Tangentially related 4: 25 November 2020
Reflecting on a drawn out pandemic and other drawn out technologies.
In this issue, I discuss:
Trends in the pandemic and autonomy,
Economics of eco-friendly autos,
Drawn out “war on cancer,”
Pandemic trends: autonomy, climate change, and more
While there are some cool news stories trickling in about breakthroughs in autonomy during the pandemic, the rate of these doesn’t seem to have ticked up dramatically (e.g. national guard fire mapping drones). What has been coming is more analyses on how the risk of the pandemic is going to be mirrored in future crises of the world. Consider this report from McKinsey, comparing climate change to the pandemic. I’ve highlighted the section matching attributes, as I think it is the best summary. The TL;DR of the article is “we are unprepared for climate change,” which I did not need a reminder of, but it is important to keep up with the different risks facing our society.
Pandemics and climate risk also share many of the same attributes. Both are systemic, in that their direct manifestations and their knock-on effects propagate fast across an interconnected world. Thus, the oil-demand reduction after the initial coronavirus outbreak became a contributing factor to a price war, which further exacerbated the stock market decline as the pandemic grew. They are both nonstationary, in that past probabilities and distributions of occurrences are rapidly shifting and proving to be inadequate or insufficient for future projections. Both are nonlinear, in that their socioeconomic impact grows disproportionally and even catastrophically once certain thresholds are breached (such as hospital capacity to treat pandemic patients). They are both risk multipliers, in that they highlight and exacerbate hitherto untested vulnerabilities inherent in the financial and healthcare systems and the real economy. Both are regressive, in that they affect disproportionally the most vulnerable populations and subpopulations of the world. Finally, neither can be considered as a “black swan,” insofar as experts have consistently warned against both over the years (even though one may argue that the debate about climate risk has been more widespread). And the coronavirus outbreak seems to indicate that the world at large is equally ill prepared to prevent or confront either.
I am working on another piece (and maybe pieces) on how COVID-19 has and has not revolutionized automation, building on my previous piece on the perceived rate of advancement. I think it is best summarized by this new app with an awful play-on-robot: Woebot (not useful, terrible name, yet people keep making them).
The pandemic’s effect on automation has seemed to be more of a climax on automated-media, given the many technology anti-trust hearings. It’s much less about physical robots coming in and taking over with new use cases. Sadly, as usage is up, there seems to still not be the right terminology to discuss the varying directions and magnitudes of risk. How are curated news feeds different in two-sided (user interacting medias) like Facebook and Twitter versus primarily one-sided feeds like YouTube and TikTok?
I actually think the latter have a much simpler system to model and should be regulated first. The scale of interactions is way narrower. Take what we learn when studying the one-sided systems to fix the majority of internet interactions that give us our current digital environment.
These types of questions are going to play out with robots too — likely in the flavor of is a robot interacting with one or multiple people simultaneously, and who gets priority.
Trends in eco-friendly-autos
The future of the personal vehicle is (at least partially) autonomous and electric. There were some interesting moves in this area in the past month. Electric vehicles overtook diesel in Europe — a trend that is definitely assisted by diesel’s decline. This inflection points matter to me, and what comes next? Will full electric or hybrid cars pass new sales of all gas cars first? I think full electric due to the reduced complexity of the drive train.
Autonomous cars are the mainstream drawn-out robotics project. Uber ATG — bye bye. In the restructuring where Uber realized it has to try and run a profitable company, the Uber Advanced Technologies Group may be getting sold. I remember Uber ATG very fondly having never worked there — I think it peaked while I was an undergrad and many of my classmates worked there. Is this part of a bigger trend where autonomy startups lack a path to profitability and an aggregation occurs? I don’t think so yet, given the pandemic has resurged funding to autonomy startups, but I think it’s coming. You’ll have big companies that define different aspects of the self driving experience: Tesla for full deep learning highway assistance, Waymo for sensor-rich, level 5 autonomy platform, TBD for the hardware company supplying chips and sensors, and TBD for the localized autonomy unit deploying taxis on closed courses in cities.
Honda is launching a full-production level 3 autonomous car. The difference between level 3 and level 2 is the need for attention (both are limited to some scenarios). Level 3 autonomy lets the driver have eyes off on top of hands off. I honestly expected companies other than Tesla to hit this first, Tesla is going for the level 4/5 moonshot with deep learning (I am bearish on this, bullish on Waymo).
Are there any companies that you think warrant inclusion in this recent-list?
Challenges of research at the junction of dreams and reality
Consider this: in some high-profile cancer clinical trials, the group of patients is so hand chosen that the control in the clinical trial outlives the real-world treatment with said drug by 50%. What this means is a situation as follows:
Research company selects say 24 patients to receive drug in trial, and 24 to receive placebo (control group).
The 24 patients in the trial live an average of 18 months more being on the the drug, compared to 12 months on the control.
Patients in the real world receiving the drug on average get a life extension of 8 months (worse than control). The real-patients are different.
The standard for cancer treatment is really that low. Only a very few, niche cancers have had dramatic extension life expectancy in the later stages (prevention and early detection methods are the most effective and not changing much).
Sources for this brief:
A recorded talk by Vijay Prasad on the failings of cancer medicine
Two podcasts with Peter Attia: 1) technical discussion with Vinay Prasad and 2) a more heartfelt walkthrough of the history and failings with Azra Raza.
Cancer has by far the most investment, and meh outcomes. What other fields will end up like this? Will sub areas of machine learning? I worry that robotics will be somewhat like this: investment way outpaces the outcomes. Robotics are so defined by science fiction and the allure of the future, which is something I have talked about frequently recently. The landscape is set up very differently from cancer research, thankfully (no clinical trials), so the financial pathways are more reasonable. That being said, the failure mode I see is overshooting — going for the goal that seems 4-6 years away, but perpetually being a little too far to make it happen.
Is there anything I missed? Would you want these issues to include recent research papers I found interesting or any other angle on robotics?
The landscape of graduate schools in the US is changing because Chinese students are still having visa problems. This is idiotic from a national security perspective (the top computing talent wants to leave China, but can’t).
I’m in a reading group on the political economy of RL with some colleagues, and a couple of readers have been forwarded to it. Check it out if you’re interested.
I liked this discussion of the philosophy of different health wearables, from the point of view of Whoop’s CEO.
Tracing AI patents. What if AI is a better currency now than gold?