Learning together is better. For robots too.

Credit: ThinkStock

Collaborative learning is more fun and it is also more effective in many cases. But this does not just applies to humans (and several animals too). MIT researchers are making the case for robots too!

Their work has been presented in July 2014 at the Uncertainty in Artificial Intelligence Conference.

What they point out is that the effectiveness in analysing data, to take decision, centrally is lower than to analyse data in many different points (locally) and then exchanging the analyses to create a synthetic view.

They are applying this finding to robots by having each of them assessing the situation (and its evolution) from the local perspective and then exchanging the assessment to learn from others.

Machine learning has been developed over the last thirty years reaching the point where machines can look at trends over time and extract experience, reaching conclusions. This has made possibile significant advances in artificial intelligence and we can see the results every time we see Siri improving in understanding our voice and our questions ... This type o technology is being used by robots to learn more about their environment.

However, machine learning is working on a single entity, the one trying to learn. it gets quickly much more complex as knowledge is distributed and even though in principle harvesting a global knowledge can be useful in practice constraints on power, processing and storage do not support this sharing.
What the MIT researchers did was to invent a way for sharing pieces of knowledge and growing it from there individually. When two robots are nearby they can exchange information and this is processed by each to augment the local knowledge. It reminds me a bit of what happens in ants. Each ant is pretty simple, although each ant posses a good "operational" knowledge and can increase this knowledge by exchanging information with the ant it happens to pass by (and touch). this is sufficient to increase both the knowledge of the single ant and the behavioural knowledge of the ensemble.

I don't know if the researchers at the MIT took inspiration from ants, probably not, but reading their work made me think about how Nature manages to create very sophisticated knowledge by having it emerging from very basic one. The idea that the learning of individuals can result in the overall learning of the emergent behaviour is fascinating and I feel this is what we will see happening in Smart Spaces in the next decade. We have just started.

Author - Roberto Saracco

© 2010-2019 EIT Digital IVZW. All rights reserved. Legal notice. Privacy Policy.

EIT Digital supported by the EIT