Artificial intelligence (AI) is a quickly growing industry.
While this technology has the potential to become a crucial aspect of education, AI systems have yet to become truly ‘mainstream’.
However, all of that is changing as breakthroughs become more common and their effects more widespread.
From training doctors to providing students and employees with personalised education plans, AI is an increasingly important part of society.
Before AI systems can be used to their full potential, they must be ‘taught’.
This includes giving them the information needed to master the niche in question, whether that means a specific field of science or a specific hobby or game (such as chess).
AI systems are excellent at mastering a specific industry or genre.
Teaching it a secondary industry or genre, however, has proven to be unexpectedly difficult.
As AI systems are ‘taught’ about a second industry, the new knowledge takes the place of the old.
Eventually, the memories of the initial area of mastery fade completely.
Scientists have long searched for a way to improve AI learning, and now a team of four researchers has made a breakthrough.
Teaching AI systems to ‘sleep’ boosts memory retention
Consisting of three scientists from the University of California and one from the Institute of Computer Science of the Czech Academy of Sciences, this four-researcher team has come up with a smart way to help AIs.
In a paper published in the journal PLOS Computation Biology, Jean Erik Delanois, Ryan Golden, Pavel Sanda, and Maxim Bazhenov describe their experiences with memory loss – also known as ‘catastrophic forgetting’ – in AI systems.
The team explained that while they initially struggled to teach systems a new ‘niche’, they looked to the human brain for inspiration.
Humans sleep for a number of different reasons – one of the most important of these is giving the brain time to consolidate memory.
Memory consolidation occurs during REM sleep and helps the brain commit knowledge and experiences to long-term memory storage in order for new information to be learned.
By applying the concept of memory consolidation to AI systems, the researchers found that teaching systems multiple areas of expertise is possible.
More specifically, the team wrote code that mimics REM sleep in the human brain.
AI systems were essentially able to combine periods of work and periods of sleep.
The periods of sleep gave them time to sort out new memories from old ones while retaining the latter.
Today’s news was brought to you by TD SYNNEX – the UK’s number one solutions distributor.
