Experts say classroom-style learning could become the norm for programming ethical, trustworthy robots.

Australian researchers have proposed a novel approach to the programming of systems powered by AI.

“Machine education is possibly our only practically sound hope to truly design ethical, trustworthy, transparent and reliable AI systems,” says  UNSW Canberra researcher Professor Hussein Abbass.

Machine education proposes a more accessible approach to unlocking the power of AI. 

It is seen as a better method than the current attitude to machine learning, which prioritises technical know-how and exclusivity in the undertaking of AI programming.

The combination of the machine education approach, together with workflows enabling two-way communication, could equip AI with the skills required to conduct ethical reasoning and logic-based synergy with their human peers.

It may also help address concerns of trust, fear, and performance expectancy - some of the biggest barriers to AI adoption across fields such as health care, manufacturing and autonomous vehicle sectors. 

AI programming and machine learning has largely been approached through a data-driven lens up until now, according to UNSW Canberra Associate Professor Sondoss Elsawah.

“Machine learning so far has just meant ‘data in, data out’,” A/Prof Elsawah said.

“However, we need to move away from the idea of AI as a ‘black box’ in order to form mutual relationships between systems and users.”

By breaking down AI-enabled autonomous systems’ desired task outputs into sets and subsets of skills, similar to a university course curriculum, the researchers propose that they can educate AI in a classroom-style fashion.

This aims to ensure autonomous systems possess a sufficiently sophisticated knowledge base to pass tests and assessments, becoming qualified to work alongside domain experts.

By embedding a structured and systematic curriculum of content and knowledge within these systems, as well as the building blocks required to understand ethics, moral values, safety, and trust, these systems may be able to more successfully socially integrate into modern working conditions.

More details are accessible in the paper, ‘A model of symbiomemesis: machine education and communication as pillars for human-autonomy symbiosis’.