🔮 Extended intelligence; Tesla’s ambition; AI vs workers; investing under climate change; calories, miners, BS++ #215
|Apr 28||Public post|| 59|
Azeem Azhar’s Weekly Wondermissive: Future, Tech & Society
This issue has been supported by our partner: Ocean Protocol.
Dive into Ocean Protocol with our free tutorials for data scientists, data engineers
and Python developers.
You are reading the Exponential View newsletter.
On Wednesday, May 1, we’re hosting a fascinating discussion with Richard Muirhead, Founding Partner at Fabric Ventures. Richard is a pioneering investor in decentralised technologies. He argues we are witnessing a paradigm shift in software archiecture which will ultimately mean a greater democratisation of services and business models.
You can attend this webinar anywhere in the world for free, by REGISTERING HERE. (2pm GMT, Wed, May 1st.)
Our State of the Exponential briefings are members-only content, but we are making Richard’s briefing available to all readers, including non-members.
If you are a non-member and want to go premium, which gives you access to more briefings and specialist insight, then click the button below.
Dept of podcasts 🎧
Economist Mariana Mazzucato is one of the key figures challenging the status quo of the mainstream economics and capitalism. In this podcast, I explore the role of governments and entrepeneurs in innovation and the economy.
If you enjoy it, let us know what you thinkin the comments below, or rate and review the podcast om Apple Podcasts. We have about 400 five-star reviews. I’d love a hundred more. :)
Dept of the near future
Joi Ito: Forget the Singularity, the future is one of extended intelligence where “we consider the system that integrates humans and machines – not artificial intelligence but extended intelligence.” (See also: a proposal from Joi’s MIT Media Lab for a new field of study, machine behaviour, to foster “more open, trustworthy, reliable investigation into the impact intelligent machines are having on society.”)
Economists, Acemoglu and Restrepo, on the “wrong kind of AI”. They argue that during rapid automation, workers can be particularly affected if new technologies do not raise productivity sufficiently. Specifically, that for workers to do well out of this phase, novel labour-intensive tasks need to be introduced. Without such novel tasks, which often require government intervention or challenging vested interests, workers will suffer relative to other parts of the economy. (Dry, readable economic paper from two important thinkers on the impacts of automation.) See also: Amazon’s fulfillment centres have productivity-monitoring technologies that automatically recommend which workers should be fired. Three hundred were axed last year. Uber drivers in seven US cities are planning to protests wages and working conditions. )
🗺️ The end of the liberal international order (if it ever existed). Slightly dense essay, but relevant:
If the international order is having greater difficulty creating rule-based [liberal] governance, it might have less to do with the weakening of liberalism and more to do with the fact that the rules that have been in place for decades were overdue for an overhaul, and especially given a shift in power from the West to the East.
And that good global governance does not need to live or die with liberalism; it may be better for ‘more experimentation, more kinds of actors, including corporations and nongovernmental organizations, and more willingness to develop alternative architectures’. (See also, Martin Wolf, “The age of the elected despot is here” and the essay below on China’s surveillance state.)
🇨🇳😨 The surveillance state is being made in China and exported over the world.
Today, 18 countries — including Zimbabwe, Uzbekistan, Pakistan, Kenya, the United Arab Emirates and Germany — are using Chinese-made intelligent monitoring systems, and 36 have received training in topics like “public opinion guidance.”
⌛🌎 Climate crisis: 413.67ppm
Each week, we’re going to remind us of the CO2 levels in the atmosphere. We must avoid a level of 450 parts per million for a chance to keep global warming below 2°C. If we don’t change how we do things, we’re likely to exceed the target in 10-15 years.
The latest measurement (as of April 24): 413.67ppm; 12 months ago: 409ppm; 50 years ago: 326.66ppm; 250 years ago, est: 250ppm.
For a clarifying and sobering visualisation of nearly a million years of atmospheric CO2, look at this.
A decent long-read is Mercer’s guide to investing in climate change, which makes a clear case for the many investment opportunities in keeping temeperative rises to below 2°C. (All sectors are negatively impacted at rises 3°C and 4°C.)
Share this with your friends by forwarding this email or via Twitter.
Dept of transportation
Tesla showed off its own proprietary chip, developed specifically for the needs of its self-driving vehicles. It’s quite an impressive piece of work. Some technical details from Tesla’s presentation are here.
It shows the benefits of vertical integration. The chip is highly specialised, specifically tuned to the power and space requirements of the Tesla car and their sensor rig. It only consumes 72 Watts and at 147 trillion operations per second, is plenty fast enough to handle the inputs from the car’s 21 sensors. It might not be the most powerful computer designed for self-driving cars—Nvidia’s has more oomph—but it is certainly built precisely for the needs of a Tesla.
It also highlights the scale and what Steve Cheney calls “Tesla’s incredible platform advantage”. In his excellent analysis, Cheney harks back to Steve Jobs’ prescient decision to have Apple makes its own chips:
It is – in fact – chip making capabilities, that Jobs famously brought in-house to Apple shortly after the launch of the original iPhone, that helped Apple create a massive profit moat between itself and an entire industry (smartphones) [...] A Tesla can be exclusively Tesla under the hood—aside from the image gathering sensors themselves. Others will be effectively running off the curve both of LiDAR vs image processing as well as the OS and chip level frameworks supplied by Nvidia and anyone else they need to depend on in the standard supply market. For Tesla this is all about custom software at the bleeding edge and strategic learning.
