Community Research and Development Information Service - CORDIS

How machines can learn from human behaviour

Contributed by: youris.com EEIG

Designing intelligent machines that can resemble and model human behaviour is closer than we think. And you can join in
How machines can learn from human behaviour
Could a human behaviour simulator be embedded into a robot or online avatar to the point that it’s hard to distinguish between a real person or artificial intelligence? Scientists have been upping the stakes in this “Turing test” for years, to the point that human-mimicking programmes are ready to answer tricky questions, assist people with online shopping or be companions.

Now researchers across a clutch of European universities have developed a large-scale experimental lab programme involving over 22,000 people to investigate the socio-economic problems that arise from human-computer interactions. At the European IBSEN project, the researchers’ goal is to provide a breakthrough towards building a future human behaviour simulator, a technology that could impact fields from robotics to economics, and offer new instruments to policy makers.

According to project coordinator Anxo Sanchez, professor of Applied Mathematics at Universidad Carlos III de Madrid, Spain, the advance needed is the capability to do experiments with their actions. “Once you have an ample repertoire of behaviours, you could go for a simulator in which there are a number of computer agents that interact with each other with the rules you have, and therefore give rise to collective behaviour which should mimic that of society,” says Sanchez.

The researchers are looking at demonstration cases such as cooperation in social networks, where participants have to decide whether they want to collaborate with teammates toward some common goals, and how to foster and maintain that synergy when some are tempted to cheat and let the others do the work.

For example, in their experiments around 1000 people have been asked to decide how much money they wanted to give to a common pot, which will be equally shared among all participants, irrespective of their contribution. They then keep whatever they didn’t contribute, along with their part of the common pot.

The researchers realise that, after a few rounds, some contributors gave no money but benefited from the common pot, leading the others to reduce their sums and, eventually, not to donate anymore. However, in their large-scale experiments if people can't possibly track down everyone else’s contributions, this makes the whole scenario different, Sanchez explains.

These experiments reveal that human behaviour depends on the way participants are informed of the outcome of previous rounds. “This is letting us model how people in a cooperative situation like this behave depending on the information received, and gives hints as to how more contributions to the common good can be promoted.”

Researchers are also looking to predict how people react when looking at online avatars in the business environment, such as online buying and selling. For example, Creative Virtual, a London-based company that offers virtual assistant avatars for customer service, already provides organisations with the capability to embed ‘personality’ and ‘emotions’ into online chatbots. “We see people interact with chatbots for longer when they are represented by an avatar that contains our small-talk module,” says Chris Ezekiel, founder and CEO of the company. “We even see people build ‘relationships’ with them, and this type of behaviour will only increase when they are combined with robots.”

Read more: http://www.fetfx.eu/story/machines-can-learn-human-behaviour/

Contributor

Organisation

    youris.com EEIG
    Drève du Pressoir 38
    B-1190 Brussels
    Belgium
    Website

Contact

Related information

Keywords

machine learning, AI, robotics, technology, human-computer interaction
Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top