New Mathematical Formula Unveiled to Prevent AI From Making Unethical Decisions

[Michie]

Researchers from the UK and Switzerland have found a mathematical means of helping regulators and business police Artificial Intelligence systems’ biases towards making unethical, and potentially very costly and damaging choices.

The collaborators from the University of Warwick, Imperial College London, and EPFL – Lausanne, along with the strategy firm Sciteb Ltd, believe that in an environment in which decisions are increasingly made without human intervention, there is a very strong incentive to know under what circumstances AI systems might adopt an unethical strategy—and to find and reduce that risk, or eliminate entirely, if possible.

Artificial intelligence (AI) is increasingly deployed in commercial situations. Consider for example using AI to set prices of insurance products to be sold to a particular customer. There are legitimate reasons for setting different prices for different people, but it may also be more profitable to make certain decisions that end up hurting the company.

Continued below.
New Mathematical Formula Unveiled to Prevent AI From Making Unethical Decisions

 

[jacks]

W=Y then Proceed.
W=N then halt.

W= Would Jesus Do it?
Y=Yes
N=N

 

[GenemZ]

Researchers from the UK and Switzerland have found a mathematical means of helping regulators and business police Artificial Intelligence systems’ biases towards making unethical, and potentially very costly and damaging choices.

The collaborators from the University of Warwick, Imperial College London, and EPFL – Lausanne, along with the strategy firm Sciteb Ltd, believe that in an environment in which decisions are increasingly made without human intervention, there is a very strong incentive to know under what circumstances AI systems might adopt an unethical strategy—and to find and reduce that risk, or eliminate entirely, if possible.

Artificial intelligence (AI) is increasingly deployed in commercial situations. Consider for example using AI to set prices of insurance products to be sold to a particular customer. There are legitimate reasons for setting different prices for different people, but it may also be more profitable to make certain decisions that end up hurting the company.

Continued below.
New Mathematical Formula Unveiled to Prevent AI From Making Unethical Decisions

There is no such thing. If someone programming has an unquestioned bias, it will result in the AI being like its maker.

 

[Caliban]

W=Y then Proceed.
W=N then halt.

W= Would Jesus Do it?
Y=Yes
N=N

What would Jesus do? = The robot starts flipping over tables, whipping people, and destroying trees.

 

[sjastro]

This looks like another application of Game Theory which I remember doing as a maths undergraduate many years ago.
Game theory's greatest success was the postponement of World War III (hopefully permanently).

 

[DebbieJ]

I prefer an AI to commit mistakes and learn from it so it will not do it again. Like touching a hot surface. It can only handle a certain temperature before damage occurs. When damage occurs, the AI will learn from it and avoid such temperature.

 

[Quid est Veritas?]

I find it intriguing that "Optimization can be expected to choose disproportionately many unethical strategies" as the researchers state. So essentially, AI seems to prefer the unethical; or to put another way, AI is evil. I don't know, I think I've seen this movie...

 

[FrumiousBandersnatch]

I find it intriguing that "Optimization can be expected to choose disproportionately many unethical strategies" as the researchers state. So essentially, AI seems to prefer the unethical; or to put another way, AI is evil. I don't know, I think I've seen this movie...

It's probably just a matter of statistics; that there are many more ways to be unethical than to be ethical (the narrow path) and/or that ethical solutions are less likely to be optimal. A simple AI will tend to favour optimal solutions.

 

[GenemZ]

Man is trying to create an artificial God that will clean man's hands of accountability..

 

[sjastro]

Since Game Theory is involved, here is a simple example of whether ethical or unethical decision making occurs as illustrated in the following game.
It’s known as the Prisoner Dilemma and was conjured up by Game Theorists in 1950.

Suppose there are two prisoners, call then Prisoner 1 and Prisoner 2 held in remand in isolated cells incapable of communication with each other.
The prisoners have committed a crime and are presented with 4 options.
Option (A): Both prisoners implicate each other resulting in a 1 year sentence for each.
Option (B): Prisoner (1) remains silent about his crime but implicates Prisoner (2); Prisoner (1) goes free and Prisoner (2) receives a 5 year sentence.
Option (C): Prisoner (2) remains silent about his crime but implicates Prisoner (1); Prisoner (2) goes free and Prisoner (1) receives a 5 year sentence.
Option (D): Both Prisoners remain silent and each receives a six month sentence.

