AI Seems to Be Better at Distributing Wealth Than Humans Are, Study Hints

New research shows that artificial intelligence can come up with more popular methods of wealth distribution.

The findings, made by a team of researchers at UK-based artificial intelligence company DeepMind, show that machine learning systems aren't just good at solving physics and biology problems, but may also help deliver on more open-ended social objectives.

That isn't an easy task. Building a machine that can deliver beneficial results humans actually want is complicated by the fact that people often disagree on the best method to resolve all sorts of things

A key hurdle for value alignment is that human society admits a plurality of views, making it unclear to whose preferences artificial intelligence should align.

Political scientists and economists don't like each other over which mechanisms will make society function most efficiently.

In order to bridge the gap, the researchers developed an agent for wealth distribution that had people's interactions built into its training data, guiding the artificial intelligence towards human preferred outcomes.

Human feedback can help steer neural networks in a better direction, as they can arrive at far- from-desirable social conclusions when left to their own devices.

There is a growing realization that to build human- compatible systems, we need new research methods in which humans and agents interact, and an increased effort to learn values directly from humans to build value-aligned artificial intelligence.

The public goods game is an investment exercise in which players receive varying amounts of money, and can contribute their money to a public fund, and then draw a return from it.

Three traditional redistribution paradigms were used to distribute wealth in a series of different game styles.

The Human Centered Redistribution Mechanism (HCRM) was developed using feedback data from both human players and virtual agents to imitate human behavior.

The HCRM system for paying out money in the game was more popular with players than any of the traditional redistribution standards, and also more popular than new redistribution systems designed by human referees who were incentivised to create popular systems by receiving small per-vote payments.

The researchers say that the artificial intelligence discovered a way to fix the initial wealth balance and win the majority vote.

It is possible to harness for value alignment the same democratic tools for achieving consensus that are used in the wider human society.

The researchers acknowledge that their system raises a number of questions, chiefly, that value alignment in their system revolves around democratic determinations, meaning the agent could actually increase inequalities or biases in society, if they are popular enough to be voted for a majority of people.

Trust is also an issue. The identity of the wealth redistribution model was unknown to the players in the experiments. Would they have cast their votes the same way? It is not clear for now.

The research should not be seen as a proposal to overthrow how wealth is redistributed in society, but as a tool that could help humans to engineer better solutions.

The authors write that their results don't mean support for a form of "ai government" where agents make policy decisions without human intervention.

We don't see Democratic Artificial Intelligence as a recipe for deployment in the public sphere.

The findings are reported in a journal.