Earlier in the summer, the MWD Advisors team presented a free webinar entitled The Rise of the Machines – looking at “Bots, AI and the future of work”.
There’s a great deal of interest in this fast-moving space, and we weren’t able to cover all your questions in our live Q&A – so we decided to follow-up on some of the points in our blog to keep the discussion going.
One thing we were asked about was why, given that in a number of the use cases we talked about in the webinar, you could still get a pretty decent outcome relying solely on a traditional programmed system (as opposed to a machine-learning system) approach, why go to all the effort of configuring a learning solution at all?
In answer, it’s worth remembering that there two scenario types to consider.
Firstly, it’s true that some applications, in tightly controllable environments, can perform well by following programmed rules. This tends to be where automation takes place in ‘programmable environments’ where you fit your environment around the restrictions of a non-learning machine so that it can cope with what it is presented with and still give you good quality results. However these solutions can also be turbo-charged with correctly trained machine-learning agents that can enable applications to operate equally well within more ‘uncertain’ environments. The resulting systems can use what they learn about the context of a problem to reject, select and weight relevant data sources for themselves.
Secondly, we have scenarios where programmatical approaches never really cut it – where approaches such as deep learning algorithms are designed to analyse and interpret video, for example (in similar ways to how a human brain would attack the problem). Here, for machines to ‘watch videos for us’, the system attempts to watch videos like us. That’s not been possible without relatively recent advances in cheap-enough compute power, but now the economics are in our favour, essentially making possible what would have been unaffordable at speed and at scale in even the relatively recent past.
If you missed the webinar, you can easily access the replay using a free site account.
What’s more, in the coming weeks we’ll be publishing a report that goes into all these topics in more detail. Watch this space!