From Ground to Sky Now – Lifelong Learning in AI

Lifelong learning in the AI era

As our shareholders arrived to our general meeting to appoint a new board member, I had a chat with one of our original team members about self driving cars. Just a month ago I had a chance to see Google’s Waymo and a couple of other autonomous driving initiatives in Silicon Valley. 

The Waymo is a formidable sensor carriage, with moving Lidar and ultrasound sensors plus cameras protruding to all directions. It moves almost majestically, carefully – and slowly. It’s trunk (the one prototype I saw) houses a rack with two datacenter-grade servers and network switches, along with a jungle of Ethernet cables. I would guess that the computing side draws some 1-2 kilowatts of electricity while the servers are loaded as the car analyzes the sensor data. With only a glimpse to the network gear, I’d say that the data bandwidth is in the gigabit range or more. 

Now back to the general meeting discussion. As usual, I told the crowd about some of the latest BaseN developments, such as the maiden flight of the spimed nDrone. Shareholders quickly pointed out that the autonomous nDrone is not much different a control challenge than a self-driving car; in a way it has more forgiving environment while in the air, compared to the car which has obstacles in most directions. Of course, a software crash, possibly by a static electricity discharge in the primary flight controller quickly reduces it to a pile of electronic waste when it drops like a stone.  

The nDrone maiden flight had some resemblance to the Waymo I saw on the Californian parking lot – slow, careful movement, abrupt stops. Both were moving like.. a toddler taking first steps. 

Now what is the main difference to a human-controlled drone or car? Yes, as one of our shareholders put it well: The autonomous nDrone lacks the combined life experience of the controlling human. A 1-year-old human is no driver nor a drone controller, even though the basic brain structure is already in place. 

This is why BaseN has brought the spime (and their containers) into existence. Any complex, AI-like function needs that full lifecycle experience, permanently stored and available to be used in present tasks. In this way, no algorithmic wisdom has to be re-invented at every technological generation. 

The spime is primarily the paradigm change; a counter force against the blind mass-production of physical and software assets. We must stop littering the world with obsolete things and bad code. This is something that can finally be called real AI. 

Missed the last nBlog? Read it here.

//Pasi 

Leave a Reply

Your email address will not be published. Required fields are marked *

More to explore