New Week #77
Google's new super-massive AI language model is better than you at explaining jokes. Shanghai citizens are being policed by a robot dog. Plus more news and analysis from this week.
Welcome to the mid-week update from New World Same Humans, a newsletter on trends, technology, and society by David Mattin.
If you’re reading this and haven’t yet subscribed, join 20,000+ curious souls on a journey to build a better future 🚀🔮
This week, researchers at Google introduce PaLM, a new language model that pushes at the boundaries of the AI uncanny.
Meanwhile, Shanghai has plunged back into lockdown — and this time citizens are being policed by a robot.
And a new report warns that we’re in danger of causing widespread forest die-off.
Let’s get into it.
🧠 Riddle me this
This week, an arresting glimpse of the intelligence revolution happening around us.
Google provided an update on its work to push back the limits of transformer-based language models. A new research paper introduced the Pathways Language Model (PaLM), a 540 billion parameter model trained on ‘780 billion tokens of high-quality text’. For comparison, OpenAI’s GPT-3 has 175 billion parameters.
If the paper is anything to go by, PaLM achieves incredible results — which eclipse even those of GPT-3 — across multiple language tasks.
One example? Members of the team invented jokes, and asked PaLM to explain them. See two examples of what happened next:
Many people would struggle to articulate such clear, pithy explanations of these jokes.
Remember, these were original jokes created by the Pathways team; no ready-made explanations for them exist online. The researchers say they didn’t cherry pick only the best responses. Across this and other tasks, PaLM displays incredible competence when it comes to context, nuance, sarcasm, and other complex forms of linguistic meaning. It amounts, as the paper says, to ‘a truly remarkable level of deep language understanding.’
It helps that this is a massive language model. But these advances are also fuelled by a new technique called chain-of-thought prompting. That means inserting helpful step-by-step reasoning into the few examples fed to the AI before asking it to start producing its own outputs. For more, see the research paper: Scaling Language Models with Pathways.
⚡ NWSH Take: This week was a goldmine for AI work; I could just as easily have led with OpenAI’s new DALL·E 2, which leverages GPT-3 to turn text prompts into photorealistic images. But the deeper signal, here? Intelligence is rapidly being decoupled from human consciousness, and the vast implications are still almost entirely unknown to us. // Some technologists are keen to strike a deflationary tone when faced with these advances. Sure, they’ll say, this AI can ‘explain’ the jokes. But it doesn’t understand them the way people do. On one level, this is true enough. PaLM certainly doesn’t find these jokes funny. Few want to argue that these AIs have true understanding the way we do, much less consciousness. // But two counter points. First, our understanding of human understanding is limited; what do we mean when we say PaLM doesn’t understand these jokes? Second, at some point what does it matter? A world in which machines can generate and simulate understanding of language with this level of sophistication is one in which so much changes for us, practically and creatively. We may come to believe that understanding and intelligence are quite separate matters, such that one does not depend on the other. That, and many other head-twisting reorientations, await.
🤖 The power of the dog
Chinese authorities put the entire city of Shanghai on lockdown this week. The move is the biggest city-wide imposition of restrictions since the pandemic began.
And this time it comes with a difference. A robot called AlphaDog is patrolling streets in Shanghai, barking out orders to ‘wear a mask’ and ‘wash hands frequently’.
The move is also a reminder of the stringency of China’s response to the pandemic. In announcing the lockdown, city authorities in Shanghai (population 26 million) pointed to soaring cases across the city: 70,000 positive tests in March. Here in the UK (population 67 million) we’ve seen between 50,000 and 90,000 positive tests per day this month, and no restrictions are in place.
🌳 Forests of the mind
A new study says large numbers of historical forests could be wiped out by climate change.
The authors of Global Field Observations of Tree Die-Off used a database of climate-induced forest death events dating back to 1970 to study forest adaptation to warmer temperatures.
Their models identified a hotter-drought fingerprint that show that at 2C of warming, forest die-off events will become 22% more frequent. At 4C of warming, that leaps to 140% more frequent.
Meanwhile, the IPCC released the second part of its latest landmark updates on global heating. The TL;DR? It’s now or never. With immediate and deep reductions in carbon emissions, we can still limit warming to 1.5C. Without them, we face 2C+ and the many crises that this will entail.
