Azeem Azhar’s Weekly Wondermissive: Future, Tech & Society
Ocean beta goes live tomorrow!
You are reading the free edition of Exponential View.
Dept of podcasts 🎧
The podcast is back! And this time we’re in partnership with the Harvard Business Review.
Our first episode is a broad discussion on how AI, blockchain, genetic engineering, renewables, and robotics are on the verge of creating trillions of dollars of value. My interlocutor? The brilliant, Cathie Wood, CEO of Ark-Invest. We had an excellent chat, do listen in.
We’re preparing to share the transcripts of my conversations with all Premium readers. If you’d like to have access to these conversations in text, and receive other special perks, join Premium today.
Dept of the near future
💭 Peter Thiel’s macro-outlook on the future includes predictions that the next Google may not come from Silicon Valley due to its groupthink; and the coming challenge of reconciling identity politics with egalitarian societies. I often disagree with what Thiel, a successful tech investor and a well-known libertarian, says. But there are some challenging thoughts in this interview, including his musings about what long-term growth prospects for Western economies means for our internal schisms. This article is a poor Google translation of the original German, but it is worth reading. Thanks to Rodolfo Rosini for the recommendation.
🔮 Ramez Naam argues that we are at a radically different point in history where building new solar or wind power (or new energy storage systems, in some cases), is cheaper than the cost of continuing to operate existing coal- or gas-fueled power plants.
If the price of new solar and wind drops another factor of 2 or 3, this will be true. (And the cost curves suggest this might happen in the next decade or so.) Remarkable.
🎱 Margrethe Vestager gave a speech on defending competition in the digital world: “one thing we may need to do, to open up competition, is to require companies to give rivals access to their data. Because in this digital age, having the right data can be one of the keys to being able to compete.”
🛑 Deep-learning pioneer, Yoshua Bengio on regulating AI: “Self-regulation is not going to work. Do you think that voluntary taxation works?” (Google’s recent attempt at self-regulating their AI reflects the challenges ahead: their external advisory council, which I wrote about last week, has imploded unceremoniously, as the firm pulled the plug on it. On the next week’s podcast, I speak to Joanna Bryson, who was one of the members of the now-defunct advisory council. See also, researchers discover that Facebook delivers ads based on race and gender stereotypes).
💸 Is the post-ownership society turning into a rentier society? We move from owners to users and lose rights in the process. (This is worthy of more discussion. After all, in business, it made tons of sense to move from owning assets to renting them and ultimately paying for outcomes. Think, for example, of Rolls Royce’s “Power-by-the-hour” programme. Perhaps, one issue is that rental services demand more than the simple subscription, but do demand our data, and limit our ability to exploit the service we are paying for. Curious to explore this point in the comments below.)
💯 Ollie Morton put together a superb overview of computational biology:
It is easy to assume that reprogrammed life is a lesser life, innately commercial and desacrilised—that as the machineries of cell and laboratory become ever more tightly bound, they will squeeze out something that is human, or natural, or both, which ought to sit in the space between them. But it is also possible that a new appreciation can grow out of that space, a sense of what life is and could be, extended and enriched by new understanding.
🔥 Climate catastrophe: 411.93ppm
Each week, I’m going to remind us of our level of the CO2 in the atmosphere. We must avoid a level of 450 parts per million for a chance to keep global warming below 2°C. We’re likely to exceed the target in 10-15 years.
- This week’s level (as of April 6): 411.93ppm
- Last week: 410.58ppm
- 12 months ago: 408ppm; 250 years ago, est: 250ppm.
The Guardian joins Exponential View in publishing atmospheric CO2 levels on a regular basis. Excited to see this. Hope other incumbents follow suit.
Please share with your network as a reminder that we must 1) pressure our local governments to act, 2) change our behaviour and habits to lower our own emissions.
