This past week seemed to follow an augmented reality theme in the regular places I read. Techmeme, A16Z podcast, and even Hacker News provided multiple references to augmented reality, virtual reality, 5G ubiquity, and the widening adoption of gaming technologies.
During recent trips on Lyft, the driver and passenger (me) conversations trended to technology discussions. I like to ask drivers if they are full time or part time with Lyft, Uber, or any other TNCs. One common theme was the prediction of when self-driving cars would be viable in any markets beyond denser and constantly mapped markets.
My anecdotes were on ideally selected AZ to NV route shipping trucks that have been developed for decreased reliance upon high cost (i.e. experienced) drivers in favor of 4G/LTE connected call center hosted drivers, aggressively funded learning models, and monitoring driver experiences.
My other anecdotal example was the controlled testing done in Las Vegas with Lyft.
Self-Driving - Lyft - Passengers
What happens in Vegas… eventually will happen everywhere else on a long enough timeline.
The driver anecdotes focused on the cost of insurance and liability for new technology in vehicles and the growing body of evidence that self-driving vehicles are less prone to accidents than human driving vehicles.
Perhaps we’ll see AR become the answer to all the problems we didn’t know we had. Getting human drivers comfortable with AR heads up display built into more dashboards and windshields might not be far away but seeing ourselves as jet fighter pilots is a flawed proposition. More distracted driving would be a bad thing. Instead, perhaps the AR for human drivers becomes the next advertising modality as we find the car driving itself and the human driver is merely along for the commercials played against the real world substrate passing them by.
AR Will Spark the Next Big Tech Platform—Call It Mirrorworld | WIRED
Or, perhaps, this can be the best test construct possible for the things we aren’t willing to unleash in the real world that could easily end badly?
Finding the right place to run a test will be a growing challenge as density of technology increases and ubiquity becomes more commonplace. The ideal routes will be harder to justify for longer term studies.
In response, machine learning against non-disruptive data sets will grow rapidly. Unfortunately, the demand to grow will have challenges. Just as there have been calls for increased emphasis on STEM and recruiting talent into emerging technology areas we will face the challenge of any new industrial endeavor: quality control and societal considerations of the result from the quality in our day to day lives.
AAAS: Machine learning ‘causing science crisis’ - BBC News
The search for a statistical outcome to match Little Britain’s “Computer says no”.
Let’s say a machine learning model renders the same result as thousands of years of humans that can be summarized as unethical, unrealistic, or unsatisfactory in some way but gets flagged for review. Let’s also say that the hunger for ML progress widens the pool of companies in pursuit of more data sets. It’s probably reasonable to assume some of the data sets will be flawed. It’s probably reasonable to assume some of the ML uses and applications will be even more flawed.
Now, if science is that which explains and predicts, then we have more work to do before we allow a model to control the behaviors of cars let alone drones capable of carrying around a piano over our heads.
Elroy Air – Elroy Air enables autonomous air transport anywhere.
An air cargo delivery drove capable of carrying 500lbs (225kg) payload for up to 300mi (500km). Amazing.
Because…. what could possibly go wrong if we feed ML garbage data?
Answer: Quite a bit.
Q: Why Do Keynote Speakers Keep Suggesting That Improving Security Is Possible? A: Because Keynote Speakers Make Bad Life Decisions and Are Poor Role Models | USENIX
Great talk by James Mickens on the implications of GIGO at scale.
For now, the right approach might be to try out more ideas in a space that is less prone to dropping pianos in the wrong place at the wrong time. Perhaps the use of AR and VR will provide places to apply ML data set findings in a safer… construct.
“The Matrix” (1999) — ‘Construct’ Scenes
Because leather lounge chairs are the real danger.
Then again, I can officially reflect on my college years when what was theoretical materials science and engineering that has now become a reality. Numerical methods indeed.
Using artificial intelligence to engineer materials’ properties | MIT News
The novel use of engineered strain was one of the theoretical topics when I was in school. Amazing.