🔮 AI & science; Apple & Ive; job transitions under automation; killing globalisation++ #225

Exponential View

Azeem Azhar’s Weekly Wondermissive: Future, Tech & Society

This issue has been supported by our partner: Ocean Protocol.

sgCarMart is building Singapore's very first 'Know-Your-Vehicle' data marketplace on Ocean Protocol to help used car buyers make smarter buying decisions

This week I am sending the same issue to free and Premium members. Like, comment or share this email by pressing the little buttons at the top of this email.

Dept of the near future

🍏 After 23 years, Apple’s design chief Jony Ive, the “most unremarkable remarkable person you could meet”, is stepping away from the empire he helped build. Ive’s departure, which according to insiders, was preceded by a long, steady withdrawal from the daily work of the company, is being read as a sign of the increasing shift away from the Steve Jobs era into one driven more by operations than inspiration.

🤭 Despite all the early hype around voice assistants, usage is abysmal. Either the industry has yet to find the right use case for voice technology, or perhaps, as one founder points out, “[n]obody wants to walk around their house talking to a computer.” Fun fact: 10,000 Amazon employees work on the Alexa voice assistant.

🚨 An exceptional piece of research from Jordan Dworkin measures the impact of automation in different professions against the ability of workers to transition to new jobs, and concludes that service and industrial sector workers are likely to struggle with transitions more than those in business, scientific and medical sectors.

Optimal recommendations should not only consider a new job’s automatability and growth potential, but also its similarity to the job a worker is leaving. Skills most associated with less automatable jobs differ across sectors, and skills that facilitate job transitions may be distinct from those that facilitate within-job skill redefinition. 

Automation will likely have complex effects on the job market that are not fully captured by the likelihood of individual jobs becoming automated. Instead, it is important that policy-makers, businesses and workers consider the relationships between jobs when determining who is at the highest risk of long-term displacement and which transition or reskilling opportunities they should pursue.

⚖️ Income inequality declined globally between 1988 and 2015, but only if we count India and China. With those two countries removed from the dataset, income inequality has worsened in the remaining 143 countries measured in the study. Growth in regional income inequality in Asia outside of China and India is particularly significant. (Sorry, paywalled!)

📧 Mike Davidson raises an important question about Superhuman, an email client favoured by the digerati. Superhuman tracks when recipients open mail. He explains: “What I see in Superhuman though is a company that has mistaken taking advantage of people for good design. They’ve identified a feature that provides value to some of their customers (i.e. seeing if someone has opened your email yet) and they’ve trampled the privacy of every single person they send email to in order to achieve that”. To his credit, Rahul Vohra of Superhuman, responded effectively ending the practice. (This is part of the norming process we are going through as we grapple with these technologies, see the Dept of Surveillance below and my End Note.)

🌡️ Climate crisis: 413.76ppm | 3,972 days

Each week, we’re going to remind you of the CO2 levels in the atmosphere and the number of days until reaching the 450ppm threshold, the point at which significant portions of our planet’s ice will melt away.

The latest measurement (as of July 1): 413.76 ppm; 12 months ago: 409ppm; 50 years ago: 326.66ppm; 250 years ago, est: 250ppm. Share this reminder with your community by forwarding this email or tweeting this.

🔥 This June was the hottest ever recorded. European average temperatures were more than 2°C above normal, and temperatures were up to 10°C above normal over parts of France, Germany and Spain. Globally, the average temperature was about 0.1°C higher than during the previous warmest June. That was only three years ago. The speed at which temperature records are now being broken is an alarming sign of the future we may be facing. 

🚫 The tide appears to be finally turning against fossil fuels, however. Capital is exiting the coal business, according to a recent report by the Institute for Energy Economics and Financial Analysis. Over 100 leading global financial institutions are pulling back from the sector, creating a “progressive strangulation” on access to capital markets and insurance for coal companies. Earlier this week, for example, insurer Chubb Ltd announced that it would not underwrite new risks for companies which generate more than 30% of their revenue from coal and that it would phase out its coverage for existing risks which exceed this threshold by 2022.

Researchers from the US and China find that committed emissions from existing and proposed energy infrastructure represent more than the entire remaining carbon budget for the 1.5 °C climate target. If we are to meet that target there is no room for new carbon-emitting energy infrastructure, and some existing infrastructure must be retired prematurely. 

Planting a trillion trees could capture as much as 200 gigatons of carbon from the atmosphere.

Dept of artificial intelligence

Unsupervised word embeddings capture latent knowledge from materials science literature. This paper had scientists abuzz with the potential. Essentially, a machine learning system, requiring little training, was able to process more than three million abstracts in materials science. By looking at relationships between words and phrases, it was able to suggest hitherto unknown materials with thermoelectric properties. (Find a less technical summary here.)

The implications for science are significant. As a reminder, this excellent essay on the stagnation of science claimed: “on a per-dollar or per-person basis, science is becoming far less efficient.” If science is suffering diminishing returns, this might mean fewer insights to inspire breakthroughs that we need as we strain our planetary boundaries.

AI is making progress in helping scientists. Think of Deepmind’s Alphafold predicting the 3-D structure of folded proteins or helping chemists understand millions of reaction pathways. And firms like Zymergen and LabGenius use machine learning tools to accelerate genomic research and drug development. The beauty of this current paper is that the technique, word2vec, is really rather simple compared to some breakthrough approaches. And the authors have made their code available for all.

