🔮 AI & climate change; blockchain and the internet; medicine meets AI; Alexa's growth, Apple's error; cheese, MOOCs, rotary phones++ #200
I received a huge dose of optimism this week when I was in Dublin
|Jan 13||Public post|| 15|
Azeem Azhar’s Weekly Wondermissive: Future, Tech & Society
This issue has been supported by Hanover Communications
Hanover prepares brands for the age of AI
Read their report on the reputational challenges that come with the AI revolution
Dept of podcasts 🎧
The Exponential View podcast is supported by Spotify.
The podcast resumes next week after a short break.
If you haven’t listened to the series, find it wherever you listen to your podcasts. And if you have, please help me spread the word by leaving a review or rating. It really helps!
Dept of the near future
🔮 Gradually, then suddenly. Tim O'Reilly on the “small changes which accumulate, and suddenly the world is a different place”, including his insights on ubiquitous AI; the decline of the US; the future of food; genetic engineering, and more.
💯 Chris Dixon on how the blockchain could allow us to build internet services that encapsulate the original intention community-led ethos of the internet. It is an appealing theory: combining the bottom-up, non-hierarchical nature of open-source with a mathematically-rigorous implementation of computational trust. (Two EV subscribers, Kevin Werbach and Kenn Cukier, discuss the blockchain, its trust and governance mechanics. Also, McKinsey & Co on blockchain’s Occam problem: the finance industry is cooling on it as little of substance was being delivered. Azeem’s comment: seems like a problem of exuberance before product-market fit. There is a high likelihood as Werbach, Dixon & I argue that blockchain is going to be a powerful, foundational technology. But technologies need time to mature and evolve, and generally one wants to find where the smartest developers are and how practitioners are evolving applications on a core technology like this. And generally, don’t follow the big dollars, shiny PowerPoint & crisp suits.)
🌍 What AI will mean for climate change. It’s complicated, argues David Victor. On the one hand, oil and gas firms are well positioned to use new analytical techniques to increase efficiency and get those wells flowing. But there are also transformative ways AI could help: by helping us focus adaptation strategies; by democratising & evolving effective local responses more rapidly; by differentially helping poorer communities to respond to the ravages of climate change. (Azeem’s comment: As a general purpose technology, AI could also reduce the cost of basic goods, through optimisations and efficiencies, a change that should benefit the poor more than the rich. I also reckon we can be hopeful that applying AI in biology, biochemistry and core scientific domains could help with specific technical developments that might mitigate or help adapt to the impact of climate change. See also, the mounting challenges for the insurance industry in dealing with climate change. Urban flood damage alone is forecast to jump 20-fold to about $1 trillion a year by 2050.)
🍿 Ocasio-Cortez vs Trump. Kara Swisher on the new tempo of politics and why the US President may have met his match—at least on social media.
💸 Venture capital is on fire. Nearly $100bn of deals in 2018, the highest level since 1999, even though the total number of deals declined. This is because of more megadeals. Seed funding actually declined as well. (See also: more startups pass on venture capital "jet fuel", and new types of capital are starting to emerge. Gerry Neumann explains the maths of venture capital and why VCs need to take high risks, with limited room for medium risk investing. Azeem’s comment: There are many sources of capital, with different risk appetites and different expectations. Choose wisely! Don't fuel your jet with forecourt diesel. And don't put rocket fuel into your moped.)
Dept of internet business
C-r-a-z-y. There are more than 100m Alexa-enabled devices worldwide, including about 150 that are not made by Amazon. The device had a bumper Christmas shopping season. I remember reading forecasts for voice devices back in 2016 which headily suggested a global installed base of 5-6m by 2018. And here we are with more than 20 times more than that. Of course, Apple via Siri, Google through Google Assistant and even Microsoft's Cortana have much larger user base that Alexa. (Siri and Google Assistant approach or exceed a billion; Microsoft announced monthly 150m users for Cortana last summer.)
And with those numbers come more predictions that 2019 will be the year voice commerce takes off. My prediction: it won’t, if I am bullish in the long term for voice as an interface. I don't believe we have adequately solved the user experience or the core natural language understanding techniques to really delight. Amazon trumpeted several Alexa use cases that were growing in this blog post, but ultimately music, timers and smart home integrations seem to be the dominant apps. Is it enough?
📺 Netflix and chill no more. As the technology commoditises, more studios are yanking content from Netflix and preparing to launch their own streaming services.
🤔 Americans over 65 are seven times more likely to share fake news on Facebook than the 18 to 29-year-olds.
Ofo, the over-expansive bike-sharing business is teetering on the brink. (It was rumoured to be burning $25m per month.)
Apple is United’s biggest corporate customer globally, spending $35m a year on flights to Shanghai from SF alone (50 daily business class seats).
Apple's new phone will have three rear cameras.
🔥 Ben Thompson on Apple's errors.
Dept of AI & data
Eric Topol’s comprehensive survey on the convergence of AI and human doctors on the path to high-performance medicine. Long, deeply footnoted, and worth reading over two cups of coffee.
almost every type of clinician [...] will be using AI technology, and in particular deep learning, in the future
He reviews applications in radiology, dermatology, ophthalmology, mental health, health systems, patient experience and more.
One useful framing is his analogy between self-driving cars and medical applications. Topol reckons a level three equivalence (conditional automation, reliant on human experts as backup) is the likely object for medical AI, and not human-free level five autonomous vehicle makers are striving for.
This is a healthy dose of sobriety for outlandish technical claims about “replacing doctors” rather than augmenting health systems and improving outcomes.
And he cautions:
The field is certainly high on promise and relatively low on data and proof. The risk of faulty algorithms is exponentially higher than that of a single doctor–patient interaction, yet the reward for reducing errors, inefficiencies, and cost is substantial. Accordingly, there cannot be exceptionalism for AI in medicine—it requires rigorous studies.
One key requirement for AI diagnostics will be explainability. Here is a profile of Been Kim, a Google researcher, working on building interpretability for famously indecipherable machine learning systems. Recommended read.
After raising $1.2bn last year, Sensetime, the world's most well-funded AI startup, is after another $2bn. One of its products is Viper which will “process and analyse over 100,000 simultaneous real-time streams from traffic cameras, ATMs, and more to automatically tag and keep track of individuals.”
🔒 The US Census Department intends to use differential privacy to manage the 2020 Census. This should allow the dataset to be shared without compromising individual user privacy.
Reconciling deep learning with symbolic artificial intelligence: a handy survey of approaches that bridge symbolic and sub-symbolic approaches to AI.