š® The future of AI; bottom-up investing, hacking Teslas, monkeys, pizza & coffee++ #258

Hi, Iām Azeem Azhar. I explore how our societies and political economy will change under the force of rapidly accelerating technologies and other trends.
This weekās edition has been supported by our partner, Masterworks
Since his tragic death in 1980, Jean-Michel Basquiat has become the highest-selling American artist of all time. Now you can invest in a painting with 18% average annual returns by the king of blue-chip art on the Masterworks trading platform. Get paid when the painting sells, or sell your shares to other investors. Skip the 25,000 waitlist with this special Exponential View link.
Be part of the change
Over the past five years, Iāve invested many thousands of hours and read tens of millions of words, as Iāve built up my thesis and exposed part of it in these emails. In addition to my time, more than $50,000 a year goes into making Exponential View worth your time. This is made possible by our wonderful readers, so thank you for making it happenāespecially the thousands of you who have already become paying subscribers.
As the fifth anniversary of EV is approaching in early March, weāll be overhauling the newsletter. The biggest change is that weāll stop taking advertising sponsorship. Ā We will focus completely on the content and interactions with you.
By becoming a subscriber, youāre supporting a business model that frees us to focus on creating the best possible version of Exponential View for you. As a subscriber, youāll get every issue of the weekly wondermssive that has become your Sunday staple. You also receive all the other members-only essays, analyses and events.
If you choose not to subscribe, Iāll still send you some of the weekly essays and occasional Sunday newsletter.
Weāre offering annual memberships at a 15% discount right now ($85 per annum, equivalent to $7-and-a-bit, or five-ish quid, a month.)
The near future
𤯠āWeāve never lived in a world where the largest economy is the authoritarian state capitalist. Weāre about to live in that world. That is, under any definition, a much more dangerous and volatile world,ā says Ian Bremmer, geopolitical expert and founder of Eurasia Group, in our latest discussion.
šÆ EV reader, Brett Bivens, examines the merits of bottom-up investing by looking at Benchmark Capital, one of the worldās top VC firms. Bottom-up investing doesnāt mean pretending to have a crystal ball, āit means being able to see the present clearly, understanding businesses as complex, adaptive systems, and recognise green shootsā startups with the ability to flourish in those systems.ā Itās a long-ish piece but full of insight.
š The business of AI is different from the business of traditional software. It has lower gross margins, requires more human support, and can be more difficult to scale. Building ādefensive moatsā for AI products appearsāat least for nowāto be more complex. This has implications for how AI companies should operate: from reducing model complexity to planning for high variable costs.
šŖšŗ The regulators are coming for big tech, but the question is which regulators and when. Europe, described as āboth gnome and giant in the tech worldā, succeeded in setting the bar for privacy regulation with GDPR, and is looking to do the same in other areas like AI and data protection. Mark Zuckerberg and Facebook are currently on a charm offensive in Europe and have released a white paper outlining Facebookās thoughts on content regulation, but so far the reception seems frosty.
Artificial intelligence
There are two broad schools of thought about what it will take to achieve a breakthrough to true artificial general intelligence. On the one hand, groups like OpenAI, a private research group funded by Silicon Valley luminaries and Microsoft, think that the basic building blocks for AGI already exist; itās just a matter of assembling them the right way and scaling. Most of OpenAIās breakthroughs have been based on this āscale and assembleā approach of throwing ever greater amounts of data and computing power at existing AI models, including their recent research into scaling laws for neural language models, in which they argue that āmodel performance depends most strongly on scaleā.
On the other hand are those who believe that true AGI needs to be capable of thinking in a fundamentally different way to the models of today. Elias Bareinboim and Judea Pearl see the inability of current AI models to truly grasp the concept of causation as the key barrier. Introducing causal thinking (the understanding that clouds are not merely correlated with rain, but are the source of it, for example) is crucial for truly intelligent cognition, in their view.
Gary Marcus takes a similar position, arguing that any hope of developing ārobustā AI (below the level of AGI) depends on moving past brittle deep learning to ādeep understandingā: the ability to move beyond correlation to answer the basic questions any human would ask, like who, what, how and why. Marcus engages with the shortcomings of models such as OpenAI's GPT-2, arguing that they represent a thin mimicry of thought rather than true understanding: āThe current paradigm ā long on data, but short on knowledge, reasoning and cognitive models ā simply isnāt getting us to AI we can trustā. Instead, Marcus is calling for a knowledge-driven, cognitive-model-based approach which builds on innovations in hybrid architecture.
Iāve discussed these issues with Gary Marcus and others. See Dig Deeper, below.
𤩠Pretty fascinating: Synthetic biologists discovered a new antibiotic for tackling drug-resistant strains of E. coli using a machine learning system that identified a candidate from more than 107m molecules.