Dept of internet access
Cloudflare is launching a free VPN service, called WireGuard. (A VPN service is an additional privacy & security layer you can wrap over your internet access. I recommend you use it when out-and-about.) WireGuard looks like it’ll be pretty innovative compared to existing VPN products, but it’s running late. I find it telling that a firm as dynamic as Cloudflare (its DNS service has been growing 700% month-on-month) should enter this market. The reason? The VPN market is rumoured to be worth $36bn in 3 years. NordVPN, which is one of the VPN services I use, is even advertising on a Premiership soccer stadium.
The VPN market is complex, and at the sharp end of the geopolitical fracturing of the Internet. Some VPN providers are pulling out of the Russian market as the government begins to ban particular websites.
Because most of the services are not based in Russia, it could make them tricky to ban in an effective way. Roskomnadzor (the government agency) has a spotty record when it comes to blocking services based elsewhere—its haphazard attempt to block the Telegram messaging service springs to mind—though that is perhaps why lawmakers are keen to make the Russian internet (Runet) separable from the wider internet.
Just over half the planet has internet access, Amazon is joining the race to connect the remainder. Its Kuiper project seeks to put 3,236 satellites on orbit to give internet coverage across the globe. Naturally, hugely beneficial, but clear concerns about the growing footprint of the most powerful companies.
This issue of EV has also been supported by Atoms.
Atoms makes the ideal everyday shoe. I know it because I wear a pair. Right now, there’s a waitlist to score a pair of Atoms—but not for Exponential View readers with this secret link!
Dept of artificial intelligence
Is black-box AI necessarily a bad thing, asks Elizabeth Holm:
In reality, scientists and engineers—like all humans—base many decisions on judgment and experience, which are the outcomes of their own “deep learning”. As a result, neuroscience struggles with the same interpretability challenge as computer science. Yet, we routinely accept human conclusions without fully understanding their origin. In this context, it seems reasonable to consider whether black box answers generated by AI systems have a similarly useful role and, if so, when we should apply them.
✨🚗 ✨ I found the claims made by Wayve, a British self-driving car firm, quite remarkable. They claim to have an end-to-end self-driving system using standard cameras, no Lidar, no high-resolution maps, all powered by reinforcement learning:
Our model learns both lateral and longitudinal control (steering and acceleration) of the vehicle with end-to-end deep learning. We propagate uncertainty throughout the model. This allows us to learn features from the input data which are most relevant for control, making computation very efficient. In fact, everything operates on the equivalent of a modern laptop computer. This massively reduces our sensor & compute cost (and power requirements) to less than 10% of traditional approaches.
Why? Well, the self-driving car theory has split into two schools. One school is that you need enhanced sensor packages, including Lidar, to do this. The other school, the Tesla school, is that you can get there with a bunch of cameras.
In both cases, other things help: HD maps and billions of miles driven, connected to several different control systems (some of which may involve machine learning, some of which won’t). And mapping the actor-state space in real-time to make a decision accurately will require teraflops of processing.
If Wayve’s claims are true and they do have “a neural network sophisticated enough to learn how to drive a car in 20 minutes using only a computer and a single camera”, with much lower power requirements, then this is a game changer. One of Wayve’s advisors is the inestimable Zoubin Gharamani, who sold his last startup to Uber. And I hear good things about other members of the team.
The challenge is moving from shiny demos to the real world. Or as EV reader, Professor Andrew Davison, says that “pure ML for the complicated real world is a very long-term game.”
(The Wayve team uses a lot of reinforcement learning. For a longish, but clear introduction to reinforcement learning (RL), read this. RL has a very close relationship with psychology, biology and neuroscience and closely mimics how humans learn to make decisions.)
🤦 IBM proudly claims to have developed an algorithm that can predict with 95% accuracy which employee will quit. It has apparently saved the firm over $300m in retention costs. (Snarky aside: this announcement probably cost them more in lost brand value and future hiring.)
JF Gagné releases his annual report on the global AI talent pool. Based on analysis of LinkedIn, 36,524 people self-identified as AI specialists, a 66% increase on last year.
