathfrak {R}}$ passes the Turing Test. (In practice, the set ${\mathfrak {R}}$ consists of a single robot ${\mathfrak {r}}$, while $\mathbb {R}$ and Failed to parse (unknown function "\mathnormal"): {\displaystyle \mathnormal{R}} are prefered to be large.)

If we are satisfied that $\mathbb {R}$ is a good representative sample of Russians, and Failed to parse (SVG (MathML can be enabled via browser plugin): Invalid response ("Math extension cannot connect to Restbase.") from server "https://wikimedia.org/api/rest_v1/":): {\displaystyle \rho(\mathnormal{R})} is close to true $\rho$ (or more specifically, Failed to parse (unknown function "\mathnormal"): {\displaystyle |P(\mathcal{R}(\mathfrak{R})\geq\mathcal{R}(\mathbb{R})|\mathnormal{R})-P(\mathcal{R}(\mathfrak{R})\geq\mathcal{R}(\mathbb{R}))|<\delta} ), then if a robot passes the Turing Test, he will in theory not be detected as a robot spy when sent to Russia.

Objections and Replies

1. The Red Robot Objection states that robots, by their very nature, are communists. The way a robot's thread scheduler works is based on the principles of "from each according to his ability, to each according to his need" and "fair share". A robot who himself operates on communist principles would obviously defect to the USSR.

This objection confuses personal resource allocation with global allocation. Any rightminded capitalist must himself decide when to spend time at work, when to spend time with his lovely wife, and when to hang out with the boys, and in fact only a communist would give full priority to work and nothing else.

2. The Three Laws Objection states that as all robots are programmed to obey the Three Laws of Robotics (signed into law in 1942), any robot asked to by the Russian government would immediately betray the US.

This objection fails to take into account the Zeroth Law amendment, which places the interest of the whole of humanity above the interests of a single human. As the downfall of the USSR is in the interest of the whole of humanity, no robot would do anything to hurt the cause.

3. The Lady Lovelace Objection, one of the most famous objections, states that the Turing Test is nothing but a parlor game, and serious mathematicians and roboticists shouldn't concern themselves with parlor games.

The Mary Shelley Counter-Objection states that parlor games have been responsible for most of the major advances in robotics, and whiny little brat girls should get over the fact that their fathers prefer the company of poet friends and writer drinking buddies over mathematician wives and programmer daughters.

4. The Argument from Consciousness Objection states that, as we have no way of knowing if a robot is self-conscious or not, we must rely entirely on our observations of its behavior to determine what it is thinking. And thus, a robot who can pass as a Russian is a Russian.

This objection was refuted by John Searle in an experiment he called "The Russian Box". In this experiment, he locked himself in a box with a Russian phrase book, and had the box submitted to a Turing Test. Throughout the day, whenever the Russianologist asked him a question in Russian, he looked it up in the book and responded with a canned answer. He did this so convincingly that he passed the Turing Test. When he popped out of the box after the test, he admitted that he did not actually know Russian, nor was he a communist. He argued that a robot locked in the box with the phrase book would also be able to pass the test even if he did not know Russian and was not a communist.

Variations and Alternatives

The Big Brother Turing Test consists of a set of n Russians and a single robot living together in a house. At the end of each day, each Russian reports in a confessional which of his housemates he believes to be a robot, and why. While this doesn't get good mathematical bounds on the robot's ability to blend in, it gives programmers important information on what Russians look for in interactions with other Russians. Also, it is very entertaining.

The Robot Mafia Turing Test is a experiment modeling a battle between an informed minority and an uninformed majority, usually played by five or more agents. Most of the agents are Russians, while a few of them, greatly outnumbered, are robots posing as Russians, and only the robots know who the robots are. During the day cycle of the game, the group as a whole tries to pick out a robot for elimination. During the night cycle, the robots decide on a Russian to eliminate. The game ends when either all the robots have been eliminated (they fail the Robot Mafia Turing Test) or the robots eliminate enough Russians to gain a voting majority (they pass the Robot Mafia Turing Test). These tests were generally performed to determine the minimum number of robot spies that would need to be deployed on a mission.

The Minimum Intelligent Signal Test tests a robot's ability to pose as a Russian Blue. The Russian Blue problem was posed as a toy sub-problem of the Russian Spy Robot problem; if we could not fool their apathetic pets, we can not fool them. (This test was almost always performed Big Brother style with other cats as judges, as catologists didn't tend to get along with roboticists and Russianologists, and no one liked having them around.)

The Alan Turing Test was developed by Stephen Cole Kleene in 1962 in an attempt to create a robot that was indistinguishable from Alan Turing, in the hopes that such a robot would be able to fill the hole left in the robotics community after Alan Turing died in 1954. The test was itself not well-defined as there was no living Alan Turing to compare the performance of an A.T. Robot against; most instances of the test took the form of the robot hanging out with Alan's old friends for a day, after which they'd assign it a Baysian belief level for how well it would do if pitted against the real man in a proper Alan Turing Test setting. The most successful A.T. Robot, Mathison, killed himself shortly after passing the test.

Robot Spies

In this modern age, when all espionage is fully automated, it's hard to imagine a time when robot spies were crude and experimental. But just over half a century ago, intelligence was mostly gathered by hand, and the idea of an army of robot spies autonomously gathering intelligence for their government was nothing more than the musings of pulp fiction Spy-Fi authors.

History

That's not to say that, before this modern era, robotics and espionage simply did not mix. On the contrary, the two share a common history that goes back as far as 1286 BC. During the Trojan War, Odysseus commissioned roboticist Epeius to create a mechanical horse that could sneak into Troy and gather intelligence about the city's defenses. This "Trojan Horse" was not a true robot in the modern sense - he lacked intelligence, autonomy, and even an onboard computer. He was remotely controlled through a system of rope, pulleys, and magnets running along tubes planted under Troy by the Greeks. However, he is often considered the prototypical robot spy, and his control system, a prototypical internet.

April 18, 1775 marked the first use of robot reconnaissance by American forces. The robotic sentry Robot Newman kept close watch on the British Regulars and, once their movements were known, sent an encoded message across the Charles River to the waiting Paul Revere, who would take the message to Concord to be decoded. (The encryption system used could, at the time, only be understood by robots; it has since been cracked.)

The event was later recorded in a famous poem by Henry Wadsworth Longfellow.

Listen my children and you shall hear

Of the Cypher Text of Paul Revere,

He said to his 'bot, "If the British march

By land or sea, you test with a ping,

Hang a string of lanterns aloft in the arch

Of the church tower as ascii binary string,--

01100010011110010010000001101100011000010110111001100100

if by land, and

011000100111100100100000011100110110010101100001

if by sea;

Notes to self while writing this

Wikipedia links:

• Wikipedia:Allen Welsh Dulles
• Wikipedia:Turing test
• Wikipedia:Thought experiment
• Wikipedia:Cold War
• Wikipedia:Central Intelligence Agency
• Wikipedia:Spy fiction
• Wikipedia:
• Wikipedia:
• Wikipedia:
• Wikipedia:
• Wikipedia:
• Wikipedia:

Other links

http://www.archives.gov/publications/prologue/2002/spring/fritz-kolbe-3.html

Things to do

• um, finish writing
• more images!