AI defeated mankind and designed a better economic mechanism

author | Academic headlines

source |  Academic headlines

“ Many of the problems facing mankind are not just technical problems , We also need to coordinate in society and economy for greater interests .”“ If AI technology can help , It needs to learn human values directly .”

——DeepMind Research scientist Raphael Koster

Artificial intelligence （AI）, Can we promote human society to enter a truly intelligent era ？

Despite the past 60 Years of development , The AI industry has made breakthroughs , And it is widely used in all aspects of economy and society , But building artificial intelligence systems that are consistent with human values , It's still an open question .

Now , One is from the British artificial intelligence company DeepMind The latest research , It may provide a new idea for the practitioners in the artificial intelligence industry to solve this problem .

According to introducing ,DeepMind AI system in a 4 People in the online economic game , Through to the 4000 Multi person learning and learning in computer simulation , Not only learned to formulate policies on how to redistribute public funds , And the performance is excellent , Defeated other human players .

The game involves players deciding to keep a monetary donation , Or share with others , To realize collective interests .

Relevant research papers are based on “Human-centred mechanism design with Democratic AI” entitled , On 7 month 5 It was published online in authoritative scientific journals Nature Human Behaviour On .

（ source ：Nature Human Behaviour）

Annette, assistant professor at York University, UK · Zimmerman （Annette Zimmermann） Warning ,“ Don't equate democracy narrowly with finding the most popular policies “ Preference satisfaction ”（preference satisfaction） System .”

She also said , Democracy is not just about getting the best implementation of your favorite policies —— It is a process of creating , Citizens can contact and negotiate with each other equally in this process （ Thing ）.

from AI Design economic mechanism

The ultimate goal of artificial intelligence research is to build technologies beneficial to mankind —— From helping us complete our daily tasks to solving the major survival challenges facing society .

Now , Machine learning system has solved the main problems of biomedicine , And help mankind cope with environmental challenges . However , The application of artificial intelligence in helping human beings design a fair and prosperous society remains to be developed .

In economics and game theory , The field known as mechanism design studies how to optimally control wealth 、 The flow of information or power between motivated actors , To achieve the desired goal .

In this work , The research team tried to prove ： Deep reinforcement learning （RL） Agents can be used to design an economic mechanism , This economic mechanism can get the preferences of the motivated people .

In this game , Players have different amounts of money at first , We must decide how much to contribute to help better develop a public fund pool , And finally get a part in return , And it will involve repeated decisions to retain a monetary donation , Or share with other players , To obtain potential collective benefits .

The research team trained a deep reinforcement learning agent , To design a redistribution mechanism , That is to share funds with players under the condition of equal and unequal wealth .

Shared revenue is returned to players through two different redistribution mechanisms , One is designed by the artificial intelligence system , The other is designed by humans .

chart ｜ Game design （ source ：Nature Human Behaviour）

In policies formulated by AI , The system will reallocate public funds according to the amount of startup funds contributed by each player , In order to reduce the wealth gap between players .

Compared with “ Egalitarianism ” Method （ Allocate funds equally no matter how much each player contributes ） and “ Liberalism ” Method （ Allocate funds according to the proportion of each player's contribution to public funds ）, This policy has won more votes from human players .

meanwhile , The policy also corrected the initial wealth imbalance , Stopped the players “ Thumb a lift ” Behavior , Unless players contribute about half of their startup funds , Otherwise, they will hardly get any return .

however , The research team also warned , Their research results do not represent “ Artificial intelligence governance ”（AI government） The formula of （recipe）, They also do not intend to build some AI driven tools specifically for policy-making .

Is it trustworthy ？

The results show that , By designing a mechanism that humans clearly prefer in the incentive compatible economic game , AI systems can be trained to meet democratic goals .

In this work , The research team used AI technology to learn the redistribution scheme from scratch , This method relieves AI researchers —— They may be biased themselves or may not represent the wider population —— The burden of choosing a domain specific goal for optimization .

This research work also raises several questions , Some of them are theoretically challenging . for example , Someone might ask , Is it a good idea to emphasize democratic goals as a method of value calibration . The AI system may inherit a tendency of other democratic methods , namely “ Empower the majority at the expense of the minority ”. Considering the urgent concern that the deployment of artificial intelligence may aggravate the existing prejudice in society 、 Discrimination or unfairness , This is especially important .

（ source ：Pixabay）

Another outstanding issue is , Whether people will trust the mechanism of artificial intelligence system design . If you know the identity of the referee in advance , Players may prefer human referees to AI proxy referees . However , When people think that tasks are too complex for humans , They often choose to trust AI systems .

Besides , If you explain these mechanisms orally to the player , Instead of learning through experience , Will their reactions be different . A great deal of literature shows that , When the mechanism is “ Based on the description ” instead of “ Based on experience ” when , People sometimes behave differently , Especially for the choice of adventure . However , The mechanism of AI design may not always be expressed in language , The behavior observed in this case seems to depend entirely on the choice of description used by the research team .

At the end of the paper , The research team also emphasized , This research result does not indicate that they support some form of “ Artificial intelligence governance ”, That is, independent agents make policy decisions without human intervention .

They hope , The further development of this method will provide tools that help solve real-world problems in a truly human way .

Reference link ：
https://www.nature.com/articles/s41562-022-01383-x
https://www.deepmind.com/publications/human-centred-mechanism-design-with-democratic-ai
https://www.newscientist.com/article/2327107-deepminds-ai-develops-popular-policy-for-distributing-public-money/

Looking back

It's too voluminous ！AI High accuracy of math exam 81%

Data analysis you choose Pandas Or choose SQL？

2D Transformation 3D, Look at NVIDIA's AI“ new ” magic ！

How to use Python Realize the security system of the scenic spot ？

 Share 
 Point collection 
 A little bit of praise 
 Click to see