New Week #118

Meta's new visual AI can tell its apples from its pears. Japan needs 11 million more humans. 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 24,000+ curious souls on a journey to build a better future 🚀🔮

*

To Begin

It’s a rapid fire instalment this week; three signals that fell on to my radar across the last few days.

Meta announced a new AI model for analysing images. But what does it indicate about the way the company’s priorities are shifting, and the broader emerging collision between AI and virtual worlds?

Meanwhile, a new forecast suggests Japan is set to face a massive shortfall of workers. And Bloomberg trains a large language model for finance.

Let’s go!

*

🤖 Vision on

This week, the hype around foundation models collides with the quest for more useful and compelling virtual worlds.

Meta announced Segment Anything, a new AI model for image segmentation. In short, this AI can look at a photo or video and identify all the objects in it.

Object identification — essentially, making some sense out of a mass of pixels — is the central challenge when it comes to machine vision. The Segment Anything Model (SAM) represents a big step forward, and an array of powerful possible uses are already clear.

To start, there’s image and video editing. But think also about autonomous vehicles, everyday helper robots and their ability to navigate their way around our homes, and our experience of virtual and augmented realities.

In future, say Meta, SAM could be used to ‘identify everyday items via AR glasses that could prompt users with reminders and instructions’.

⚡ NWSH Take: We’re hearing far less about the metaverse these days. After the hype of 2020/21, reality had to set in; we’re yet to see an AR or VR innovation compelling enough to weave itself through the lives of billions of people in a lasting way. Decentraland’s Metaverse Fashion week made headlines a few days ago when hardly anyone attended. // Still, I remain convinced that the underlying promise of virtual worlds is real and will eventually be manifested — probably in some ways that aren’t yet on our radar. And SAM is a reminder that 2023’s technology of the moment (AI, if you hadn’t noticed) will help unlock that promise. When machine vision can uniquely identify every object in an AR or VR world, that world becomes vastly more useful and usable. // In short: Meta’s much-discussed ‘pivot to AI’ isn’t as straightforward as some are making out; much of the AI work they’re doing could end up fuelling the broader mission around virtual worlds. // Meanwhile, want another glimpse of the way generative AI is colliding with virtual space? Check out this prototype text-to-worlds tool from Blockade Labs.

*

👨‍💻 Work it

Also this week, a glimpse of a demographic reality set to reshape life across the Global North.

A new study by the think tank Recruit Works Institute says that Japan could face a shortfall of 11 million workers by 2040. Their new report forecasts that demand for labour will remain broadly level across the next 17 years, but that a rapidly ageing population will see the workforce shrink by around 12%.

This is a huge shortage of labour.

For Japan, dire demographic forecasts are an ongoing crisis; the current Prime Minister, Fumio Kishida, has warned of ‘societal collapse’ if the issue is not addressed.

For the others in the Global North, though, they are a warning: this is how bad things could get if the birthrate falls even lower.

⚡ NWSH Take: Last week I wrote on a new study from the Club of Rome, which forecasts that global population will peak at around 8.8 billion in 2050. That’s far lower than the consensus UN forecast, which has global population peaking over 10 billion. The big question made acutely visible by that study: is a far smaller population good or bad news? // Japanese leaders are in no doubt what a shrinking, ageing population means for them, at least economically: it’s a massive productivity problem. Pretty soon, most other countries in the Global North will be facing the same kinds of numbers. The obvious lever to pull is immigration: Japan is famously resistant to that, and it’s hardly uncontroversial in the US and Europe. The other? It’s robots. If people can’t do the work and the economy is to remain as productive as it was, then machines will have to step in. Japan is choosing robots over immigration; it’s the world’s top manufacturer of industrial robots. Here in the UK, fruit growers want the government to allow for more immigration until the industry can automate fruit picking some time in the 2030s. Some version of this argument — give us more immigrants, or give us more robots — is coming to a whole ton of industries across advanced economies in this decade.

*

🧠 Money mind

A few days ago the financial research and media giant Bloomberg dropped this.

Typically I’d embed the tweet, but Elon Musk has disabled that function as part of his new war on Substack. So here’s a screenshot:

BloombergGPT is a 50 billion parameter (GPT-3 has 175 billion) large language model, built from scratch and trained on a massive tranche of financial data. It’s being billed as ‘ChatGPT for finance’.

Bloomberg say the new model will help their clients ‘unlock new opportunities for marshalling the vast quantities of data available on the Bloomberg Terminal’.

⚡ NWSH Take: The rise of potent LLMs is hugely empowering for those who hold proprietary, domain-specific data or content. The holders of that data can use it to train their own model, which will then outperform general purpose models when it comes to the relevant domain. Google have already conquered health with Med-PaLM 2. But who will train the definitive model for law? For teaching physics? For cognitive behavioural therapy? There are worlds, and fortunes, to be won. But as it has proven here with Bloomberg, the spoils will often go to incumbents sitting on data goldmines that were decades in the making.

*

🗓️ Also this week

🙊 Samsung employees leaked confidential company data via conversations with ChatGPT. Confidential information on Samsung’s semiconductor equipment measurement data is now reportedly part of the ChatGPT learning database, and Samsung has encouraged employees to exercise caution during their interactions with the platform.

🚀 Virgin Orbit declared bankruptcy in the US after funding dried up. The satellite launch company was part of UK billionaire Richard Branson’s Virgin Galactic; it has now laid off its approximately 800 staff and will try to sell its assets.

🌖 NASA named the Artemis 2 crew that will fly around the Moon in late 2024. It’s the first lunar astronaut crew announced by NASA in over 50 years. Back in The Future that Never Came I wrote about the idea that, for decades now, IRL innovation has stagnated.

🤖 President Biden says the AI revolution ‘potential risks to our society, to our economy, to our national security’. Last week I wrote about a petition to pause research on the training of AI models larger than GPT-4. Also this week, OpenAI published Our Approach to AI Safety; the statement says OpenAI spent more than six months ensuring that GPT-4 was ‘safer and more aligned’ before they released it to the public.

🦠 Scientists in Israel have created a micro-robot the size of a single biological cell. The robot-cell is inspired by biological ‘swimmers’ such as bacteria, and can identify and capture a single cell. The researchers, from Tel Aviv University, say these robots could make possible new diagnostic procedures, and new and more targeted ways to deliver medicines.

*

🌍 Humans of Earth

Key metrics to help you keep track of Project Human.

🙋 Global population: 8,026,095,380
🌊 Earths currently needed: 1.80148167

💉 Global population vaccinated: 64.5%

🗓️ 2023 progress bar: 27% complete

📖 On this day: On 7 April 1795 the French First Republic adopts the kilogram and gram as its primary unit of mass.

*

All Seeing Eye

Thanks for reading this week.

The quest to create virtual worlds that become meaningful domains of experience is yet another case of new world, same humans.

This newsletter will keep watching. And there’s one thing you can do to help: share!

Now you’ve reached the end of this week’s instalment, 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. Remember: the larger and more diverse the NWSH community becomes, the better for all of us.

I’ll be back next week. Until then, be well,

David.

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.