The road to Edge AI Computing

OK, I already missed the second day of my everyday blog-post challenge about “AI in Europe”. But here we go:

The world of Artificial Intelligence has many components. I believe the popular image of this world is a bunch of chatbots, large corporations and big data-centers.. I want to show people a much wilder image of AI. Not only what it could be, but what it already is!

The Chatbots and the GPTs

The GPTs ¹ are the engine of most of the chatbots and other text-based applications. Us ML, Deeplearning and AI fans were following the progress on the sideline, seeing them evolve from GPT1 to GPT2, to GPT3, from which ChatGPT were born, and beyond. The GPT models were enabled by an exponentially growing data-set of digitalized books, articles, music and videos. Alongside a rapid development of GPUs and dedicated versions of them called TPUs. Parallel to this we have been following a similar emergence of intelligence.

The Everywhere Computer

All around us computers have been inserted into products that used to be thought complete without further intelligence. Smart home devices like TVs, Fridges, Lawn Mowers, teddy-bears and light bulbs are just some of them. In the public we have: Light posts, traffic lights, sensors for weather applications, railway-switches and much more.

Connected to the internet, this, in some way, becomes one computer that is everywhere. The Everywhere Computer.

The Everywhere Computer does not have single mind of it own. It is scattered and fragmented - from a smart factory with a local network controlling the robots inside, to a municipal light post collecting traffic data, to an amateur plane watcher with a Raspberry Pi logging flight data and sharing it online. It has no apparent brain.

At first glance, the data that the Everywhere Computer seems as incomprehensible machine data. But many places, like the website Danish Meterological Institute ², you can go and have this data presented in very human-friendly form. This is very powerful.

This power enabled by data from many smaller machines being transferred to central computer systems for insights (this dmi.dk) and control (think trafic-lights), is the enormous field of IoT, Internet-of-Things.

Combining IoT and AI

IoT very much parallels the foundation that the GPT-models where built upon. Massive amounts of data, that a central system tries to make sense of. However, where GPTs first data was PDFs, YouTube-videos and news-articles and in general being web-pages readable by humans, the data for the Everywhere Computer is machine data from the IoT. The Chatbots are getting upgrades. They are now able to search the web faster than you ever could. They can read your calendar, if you let it, fortunately. It can not turn on your coffee machine or the traffic lights as a part of superficial prompt you made. But it is not difficult to imagine it. The Everywhere Computer gets a brain. This is a bridge that is happening as i write this. And when this bridge is paved, the implications are huge.

EU - The next generation of IoT

The European Union are fortunately working on how to tackle the bridging of AI and IoT³. The next big thing is Edge Computing, which is getting compute power from the cloud to the machines themselves before sending data to a central place. Putting more compute power will naturally evolve into also putting AI into the machines themselves for faster and local decision making and processing of data. We call this Edge AI. Edge computing and Edge AI is also seen as a possibility to meet privacy concerns and still get the benefits of IoT and AI.

I will explore both Edge AI, privacy and EU’s policy and strategies for AI IoT in the coming blogpost.