The brain can process high volumes of data faster than any single computer. It also uses a surprisingly low amount of power. A shortcut to replicating these features may come sooner than expected, thanks to “neuromorphic” chips envisaged by Samsung and Harvard researchers.
It is the fourth blue sky idea Lex examines in today’s thematic column.
Such human-level artificial intelligence would have broad applications. In medicine, for example, where it could provide diagnoses. The biggest challenge has been training machines to process information in a “brain-like” manner. Computers are linear and methodical. Brains are lateral and make sudden leaps.
AI researchers try to get round this by feeding computers massive amounts of data. Programmes are only as good as this information. The limited amount and quality of available data has often sent human-like AI down the wrong path, prompting them into illogicalities of expression and offensive opinions.
It has also been difficult for current AI models to make their own decisions. They have trouble prioritising data, such as the information needed to learn a certain function more efficiently.
The brain, in contrast, absorbs new concepts without explicit instruction and often reach intuitive conclusions that reduce the need for background data.
Samsung, working with Harvard researchers, is seeking to overcome these challenges with a scheme for “copying and pasting” the brain’s over 100bn neurons and their connections to new types of 3D “neuromorphic” chips. These are supposed to replicate the brain’s complex interconnections of neurons and synapses, overcoming the limitations of conventional microprocessors.
But allocating chip manufacturing production to support this blue sky research may be difficult amid shortages of normal semiconductors. Moreover, advanced AI is fraught with ethical and legal issues. These may be tricky for Samsung — a company focused on engineering and manufacturing — to navigate.
The swing factor, however, will be humanity’s limitless appetite for data processing capacity. Replicating human thought will remain difficult given that scientists are struggling to understand it. But chips that came anywhere near doing this would have a huge potential market.
This is the fourth article on blue sky thinking published by Lex today. Look out for the others in Lex online.