It does provide another example that proves that this phase of the technology industry favours scale. Smaller entrants to the auto industry don’t have Tesla’s scale. It is hard to see how traditional automakers, so heavily reliant on OEMs, like Bosch and Valeo, can easily or quickly get in the business of custom silicon.
It also asks some really difficult questions about Level 5 driving. While there is a lot new in this chip, to some degree there is nothing new. The sensor package is mature, the performance of the chip (in terms of perception, inference, decision making) is not ground-breaking: why should this suddenly enable Level 5 driving? What will change? (It could just be a case of needing very much more training data for the models, but this new Tesla’s chipset doesn’t close the shortfall in training data. And it’s not clear Tesla, or anyone else, has a pathway to doing that in the next 18 months.)
One of Musk’s other claims that Tesla would have one million robo-taxis on the road by 2020. Owners could simply add their cars to the Tesla Network platform and start to earn up to $30k per annum when the car is not in use. It’s a bold vision. I find it baffling, with far too many hurdles to overcome in the short period of time between now and December 2020. (Hurdles are regulatory and technological; but also persuading people who can pony up to $100k for a luxury car to let total randoms ride in them.)
That presentation nearly veered off-the-rails at times. Tom McKay does a good job ranking the claims made.
Tesla’s dominant market share looks really vulnerable as soon as there are other luxury electric vehicles available. In Norway, Tesla Model 3 sales are down 82% and Audi eTron and Jaguar i-Pace are outselling the Tesla X and S by nearly six-to-one.
Driving? The kids are so over it, argues Adrienne Roberts. In 1983, more than 92% of Americans between 20 and 24 had a drivers license. That number is now down to 80%. The proportion of 16-year-old with a license has halved.
A hacker finds a way to break into GPS tracking apps remotely: “I have fully [sic] control hundred of thousands of vehicles, and by one touch, I can stop these vehicles engines.”
Dept of AI & bots
MuseNet from OpenAI: a deep neural network that can generate 4-minute musical compositions with 10 different instruments, and can combine styles from country to Mozart to the Beatles. Quite stunning. Be sure to play about with the composer mappings, as well.
💬 A new paper uses machine learning to decode neural activity in speech centres into audible speech. (See also a Scientific American article about this, without a paywall.)
⚕️ A commercial artificial intelligence (AI) system matched the accuracy of over 28,000 interpretations of breast cancer screening mammograms by 101 radiologists but “the performance of the AI system was, however, consistently lower than the best performing radiologists in all datasets.”
But caveats abound for implementing AI systems into real-world systems. After all, IBM’s Watson over-promised and under-delivered in healthcare: “the Watson Health story is a cautionary tale of hubris and hype. Everyone likes ambition, everyone likes moon shots, but nobody wants to climb into a rocket that doesn’t work.”
Using adversarial machine learning, researchers at KU Leuven create colour printouts that trick image recognition in security cameras.
The power of large-scale data analysis: researchers at Stanford and the University of Warsaw predict car accident risk from Google Street View images of properties and people on them. “[F]eatures visible on a picture of a house can be predictive of car accident risk, independently from classically used variables such as age, or zip code”.
Short morsels to appear smart at dinner parties
The death of the calorie: Good long-read on the limitations of using the calorie to guide eating.
The Dutch East India company was richer than the Apple, Google and Facebook combined.
⛏️ The Mueller Report identified that Miners for Trump was actually a Russian troll network.
🔒 Governments are shutting down the Internet more frequently.
More than half of UK digital ad spend is on mobile phones. 📱
🤯 What does brain-stimulating memory enhancement feel like?
🤣 Princeton researchers find that rich guys were more likely than other groups to feign expertise in fabricated topics.
The winner of the French Scrabble championship does not speak French.
Change happens slowly, than all at once. It is quote attributed to various writers, the best I can find is in Ernest Hemingway’s novel, The Sun Also Rises, where a character describes the process of going bankrupt as “[g]radually and then suddenly.”
As I enter the fifth year of chronicling my learning through Exponential View, I’m seeing this dynamic. More and more of the good and bad themes identified in the first year or so have turned from tendrils of the possible to a torrent of the actual. Evidence and reality replaces conjecture and hypothesis. In some cases, we were wrong. For instance, fully self-driving cars seem no closer to reality today than they did in 2015. In others, the themes have played out as they were first identified. Take the emerging fracturing of global agreements, and how the new geopolitical game manifests itself in technology. Consider the kerfuffle of Huawei’s 5G ambitions. Or consider, the increasing rise of disinformation campaigns, or the marked shift towards populist politics around the world.
So are at the gradual, slow stage? Or at the all at once, suddenly stage? Hard to tell but 2019 feels much more febrile, jittery, disjunctive, than 2015. Interesting times…
Have a great week!
P.S. Any comments on this week’s themes, click below.
This issue has been supported by our partner: Ocean Protocol.
How can we scale up AI for Good? Here is the latest from Trent McConaghy.
What you are up to—notes from EV readers
Marko Ahtisaari is directing Finland’s largest art festival this year, and the lineup is now live.
Rachel Coldicutt and her team at doteveryone have published a Consequence Scanning manual for companies and individuals who want to ensure their products serve the well-being of society.
Have news to share? Email firstname.lastname@example.org, and we’ll share it with other readers.