Now the AI technology is established at a basic level, 2018 promises
to be a seminal year, a year in which the industry will race to establish the
standards that could determine the entire future of the technology and
what companies benefit the most.
The first trend to watch is the transition from training to doing, a
threshold more and more machines are reaching. That is, from teaching
machines how to think to seeing them actually accomplish tasks. The
catalyst for this will be new chips, the so-called specifc integrated
circuits, due to be released this year.
That is important as different types
of computer processors are useful for learning and doing. For example,
training and deep learning functions are best accomplished by graphics
processing units, such as those made by Nvidia. These are units that were
originally designed to make video games look better and are able to
process diferent types of calculations at the same time. In a test between
regular computer processors and graphics processing units (GPUs), the
latter was able to learn how to recognise people by watching films for just
one per cent of the cost of regular processors.
That has made GPUs extremely important for the current phase
of artifcial intelligence research which involves the use of large amounts
of data to give machines the information to make decisions. The next
step, however, involves translating those decisions into action. That
requires a different type of computing power and GPUs may be less
important than the specifc integrated circuits which are made by the likes
of Google and Intel. As the shift towards implementing learnt knowledge
becomes more widespread, so too will the need for these specific
integrated circuits.
EU Forecast
euf:b18:96/nws-01