⚡ NWSH Take: Across the last few instalments of New Week, I’ve covered research showing the Amazon rainforest approaching a lethal tipping point, and freakish simultaneous heatwaves at the Arctic and Antarctic. Add this new study on forest death to the list. Taken together, they’re a portrait of a natural environment passing into history; ‘granddad, tell us about back when there were forests’. // The IPCC’s conclusion, that we’re at the crossroads now, runs counter to the line I’ve been articulating recently: that heating over 1.5C is inevitable, and we should prepare. I understand that the world’s leading climate scientists want to stress that we still have a chance, however slight, to avoid catastrophic warming. But read the report, and decide for yourself whether the decarbonisation needed to stay inside 1.5C — including a global greenhouse gas emissions to peak before 2025 at the latest — is likely to be enacted.
👨💻 Working for the man
New data from global human resources consulting firm Randstad sheds light on the Great Resignation.
The firm’s latest Workmonitor report — the first was published in 2003 — saw them survey 35,000 workers across 34 markets in Europe, Asia-Pacific, and the Americas. Randstad point to results that they say show attitudes shifting around work and fulfilment, particularly among young employees.
A full 56% of those aged 18 to 24 said they would quit a job that left them unable to enjoy their lives. That was against 38% of those aged 55 to 67.
Meanwhile, a sign of the times? Google is bribing its US staff with free escooters in an attempt to lure them back to the office. The tech giant announced last month that it wants US staff back in the office at least three days a week from April.
⚡ NWSH Take: Randstad is a corporation with an agenda; view its conclusions accordingly. After all, there are many reasons why 18 to 24-year-olds are more willing than older workers to leave a bad job. // Still, something is clearly afoot. Back in 2019, if you’d told me that Silicon Valley tech giants would soon have to negotiate with their staff on whether those staff can hardly ever be at the office and still remain employed, I would have found it hard to believe. Now, that’s where we’re at. Young knowledge workers are rethinking work/life; organisational policies are being shaped by early data showing productivity may improve when people work from home. // See also the rise of the lie-flat trend in China, which is seeing young people reject the idea of a life centred around career. // Where does all this end? Thousands of US workers are currently participating in a global trial of the four-day work week. Perhaps the 15-hour working week that JM Keynes famously predicted in 1930, and that we’ve long bemoaned the absence of, is finally, and by degrees, starting to emerge.
🗓️ Also this week
🐦 Twitter co-founder Jack Dorsey says he regrets his role in building an internet centred around powerful technology companies. This comes amid news that Elon Musk has purchased 9% of Twitter stock, and will sit on the firm’s board. Musk is talking of ‘significant improvements’ to the platform, including an edit button.
🤦♂️ Amazon’s planned new employee chat app will ban the use of words including ‘union’, ‘restrooms’ and ‘pay rise’. That’s according to leaked documents seen by news organisation The Intercept.
🇫🇷 President Emmanuel Macron’s election campaign team have created an idyllic town inside the video game Minecraft. Visitors to the ‘Macronverse’ will find a town square, a number of cafes, and a huge billboard of the man himself.
🌓 Stanford University engineers have created solar panels that generate electricity at night. The process relies on the temperature difference between the ambient nighttime air and the surface of the panel.
👾 Epic Games unveiled the long-awaited Unreal Engine 5. It’s an advanced video game building tool that allows users to create photorealistic 3D worlds and characters.
🏙 Researchers at Waymo used AI to create a massive virtual 3D model of a San Francisco neighbourhood. The team used over 2.8 million photos collected over three months.
👁 Controversial facial recognition startup Clearview AI says it wants to offer its service to banks and other private clients. In New Week #74 I wrote on how the platform, which has scraped more than 10 billion images of faces from social media, is being used by the Ukrainian military.
👨🚀 The first ever private mission to the ISS is planned for launch on Friday. Four astronauts from Houston-based startup Axiom Space will be delivered to the space station in a Falcon 9 rocket created by Elon Musk’s SpaceX.
🌍 Humans of Earth
Key metrics to help you keep track of Project Human.
🙋 Global population: 7,938,571,305
🌊 Earths currently needed: 1.8065658638
💉 Global population vaccinated: 58.2%
🗓️ 2022 progress bar: 26% complete
📖 On this day: On 7 April 1948 the World Health Organization is established by the UN.
Thanks for reading this week.
The emergence of AIs capable of simulating deep understanding of language is set to help reshape our world in the years ahead.
New World Same Humans will keep watching, and working to make sense of what it means for our shared future.
Want to help advance that mission? You can!
If this week’s instalment resonated with you, why not forward the email to someone who’d also enjoy it? Or share it across one of your social networks, with a note on why you found it valuable. All you have to do is hit the button below. Remember: the larger and more diverse the NWSH community becomes, the better for all of us.
I’ll be back soon. Until then, be well,
P.S Huge thanks to Nikki Ritmeijer for the illustration at the top of this email. And to Monique van Dusseldorp for additional research and analysis.