How will crypo-funding disrupt venture capital? How does AI boost profit margins? The promise of solar PV. Intel's new AI chip. Carbs vs. fat. Hacking pacemakers.
Enjoy the issue, ask questions, 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
🤑 As money raised by crypto-token offerings exceeds venture capital funding, Noah Jessop asks “What next?”
access and liquidity, afforded by decentralized technologies, is what could be so disruptive to [the venture capital industry].
(See also: Fred Wilson on how do we value crypto-assets; a skewed sample of my followers on Twitter plans to move many of their assets into crypto over the coming years. Meanwhile, hedge funds are increasingly moving into crypto to juice their returns. Turns out these highly-volatile neoassets are uncorrelated with every other asset class.)
💯 Firms with proactive AI strategies have profit margins 3-15% above their peers in across all industries. And there is a widening gap between digitised early adopters and others. According to McKinsey, the total value of investments in AI has trebled in the past three years with nearly 3/4s coming from tech giants rather than startups.
🛒 Oxford economist, Carl Frey, doubles down on his prediction on mass technological unemployment. In particular, he still warns that
retail is one industry in which employment is likely to vanish, but unlike manufacturing jobs which are highly concentrated, the downfall of retail unemployment will affect every city and region.
(The full report, written with Citi, which I have not read, is available here.)
Adapting to a future in which a millennium-scale flood can wipe out a major city is much harder than preventing that flood in the first place. By and large, the built world we have right now wasn’t constructed with climate change in mind. By continuing to pretend that we can engineer our way out of the worsening flooding problem with bigger dams, more levees, and higher-powered pumping equipment, we’re fooling ourselves into a more dangerous future.
☀️ The under-estimated potential of solar PV. Academic review, including fascinating data on why forecasts consistently underestimate the capacity of solar PV.
🏛️ The long history of (political) corporate personhood. (I believe that businesses are going to be increasingly visible and important political actors.)
🌊 Is the future of publishing a Spotify-like model? Interview with Ev Williams on where Medium goes next.
⚖️ How can competition law avoid being anti-competitive?
DEPT OF ARTIFICIAL INTELLIGENCE
How fast is AI really progressing? This is a key question for answering about the timing and impact of automation, as well as the appropriate policies for mitigation. Tom Simonite reviews the efforts of the AI Index to track progress in various AI domains.
🎓 Fascinating review on the Allein Institute AI Science Challenge, which pits AI systems against 8-grade science test papers. It’s a more structured (better?) version of the Turing Test. The top teams achieved a 59-60% score (where 25% would be random selection, and 40%-ish the result of running a search engine over Wikipedia). These types of questions which required inferencing and the winning teams used combinations of pretty traditional information retrieval techniques and close hand-tooling for the problem domain. Impressive? Yes. General AI? No.
Inside Facebook’s AI workshop: How did Facebook so fully embrace AI and invest in an algorithmic future? In part by focusing on acquiring more data and increasing the speed of iteration, rather than betting on a better algorithm. According to Facebook’s Joaquin Candela:
I’m not saying don’t work on the algorithm at all. I’m saying that focusing on giving it more data and better data, and then experimenting faster, makes a lot more sense.
In addition, Facebook has a structured approach to implementing AI that clearly delineates between application, engineering, and research with an emphasis on making application of AI widely available.
🔊 Microsoft and Amazon are announcing that their voice assistants, Cortana and Alexa, will work together. Details are scarce, and I’m kind of baffled by this. Was it a slow news day? (John Mannes suggests that asking your personal assistant for what amounts to a warm intro to another assistant is absolute lunacy.)
🇩🇪 Germany to implement 20 guidelines for self-driving cars (damage property before animals; animals before people).
Is romantic desire algorithmically predictable? A new study suggests that while we can predict the likelihood someone is desirable, it was unable to predict the likelihood that two people would ‘click’. Perhaps this frontier will remain hidden from machines for a while, or more likely there just wasn’t sufficient data to model the complexity of human-attraction dynamics.
Neuroevolution, a subfield of ML, which uses evolutionary process to elide the architecture of a neural net is getting more important.
Today the connections in deep learning come from a human researcher who decides… what the architecture should be. The 100-trillion connection architecture of our human brain evolved over a Darwinian process over many millions of years…. neuroevolution focuses on the origin of the architecture of the [network] itself.
🇨🇳 Shanghai will be the hub for China’s AI developments.
HireVue is an AI firm claiming to hire people based on their face. Sarah Butcher, on how to handle being interviewed by an AI. (This is clearly an area which provides firms with great efficiency savings, and at the same time will be sink-hole for potential biases and prejudicial implementations, whether by accident or design.)
🔦 Technical: Intel’s new AI chip.
SHORT MORSELS TO APPEAR SMART AT DINNER PARTIES
🎯 1983 paper by MIT computer scientists on how women are being pushed out of the historically female field - computer science - and what both men and women can do about it. Tweetstorm
🐦 Twitter user growth has stalled completely. And it continues to lose money.
Facebook says fake news sites can’t buy ads.
Just how much will Europe's GDPR regulation threaten Facebook & Google's business models. Possibly a lot.
🍝 Very large scale study suggests it’s carbs, not fats, that increase the risk of premature death.
🥄 Can the diet of the Hadza, a hunter-gatherer people, teach us to maintain healthy gut bacteria?
☕ Apparently, coffee naps improve memory and alertness. You should try it. (I do this all frequently!)
Crows rival monkeys in cognitive capacity. 🐒
The “Distracted boyfriend meme” explained.
🚪 Was the US housing crash caused by house-flipping middle-class speculators rather than subprime buyers?
How to spot a Twitter bot.
Vacuum-sealed couples trending among Japan’s newlyweds. Do not try this at home.
I'm reflecting on the crypto-boom. I haven't written much about the impact of blockchains, decentralised apps and the implications of the novel funding methods. But I have been thinking about it for a while and I'll form a hypothesis over the next couple of months. Ping me if you have a particular point of view to share.
Also, I'm pleased to welcome Unbabel as a supporter of Exponential View. Unbabel is a player in machine translation, a topic we've covered regularly in EV. (They actually have quite a helpful video where machine translation has come from, and where it is going.) If your firm is interested in supporting us, please drop Marija a line.
This week's issue is brought to you with support from our partner, Unbabel.