🔮 The productivity paradox; ecosystems; legacy grids; mRNA goes peanuts #417
Hi, I’m Azeem Azhar. As a global expert on exponential technologies, I advise governments, some of the world’s largest firms, and investors on how to make sense of our exponential future. This Sunday newsletter is a weekly brief for you and 80,000+ readers.
In today’s edition:
How to think about LLMs & their impact on productivity,
The AI ecosystem might not be ruled by one all-powerful model,
Covid vaccine mRNA technique is inspiring new medication.
Total reading time: 6 minutes
Today’s edition is supported by our knowledge partner, Singularity Group.
80 Leaders. 20 Experts. 5 Days. 80 leaders like you who share a passion for innovation and a curiosity about the future. 20 experts to teach you about the next wave of exponential technologies. 5 days of intense collaboration that will change how you see the world and your role in it. This is the Singularity Executive Program. Limited seats remain for our April and June cohorts.
Sunday chart: Will AI really improve productivity?
Nobel laureate Paul Krugman argues that AI would not have any sizeable impact on the American economy for at least another decade. Maybe not.
Krugman makes the case that productivity tends to leg new technologies because firms need time to adjust their processes to those technologies. Before digital transformation projects were electrical transformation and typewriter transformation. However fast the adoption, productivity takes some minimum amount of time to increase as organisations figure out how to use the technology and build the infrastructure to support it.
Indeed, as the chart above shows, electricity was introduced into use in 1892 and we didn’t see a rise in productivity growth until 1920, nearly three decades later. In that time, factories had to be entirely redesigned and assembly lines assembled. (And, roughly, in the era of electrification productivity growth accelerated most as electricity adoption approached 25% of households.)
In the Information Technology revolution, workplaces had to be entirely rethought, rebuilt. Computers were introduced more widely in the 1970s and 1980s to American businesses, preceding an increase in productivity growth. (I don’t show computer penetration for ease of understanding).
We see another spike in productivity growth when Internet penetration passes 25% of American households. I take this 25% level to reflect two things: the first is sufficient time for sufficient learning about the technology for firms to use it usefully and enough penetration to actually have an impact on day-to-day activities.
But why would AI be any different to electrification or IT which took decades to have an impact? (And, in the case of IT, whose impact was short-lived?)
First, getting this technology in the door is pretty easy. Unlike electricity, or even the Internet, AI services will be available without having to build new infrastructure (or even insert an AOL CD into one’s PC). ChatGPT, for one, is spreading at an unprecedented speed: it took only 60 days to go from 1m to 100m users and in March saw more than 1.5bn visits. In my discussions with LLM vendors, systems integrators, consultancies and firms themselves, using generative AI tools is an exceptionally high priority. They are already persuaded.
Second, companies are ready for AI. Three decades of business process reengineering, outsourcing and digitisation have meant firms are better positioned than ever to make changes to functions. In many cases, an LLM used to triage inbound customer requests will simply replace an old-school keyword-based system. The LLM will just handle 95% of inbound tasks without human intervention when the previous tool handled 60% automatically. Plugging an LLM in is much easier than replacing a steam-powered drive shaft with snaking cables of alternating current.
Will this mean a quicker shift in productivity? There are many other confounding factors that could drag productivity growth back to earth. But all other things being equal, the conditions may be ripe for a surge.
See also: Samsung is debating whether to ban employees from using ChatGPT, after workers started uploading confidential information to it.
Key reads
Winner takes all? Maithra Raghu and her co-authors examine whether one huge general-purpose AI model will dominate the entire ecosystem — a monopolistic climax. They believe the contrary: a future made up of a collection of high-utility, specialised AI systems, which will be powered by multiple AI models. See also, a summary of an excellent conversation between Andrew Ng and Yann LeCun.
Fossilised energy grids: This Economist piece explains why it’s difficult to add capacity to the electricity grid, particularly when the power comes from renewable sources. Grids have historically not needed to change their power source or expand much at all. Therefore, specialised components like transformers are tough to get by as there was never any real incentive to invest in the efficiency of their provision. There’s a strong need to update these legacy systems, make grids as flexible as is needed for renewables, and reverse conservative regulation. This needs to happen fast: in the US, the capacity of generation and storage projects (a total of $4.3 trillion of projects) waiting to be connected to the grid far exceeds the capacity of all existing power plants!
Weekly commentary
🐣 This week, I explain why LLMs sometimes surprise their creators - and why such surprises will continue as the models get bigger and bigger. The reason? Emergence.
Market data
VC funding has halved in the past 12 months from its 2022 peak.
88% of developers say they are more productive using Github’s Copilot. The majority (60%) find it increases job fulfilment, with 96% saying that it makes repetitive tasks faster.
69% of US adults would support a 6-month ban on some kinds of AI development, while 13% oppose it.
89% of Parisians voted to remove e-scooters from the city in a referendum.
A study found that inventors who work for incumbents, as opposed to a young firm, have their innovation output decline by 6-11%.
As high as 87.7%, Kima Ventures shares the remarkable performance of their early-stage funds.
Short morsels to appear smart at dinner parties
🥜 Inspired by how Covid-19 mRNA vaccines work, researchers are testing mRNA-based medicine for peanut allergies.
🙋 Bye, bye, CAPTCHA. We no longer know how to prove humanness online.
📱 An exploration of the future of smartphones: from broken-up app stores to AI.
🇨🇺 What declassified Soviet documents about the Cuban Missile crisis teach us about the past - and about the present.
🥀 Plants produce “ultrasonic airborne” sounds when they are stressed.
👁️ Good reasons why not to buy what is advertised to you online. (The products are more expensive.)
End note
Thanks for all the fantastic feedback on the essay+video combo. I’ll keep doing them. I’ll be looking at topics other than AI in the next few weeks too!
Let me know in the comments if there’s a particular issue or question you’d like me to tackle.
Best,
A
What you’re up to — community updates
Claudia Chwalisz’s organisation DemocracyNext continues to shape the deliberative momentum in Europe.
Glenn Leighton has started piloting a new carbon utilisation technology at the desalination plant in Adelaide South Australia capturing around 0.5t of CO2 in the form of calcium carbonate.
Nicholas Niggli, former Chairman of the WTO Government Procurement Agreement, discusses his work and the love of cycling in this interview.
Jeff de Cagna released an open letter on AI for the association community, advocating for a slowdown in development of more powerful models.
Share your updates with EV readers by telling us what you’re up to here.
As always an insightful issue of EV (and the videos throughout the week). Thanks! An anecdotal view but sharing here as it seems to align to what Azeem posted above on Krugman's analysis: In my conversations with people on whether AI will have a real impact on productivity (and others) there seems to have a been a shift over the last 2-3 weeks towards a "it's just hype like crypto", particularly in my French-speaking circles (for some reason). A few days ago I was having the same conversation with a rather well-versed friend who has been working on innovation for many years and I was surprised to get the same "it's just hype" reply. I insisted that the triad ease-of-availability, real-concrete-value, ease-of-usability by virtually anybody made it different, to which I didn't get much of a rebuttal. In any case, my current sense (really and truly "gut feel") is very similar to what I felt in early 90s with WWW and circa 2007 with Cloud / iPhone: it's a significant shift.
If we were having this conversation three years ago, I might agree with Krugman's assessment. But we aren't in a iPhone 2007 situation where the potential is yet to be realised. AIs will get better, of course - and who knows where we are on the S-curve - but they are already incredibly useful. But who is going to look at their graduate level workers and think "Nah, I think I'll pass on a potential 40% increase in their productivity and wait to see what happens"?