What does the industrial revolution tell us about automation? How should we predict AI? Do addictive mobile apps make for unhappy users? Big data, machine learning and privacy: an irreconcilable triumvirate?
Have great conversations!
🌎 Would your friends enjoy this issue? Forward it!
💌 Share this issue *|SHARE:facebook|* *|SHARE:twitter|*
📣 Supported by Unbabel, AI-powered translation.
DEPT OF THE NEAR FUTURE
🏭 What does the industrial revolution really tell us about automation and the future of work:
We should listen not only to economists when it comes to predicting the future of work; we should listen also to historians, who often bring a deeper historical perspective to their predictions. Automation will significantly change many people's lives in ways that may be painful and enduring.
⚔️ Rodney Brooks: The seven deadly sins of predicting the future of AI. Awesome essay.
🕸 Neha Narula & team on the fundamental challenges facing decentralised social networks in their ambition to take on the platform monopolies: "there is no straightforward technical solution to the problem."
🤳 The more frequently an app is used, the more likely its users report themselves as unhappy. Most unhappy users: Grindr, CandyCrush, and Facebook; happiest: Calm, Google Calendar and Headspace.
🏎️ Jean-Louis Gassee: Will Alphabet be the runaway winner in the self-driving car race? (It may be a trillion-dollar opportunity, so a prize worth chasing.)
Nation-states are unlikely to collapse overnight. There are no barbarians at the gate. But it evolved during a time of industrialisation… and national loyalty. Modern technology tends in the opposite direction. The future…looks far brighter for the modern, connected agile city.
🌬️ The Rocky Mountain Institute has some optimistic models for energy transition paths which may keep us below the 2°C limit. The full PDF in an enjoyable read on the current state of solar and the challenges of forecasting exponentially changing technologies. (And also, while we won’t run out of lithium for electric vehicles, we may need more mines.)
DEPT OF DATA AND PRIVACY
Machine vision is getting progressively more powerful with significant implications for privacy. A new paper demonstrates a machine vision system that can accurately identify people even when their faces are masked (with glasses, head scarves or fake beards.)
Michal Kosinski (who demonstrated the ability to predict personality traits from Facebook likes) has new research which demonstrates algorithms which accurately predict sexuality from photographs:
When shown one photo each of a gay and straight man, both chosen at random, the model distinguished between them correctly 81% of the time. When shown five photos of each man, it attributed sexuality correctly 91% of the time. The model performed worse with women, telling gay and straight apart with 71% accuracy after looking at one photo, and 83% accuracy after five.
Turns out that humans do significantly better than chance (predicting women’s sexuality 54% of the time from a photo, and men’s 61% of the time). This suggests there might be particular facial structures that are affected by in utero exposure to testosterone and “may similarly be involved in determining sexuality.”
👄 And, Oxford University researchers recently demonstrated an algorithmic system that could lip-read twice as well as a professional human lip reader.
These early approaches will only get better because we have more and more data and because the algorithms used are widely understood and easily accessible.
😬 The policy implications are huge. Governments (not just authoritarian ones) may delight in the ability to identify their citizens under any circumstances. Just this week, the European Court of Human Rights ruled that employers must warn employees before they monitor communications because of the right to “private and family life, his home and his correspondence.”
The urgency to have a debate about how these new capabilities need to reframe the relationship between a citizen and the state only increases. The growth of identity-based, exclusionary politics over liberal politics, as witnessed in the UK and US, adds tinder. Imagine an ultra-conservative, anti-LGBTQ government with access to enhanced profiling tools?
(For more details, see this recent warning from Privacy International on how “Throughout the world the police are acquiring new technologies which increase their surveillance powers on a scale little understood by those they serve. The law is slow to keep pace.”)
🚨 Equifax, one of the oligopolistic credit agencies, was hacked and up to 143m personal records were compromised. Why do these hacks happen? (Three senior Equifax execs were lucky enough to sell stock in the days before the stock-shattering announcement. Equifax isn’t covering itself in glory with its consumer response.)
The GDPR was designed using the assumptions that custodians of data would continue to be centralised entities. However, technologies such as blockchain are facilitating a move towards a decentralised model of data management…[and further, there exists an] incompatibility of immutability and the right to be forgotten.
✍️ Has the NSA identified Satoshi Nakamoto using writing-style analysis? (I can’t tell if this article is a spoof or not.)
DEPT OF COMMUNITY SUPPORT
Exponential View is in hock to the vagaries of incontestable algorithms that determine whether it should be treated like a hand-crafted mail (which it is) or a Groupon coupon (which it isn’t). It’s a constant battle to keep it in your inboxes, away from the shellac discounts. You can help teach the machines a thing or two.
- Best is to refer EV by forwarding the email to friends. We think this is the strongest signal.
- If you can’t do that, then moving it into your inbox or “starring” it helps tremendously.
Both of these actions can ensure that other members receive the newsletter. Please take a moment to help.
DEPT OF AI
[Bit nerdy] A really wonderful overview of Michelangelo, the machine learning platform at Uber. This is a great insight into how you can deliver ML tools to all parts of a business and at scale. Michelangelo hands all parts of an ML workflow including managing data, training models, evaluating models, deploying them, making predictions and monitoring prediction performance.
💉 IBM has been pitching its Watson toolkit as a cancer-busting technology for years. EV has covered it and I’ve used it as an example in multiple presentations. Now longer term results are in and an investigation:
has found that the supercomputer isn’t living up to the lofty expectations IBM created for it. It is still struggling with the basic step of learning about different forms of cancer.
Watson seems to stumble in four areas: (a) difficulty in understanding doctor’s notes (b) ability to guarantee it would ‘first do no harm” (c) a paucity of data with which to train the systems and (d) a bias in the training data resulting from close collaboration with one hospital.
🇷🇺 Putin: Whoever leads AI will be leader of the world. (Russia already leads in cyber-information operations, see the piece on ‘fake Americans’ below.)
Carlos Perez: three dimensions for tracking progress to artificial general intelligence.
Tractor company, John Deere, adds an AI-weed killing firm to its autonomous toolkit.
Huawei, the Chinese telecoms equipment manufacturer, released a 10nm chip optimised for deep learning.
Could machine learning help spot Alzheimer’s disease five years before symptoms are visible?
SHORT MORSELS TO APPEAR SMART AT DINNER PARTIES
🎭 A gripping insight into the fake Americans created by Russian information operations.
🛏️ Lessons from scaling Airbnb 100-fold. Fascinating recent history.
🕳️ Game of Thrones season 7 has been illegally downloaded more than 1bn times.
Welcome to the H-Bomb club, North Korea. Now what?
🎗️ Nice example of how a blockchain-enabled bank card is supporting refugees in Finland.
👕 Clothes that grow with kids (they fit one kid for three years).
How stock photos have changed their representation of women from sex objects to gritty achievers.
Video: cats love fidget spinners. 🐈
WHAT YOU ARE UP TO
Long-time EV supporter, John Battelle, is hosting the second edition of NewCo Shift Forum in San Francisco in February next years. Last year's event (at which I spoke remotely) was superb. I am happy that John and I have agreed that EV readers can get a unique rate on the tickets. Use the code EXPVW when you register to get $500 off.
Thank you to everyone who responded to my request for thinking around blockchains, tokens, and crypto-markets. It’s interesting food for thought and helped expose some deeper thinking on the subject. We’ll return to that.
P.S. As the old adage goes, sharing is caring. Tweet this to recommend Exponential View to your friends and foes.
This week's issue is brought to you with support from our partner, Unbabel.