Question of the Month: How will AI be deployed in embedded applications in five years’ time?

e14 Question of the Month

Artificial Intelligence has the potential to change many of the things we do. Take the poll and let us know how you think AI will be deployed in embedded applications, and please tell us why in the Comments section below!

Top Replies

DAB over 2 years ago +3

The problem with using AI with sensors is that you do not have control over the result. If you cannot predict what the output will be, you have a high risk system, which should be avoided for anything…
bradfordmiller over 2 years ago in reply to dougw +3

I think it's essentially "applied statistics"; a turn on prior work on data mining. I did research in symbolic reasoning for about 30 years, and while I think learning is essential, what's being called…
dougw over 2 years ago in reply to bradfordmiller +3

Yes. Unfortunately the tech crowd want their software to seem snazzy, so they call it AI. Probably not an ideal image because the average person doesn't think AI is snazzy, they just think it is scary…

DAB over 2 years ago

The problem with using AI with sensors is that you do not have control over the result.

If you cannot predict what the output will be, you have a high risk system, which should be avoided for anything involving humans.
Cancel
Vote Up +3 Vote Down

Sign in to reply

Cancel
dougw over 2 years ago

General purpose processors will increasingly include AI.

There are a lot of big applications for AI in the cloud but there are far, far more small processors at the edge.
Cancel
Vote Up +1 Vote Down

Sign in to reply

Cancel
dougw over 2 years ago in reply to DAB

What they call AI these days isn't really what AI was originally conceived to represent. The concept has mostly been dumbed down to practical and specific correlations found in limited datasets.
Cancel
Vote Up +2 Vote Down

Sign in to reply

Cancel
shabaz over 2 years ago

All of the above. There are already sensors that use AI, they give good results therefore they are being used. Even financial and government orgs use them, they are trusted (they are vetted just like any other sensor would). As Doug mentions, the AI performs a specific function (and does it very well otherwise people would be sticking to older systems). It would be a surprise if AI usage didn't increase.
Cancel
Vote Up +1 Vote Down

Sign in to reply

Cancel
Gough Lui over 2 years ago

I think it will likely be a distributed mix of cloud (for non-latency sensitive, large models), edge (for latency-sensitive, medium-sized, privacy-concerned models) and on-device (for small, data-reduction, privacy preserving models to reduce bandwidth and transmission costs). No one-size-fits-all, but all aspects are under active development at this time.

Dedicated AI processors will take some time to become mainstream, and while existing general purpose processors may be able to run small models relatively slowly, I think the move to AI processors will mainly be made for performance and energy efficiency reasons (but in turn, may introduce hardware-related model size and precision constraints).

- Gough
Cancel
Vote Up +1 Vote Down

Sign in to reply

Cancel
beacon_dave over 2 years ago in reply to dougw

AGI appears to be the latest buzzword. The 'G' being for 'General'.
Cancel
Vote Up +1 Vote Down

Sign in to reply

Cancel
bradfordmiller over 2 years ago in reply to dougw

I think it's essentially "applied statistics"; a turn on prior work on data mining. I did research in symbolic reasoning for about 30 years, and while I think learning is essential, what's being called "learning" is what we used to call "adaptation". It's useful, of course, but not really representative of human or even animal intelligence.
Cancel
Vote Up +3 Vote Down

Sign in to reply

Cancel
bradfordmiller over 2 years ago

My expectation is that AI will have more of an effect in the tools used to build embedded applications than in the applications themselves.
Cancel
Vote Up +2 Vote Down

Sign in to reply

Cancel
dougw over 2 years ago

Clearly one of the ubiquitous places you will find AI is in smart phones, but it won't be clear whether it is in the phone's processor or in the cloud or a hybrid of both.
Cancel
Vote Up +1 Vote Down

Sign in to reply

Cancel
dougw over 2 years ago in reply to bradfordmiller

Yes. Unfortunately the tech crowd want their software to seem snazzy, so they call it AI. Probably not an ideal image because the average person doesn't think AI is snazzy, they just think it is scary.
Cancel
Vote Up +3 Vote Down

Sign in to reply

Cancel