The options can be expressed as Game Theorists term a payoff matrix.

Which is the favourable option?
In terms of individual outcomes Options (B) and (C) might be considered but it is highly likely each prisoner will implicate the other and both will receive a 5 year sentence.
Options (B) and (C) are effectively eliminated leaving option (D) as the best outcome.

If a human and AI Game Theorist analyse the Prisoner Dilemma both should come to the same conclusion.
The human Game Theorist might consider the prisoners have been rewarded for not confessing in which case Option (A) is ethically the best decision.
The AI Game Theorist on the other hand purely analyses for the best outcome which is Option (D).

This raises an interesting subject where defense lawyers are the human analogue of the AI Game Theorist; their objective is to get the best outcome for their clients despite examples of overwhelming evidence and the heinous nature of the crime.
Is this an unethical activity?

 

[FrumiousBandersnatch]

This raises an interesting subject where defense lawyers are the human analogue of the AI Game Theorist; their objective is to get the best outcome for their clients despite examples of overwhelming evidence and the heinous nature of the crime.
Is this an unethical activity?

It's adversarial - the prosecution lawyers are similarly vested in ensuring a guilty verdict despite examples of a paucity of evidence - is it doubly unethical, or do they cancel out in the long term?

 

[sjastro]

It's adversarial - the prosecution lawyers are similarly vested in ensuring a guilty verdict despite examples of a paucity of evidence - is it doubly unethical, or do they cancel out in the long term?

It doesn't have to be adversarial where total victory is the objective; defense and prosecution lawyers engage in plea bargains which is a compromise.
This is an example of a cooperative game.

I think the question remains unanswered as the subject is still a work in progress.
Game Theory - Its Applications to Ethical Decision Making.

The application of game theory according to Hargreaves-Heap and Varonfakis (1995) to understand human behaviour, and in particular ethical behaviour, is a valuable development, as game theory has gradually become one of the key frameworks to assist us in the understanding of social sciences. Esther (1982) and Aumann and Hart (1992) show that there are several studies that indicate the importance of a game theoretic framework in advancing our understanding of social behaviour and evolutionary sciences. Although the application of game theory in the above areas has largely been not formalised, its application in the fields of ethical conduct and human behaviour is at present developed in several respects with the gradual assistance of advances in related areas such as evolutionary biology and our understanding of group social behaviour.

 

[sjastro]

In the 1950s Game Theorists developed the Madman Theory where if leaders portrayed themselves as irrational maniacs this would provide a disincentive for rival powers to engage in provocative action fearing retaliation.
Richard Nixon used this behaviour in his foreign policy and given Henry Kissinger was very much into Game Theory it could be mathematicians were behind this (warning possible conspiracy theory).

It might also explain the motives behind Nikita Khrushchev's academy award winning performance at the 1960 UN conference.

Even Boris Johnson's Brexit strategy may have been influenced by Game Theory.
Boris Johnson and game theory - Oxbridge Applications

Perhaps Donald Trumps irrational, contradictory behaviour is a masterful example of Game Theory in action; although I doubt it.

 

[stevevw]

Researchers from the UK and Switzerland have found a mathematical means of helping regulators and business police Artificial Intelligence systems’ biases towards making unethical, and potentially very costly and damaging choices.

The collaborators from the University of Warwick, Imperial College London, and EPFL – Lausanne, along with the strategy firm Sciteb Ltd, believe that in an environment in which decisions are increasingly made without human intervention, there is a very strong incentive to know under what circumstances AI systems might adopt an unethical strategy—and to find and reduce that risk, or eliminate entirely, if possible.

Artificial intelligence (AI) is increasingly deployed in commercial situations. Consider for example using AI to set prices of insurance products to be sold to a particular customer. There are legitimate reasons for setting different prices for different people, but it may also be more profitable to make certain decisions that end up hurting the company.

Continued below.
New Mathematical Formula Unveiled to Prevent AI From Making Unethical Decisions

It would be impossible to make ethical decisions when it comes to making commercial ones. They are incompatible. If they make ethical decisions then it becomes costly for the bottom line. Any successful company today has made unethical decisions in putting profits before people. We would have to change the entire economic system from a capitalist one which I doubt people will allow.

I wouldn't be surprised that the whole idea of a mathematical equation for ethical decision making is designed to be sold and to make profits itself. Otherwise, why do it? They surely would not be spending all that money and time to just give the idea away to help others.