Women were underrepresented, making up just 18% of the researchers publishing in these conferences. We found that the AI talent pool is highly mobile, with about one-third of researchers working for an employer based in a country that was different from the country where they received their PhD.
⚠️ A group of concerned researchers has called on Amazon to stop selling its face recognition service Rekognition to law enforcement as “[appropriate] legislation and safeguards are not in place.”
Spreading intelligence from the edge-to-the-cloud. The explosion of data is forcing significant changes in regards to where processing is done. We’re increasingly needing to move the computation to where the data is—on the edge of the network.
Short morsels to appear smart at dinner parties
🤔 Would you give up Facebook for $48 a month?
Fascinating organisational teardown of China’s ByteDance, which is behind massive social app hits such as TikTok & Jinri Toutaio. (In that past year, revenue tripled to $7.2bn & headcount doubled to 40,000. About a quarter of the employees are involved in content moderation, another quarter in ad sales.)
🚍 How bad public transport impacts economic productivity. An extremely interesting case-study on Birmingham, UK.
😮 GOJEK may have added $3bn to Indonesian GDP last year.
Robot density in the US manufacturing industry reached 200 robots per 10,000 employees vs. 97 in China (2017). (China is a laggard. South Korea runs to 631 robots per 10,000 employees.)
🕳️ In the VC industry, “Nepotism is rife, innovation is rare, and diversity is a sham”.
🐙 Octopus skin can detect and respond to light, no eyes or brain required.
Lithium-ion battery prices have come down six-fold since 2018, and looking to break the $100/kWh barrier in 2024.
A reason to learn to code: escaping awkward situations.
🎪 The most hyped technologies of the century. Fun infographic.
Data visualisation is hard, even when for The Economist.
Research shows how neurons decide when our thirst and hunger have been saturated. Fascinating.
This week was a big week as we launched a new podcast season with the Harvard Business Review!
The newsletter is shorter than usual for two reasons.
Number 1: I would really like you to give forty minutes over to listen to the podcast with Cathie Wood.
Number 2: The extended essay on computational biology is absolutely fantastic. You need time to take it in! I’m running a stream on computational biology at the CogX festival in London in June, and as an EV reader you get a unique discount for tickets. Cathie and I also touch on some commercial opportunities in computational biology space in the podcast—and we’re preparing a couple future episodes of the podcast on this subject.
Finally, if you have done number 1 and 2, please take a second to rate the podcast on iTunes. It makes a huge difference to help people find it. I know I beg for this a lot, but it really does make a big difference.
Off on holiday next week. EV will be brought to you by a special guest editor.
P.S. Do scroll down to check what your fellow readers are up to!
New data era starts tomorrow.
Join the revolution and help build Ocean.
What you are up to — notes from EV readers:
Congrats to Sandy Speicher for being appointed CEO of Ideo, the world’s foremost design consultancy.
Congratulations to Matthew Gould for being appointed CEO of NHSx.
A profile of Matt Clifford’s plans to “build the next Google” in the Evening Standard.
Stephen Hsu’s podcast season is well under way. Quite an interesting discussion between him and a philosopher on the Ted Kazcynski’s critique of technological society.
Lisa Bari launched Centers for Medicare & Medicaid Services) AI Health Outcomes Challenge, a $1.65M prize competition designed to encourage development of AI-drive solutions to predict health outcomes from Medicare data. Applications close June 18.
Quinn Emmett has been nominated for the “Best Host” for his science podcast Important, Not Important. Vote here to help Quinn get it!
Christina Chipurici’s CEO Library is a treasure trove for bookworms.
Sarah Spencer is inviting you to apply (or nominate someone) for the Next Generation Foresight Practitioners Award.
James Creech is moderating a panel on the challenges to tax administration in the digital age at the 4th annual international Taxpayer Rights Conference in Minneapolis, MN.
Tom Wehmeier: Why do we talk more about blockchain than diversity in tech?
Share your news and projects with firstname.lastname@example.org