The researchers trained its neural network to spot molecules that inhibit the growth of the bacterium Escherichia coli, using a collection of 2,335 molecules for which the antibacterial activity was known. This includes a library of about 300 approved antibiotics, as well as 800 natural products from plant, animal and microbial sources.The algorithm learns to predict molecular function without any assumptions about how drugs work.
Also quite fascinating: Google has been running machine learning models to help design chips better optimised for running machine learning.
š Climate catastrophe: 414.65ppm | 3,749 days
Each week, weāre going to remind you of the CO2 levels in the atmosphere and the number of days until reaching the 450ppm threshold.
The latest measurement (as of February 12): 414.65ppm; February, 2018: 411.37ppm; 25 years ago: 360ppm; 250 years ago, est: 250ppm. Share this reminder with your community by forwarding this email or tweeting this.
Martin Wolf: The last chance for the climate transition. (Excellent)
š Thwaites Glacier is often seen as a fat canary in the coal mine of climate catastrophe. Its collapse could raise global sea levels by half a metre. Bad news. Robosubs have found warm water welling up from three directions below the glacier.
š” Iron and steelmaking make up about 5 per cent of global CO2 emissions. In our next live expert briefing, weāll discuss innovations in low-carbon steel production with MIT Professor Donald Sadoway. Join the members-only webinar, RSVP here.
Chart of the week

More than half of Americans think tackling climate change should be a top priority for Congress and the president, but the partisan gap is larger than ever. Although, 52 per cent of Republicans under 38 believe the government isnāt doing enough to address climate change.
Short morsels to appear smart at dinner parties
š Things are not going well at robot pizza startup Zume. Despite a $375 million SoftBank investment, the business seems to have fallen over before it ever really got started. Thankfully, it seems no amount of money could realise this daft idea.
A teardown of a Tesla shows that the company is still far ahead of its competitors when it comes to the electronics that are essential to the modern car. By one estimate, Tesla has a six-year lead. The firm even has an ambitious plan for cobalt-free batteries for its Chinese EVs, which will use lithium-ion phosphate batteries instead.
Security researcher, McAfee, turned white-hat hackers loose onto a Tesla. By using a small amount of sticky tape on a road sign, they got the car to autonomously accelerate from 35 to 85 mph.
š¦ Fascinating analysis on the growth of alternative data investors are using to understand the impact of coronavirus.
š¤ On Shenzhenās smart city.
š Scientists have figured out how to spark consciousness in anesthetised monkeys, raising questions about where exactly consciousness resides in the brain.
Great news! Coffee is good for your health (sort of).
Four Anker USB-C charger chips give you enough computing power to reach the moon. š
Pop songs are getting sadder. Why?
End note
I was on a short break this week. I asked James Wise to add some depth to the week and write about regional disparities in innovation and technology. This has become a hot political topic in many countries. This essay will show up in your inboxes shortly.
Cheers,
Azeem
This weekās edition has been supported by our partner, Masterworks
You might have $100M lying around to own a Basquiat, but now you can own part of one. Masterworks is the only platform that allows anyone access to this ultra-exclusive asset class. You might not be able to hang it on your wall, but it can still make you a lot of money. Get started today and skip the 25,000 waitlist.
What youāre up toānotes from EV readers
Francois Candelon: Data sovereignty is a first step towards AI sovereignty and AI competitiveness.
Farhan Lalji on why his team at Anthemis invested in Wollit.
Congrats to Johnny Boufarhat on raising a $6 million seed round from Accel.
Chris Locke shared new research from Caribou Digital on how digital platforms in Africa contribute to upskilling their workers.
Emmanuel Omoruyi reflects on Thomas Pikettyās writing.
Send your news and project updates to marija@exponentialview.co
Dig deeper
I have had several conversations about tech companies and their relationships within sovereign nations and supranational entities:
- With the International Policy Director of the Stanford Policy Center and former Member of the European Parliament, Marietje Schaake, on regulating the cyberspace.
- With Danish Technology Ambassador to the Silicon Valley, Casper Klynge. Casper has since moved to Microsoft to work for Brad Smith.
- My conversation with former British Prime Minister, Tony Blair, about governing in the age of AI.
- With the President of Microsoft, Brad Smith, about regulation and leading a tech giant in the times of decreasing trust in technology companies.
- With Parag Khanna about technology diffusion and the rise of Asia.
Iāve had several discussions with leading figures in the AI space which add colour to the discussions above:
- With Gary Marcus who argues that we need a new set of disciplines to move the field of AI forward.
- With Jack Clark, Policy Director at OpenAI.
- With Jürgen Schmidhuber, one of the pioneers of modern deep learning. Jürgen makes the point that the ever increasing scale of computation at our disposal will continually enable more complex AI systems.
Comments
Sign in or become a Exponential View member to join the conversation.