Tangentially related #3: October 7
Industry-academic complications, the big picture, more on Amazon, drones, and snacks.
|Nathan Lambert||Oct 7, 2020|
Following the world of autonomy is fun — it’s a roller coaster, and I think my takes may reflect a sense of optimism that practitioners had a decade ago, even though many of the open problems were never solved (ahem, when are robots going to be in our homes). Robots really still only work best on closed loops, $15/hour humans are hard to beat, and we are ready for reinforcement learning to change the world of robots.
In this issue:
Some big picture trends on how AI research in academia and industry is affecting the broader industry,
Advancement in adoption of AI globally and how this relates to geopolitical stability,
More on Amazon’s robots,
Drones as an asset, risk, and opportunity,
Snacks (quick news takes).
Transparency in industry-academic AI
Among a smaller group of faculty that works on AI ethics, they also found that 58 percent of those had been funded by Big Tech. When any source of funding was included, including dual appointments, internships, and sabbaticals, 32 out of 33, or 97 percent, had financial ties to tech companies.
AI has a substantial brain drain from academia to the point where those who remain still are tied at the hip (Wired). It’s interesting to note that the percentage in AI Ethics has much higher prevalence of ties to industry when compared to the general AI research community (58%). At Berkeley, we are currently watching Apple buy out the hardware EE program to be a feeder for internships, jobs, and more (by buy out, I mean the classes have lectures from Apple, fast tracked internships, and direct overlap in work with Apple’s research). I have personally benefited from this, and still have an optimistic lens where it is normally a win-win for those involved, but I think it raises the bar for the minimum amount discursive evaluation for practitioners — with the benefit or industry support comes more responsibility to evaluate if their work will do harm.
Taking this trend one step further, Google is now offering an AI ethics consulting gig (bonus points to the reader that helps me dig through Twitter, but Timnit Gebru who leads Google’s AI Ethics Research had no idea this program existed before it launched). This is pretty odd, as Google surely can cause more benefit by looking at their system with a finer lens, and is somewhat hilarious that they didn’t loop in their researchers in launching a new “product.” Honestly, classic Google — something that seems to hit the mark, but misses a couple of softballs, and no one expects it to be supported in a few years. My take: I want to trust them, but I don’t and that tells us a lot about the state of AI research and its overlap with real world applications.
Future of robotics, automation, and geopolitics
I think the current administration did something right (investing in more AI research nationally), but is $1billion enough, or even remotely close? China is expected to spend $70billion dollars total. There are some key facts missing here, but it’s an area worth keeping an eye on:
How much do these numbers reflect the total (mostly for the US)?
What proportion is defense-oriented vs consumer?
How long will the US’s talent lead last with the immigration changes?
An interdisciplinary group at Georgetown said AI advantages are “messier than oil to assess”
There was a Roadmap for US Robotics released in 2020 to predict the future of autonomy, and its takes on robots taking over are pretty grim (for roboticist). Here is their bland take on the state of robotics integration — growth is consistent, never has taken off, and penetration is low:
Over the last decade a tremendous growth in utilization of robots has been experienced. Manufacturing has in particular been impacted by the growth in collaborative robots. There is no longer a need for physical barriers between robots and humans on the factory floor. This reduces the cost of deploying robots. In the US the industrial robotics market has grown 10+% every year and the market has so far seen less than 10% penetration. We are thus far away for full automation of our factories. US is today using more robots than it has even done before.
I think while I am an optimist for robots having a role in the future, I am obliged to remind my audience that it is not a guaranteed conclusion. Look at this plot of robot sales versus manufacturing employment (not much beyond the status quo here):
Amazon scales up autonomy: for better and worse
“Hands off the Wheel” is Amazon’s program for reassigning workers to more creative and impactful tasks while replacing them with autonomy — its name also doubles as a future PR nightmare for their driving startup Zoox (source).
The purpose was not to eliminate jobs but to automate tasks so that the company could reassign people to build new products — to do more with the people on staff, rather than doing the same with fewer people.
I think this is a good way for a company to keep talent it worked hard to get onboard, but not sure it will work everywhere. It seems sufficiently Amazon to say, “eh, that’s good enough, try taking over this industry.”
I think we really need to see this program in the context of practical numbers: their serious injury rate is way above the industry average for warehouse management (source). Combine this with Amazon considering delivery drones, and there're some storms brewing that I would be surprised if they get away with all of them (source). 0% of me wants drones delivering to my door, and especially not my neighbors. They are so loud; I want slow creepy-crawly drivers with solar panels.
More on drones
A founder of Oculus VR, who is still only $28 and a multi-billionaire, now is founding a military contractor called Anduril. There are some direct links here, like the company having VR simulation tools for the military, but they are also innovating in drone technology. They made a large, quiet helicopter drone for numerous military applications (suitably named ghost), and I am sure this is early stages for the company founded only in 2017.
I am floating the idea of writing a white-paper on the writ large concerns when making consumer drones extremely accessible (along with the internet providing more information) over the next decade. Making drones is easy, and preventing them is hard. Hell, even the Air Force is starting to consider removing pilots from their planes (may have already mentioned this), but what is new is that they released some results.
Apple started an ML Residency program. If I was trying to get into ML, I would apply to things like this.
Weird AI for the month: an AI chatbot that the creator says will improve people’s mental health.
A database of AI applications in healthcare was published in Nature.
Machine learning finds more planets.
Want to support this Democratizing Automation experience and more high-signal content on AI, robotics, and automation?
Click the share button and send it to one of your buds.
Tap the cute heart button at the bottom — making an account lets you comment too!
Should I start a podcast? I think it’s a matter of when, if not if.