Elsewhere:

• An excellent breakdown by Jaspreet Sandhu of the concepts and current state of knowledge about the role bias plays in machine learning. Sandhu draws out the point that some degree of bias is inherent and even necessary—the important thing is to explicitly recognise and understand which biases are present, how they're influencing the output and what the implications are.
• MIT researchers reckon they can predict emotional states by reflected wifi signals.

Dept of surveillance

Worth noting just how surveillance, by government and private sector, has exploded as an issue. We covered this in last week’s special issue, but there are a handful of more interesting developments. These indicate we need a new agreement around consumer and citizen privacy, and more clarity on how these technologies could be used. Partly it is about law and regulation. Partly it is about norms and behaviour. See my End Note. 

👎 Dave Lee: Facebook may be pivoting to something worse. “Is Facebook fixing itself, or merely making it harder for us to see it’s broken?”

Shoshana Zuboff doesn’t think we’ve failed to rein in Facebook and Google; she thinks we haven’t really tried. She argues that lawmakers need to “outlaw the secret theft of private experience”, disrupt revenue flows and business models based on surveillance, develop laws and regulations which protect users and empower new forms of grassroots collective action against the tech giants. 🔈Shoshana and I had a great conversation about surveillance capitalism a couple of weeks ago.

🚓 The London Metropolitan Police’s facial recognition program continues to draw attention for all the wrong reasons. The system’s sky-high levels of inaccuracy (with a 98% false positive rate) have been repeatedly highlighted in the past. In that context, the 81% inaccuracy rate revealed in the latest independent report commissioned by Scotland Yard almost looks like an improvement—although, in fact, it could be higher than that, as some of the “suspects” could not be found to verify whether or not the program had accurately identified them. The report also identified significant operational problems and issues around informed consent, or lack thereof.

In the US, a consumer advocacy group has asked the Federal Trade Commission to investigate the practice of “surveillance scoring” customers’ financial status. The group points to a well-known study which showed that major US retailers, Walmart and Home Depot, offered lower prices to users browsing anonymously. (Some anonymous users were offered 15% lower prices than identifiable users.)

🇪🇺 An independent group advising the European Commission on whether and how to regulate AI has recommended an outright ban on the use of AI for mass-scale credit scoring and blanket surveillance. The group are concerned about both state and commercial surveillance, but their report overall encourages the uptake and expansion of AI across multiple levels of society.

Short morsels to appear smart at dinner parties

💔 Globalisation is already dead.

🙈 Waymo has been given the go-ahead by Californian regulators to carry passengers in its self-driving vehicles on public roads.

Women score higher than men in most leadership skills. 

Could blockchain help save politics from big money?

👀 Apple is trying to make eye contact using FaceTime way less awkward.

🙀 It wasn't your imagination: wiggling the mouse in Windows 95 really did make it run faster.

The world's smallest MRI visualises the magnetic field of individual atoms.

👂 Five couples lined up at Russia’s largest fertility clinic for CRISPR babies to avoid deafness.

End note

How should we make sense of the surveillance-privacy axis? Well, it is another instance of the ethics of invention. To quote Sheila Jasonoff from her book, The Ethics of Invention (See what I did there?):

The challenge for modern societies is to develop sufficiently powerful and systematic understanding of technology for us to know where the possibilities lie for meaningful political action and responsible governance. The bargains struck in enhancing human capability do not have to be Faustian, ratified between unequal bargaining partners under conditions of blind ignorance or irreducible uncertainty.

I want to talk about email tracking as such a technology. The story about Superhuman broke this week. The firm’s email client did, apparently, cross a boundary. The product had a tracking capability which meant that a sender could see a recipient’s rough location and online status, without opt-in or opt-out. But understand that elements of this capability are now foisted on us across the board. Superhuman was far from unique. Add-ons like Mailtrack and Boomerang have provided similar email tracking capabilities for close to a decade. Messengers like Signal, WhatsApp and iMessage also give away the recipient’s status information.

Superhuman’s response has been rapid and significant. The firm put a moratorium on that particular technology while they figure out what the non-Faustian bargain should be. (I offered Rahul, the CEO, some typo-ridden suggestions about what a product fix might look like.)

The reality is that we are all in a process where we figure out positive-sum arrangements for these enhancing technologies. The technologies will shape our norms, yes. But our norms should also shape the technologies.

Incidentally, you may wonder what kind of tracking happens on emails sent by Exponential View? Well, in the case of this newsletter, every single recipient is on here via a double-opt-in. You have selected to receive the newsletter and confirmed you wanted to receive it. 

In common with virtually every other newsletter, our mailing platform provides us with operational data that includes which users opened the mail, and which links they clicked on. We use this data for operational reasons. Most commonly, we use it to track open and click rates. If we see these changing, usually declining (!!), we can make course corrections. Often a declining open rate means we are being caught by spam filters, and need to take action. Sometimes a declining open rate means that I’ve started to bore you, and we need to shake things up.

We have also drilled down into per-user open rates to understand cohort behaviours. This helps us to talk to you about improving Exponential View. 

We’ve also used open rate information as a way of identifying the most engaged readers. These users have become eligible for perks like early-access copies to new books or exclusive invites to events and conferences. Many hundreds of you have benefitted from this in the past few years. Of course, today those occasional goodies are now concentrated on members of our Premium tier.

We don’t provide an opt-out for tracking these metrics. It is essential operating data for us. If you are uncomfortable with it, you could also use a separate email box to receive your weekly wondermissive. ProtonMail is rather good for those types of things. If your heebies are still jeebieing, I encourage you to use the unsubscribe link at the foot of the email. 

Have a great week!
Azeem