🔮 The brave new knowledge ecosystem; Chat Xi PT; deglobalisation; electric Ethiopia ++ #475

An insider perspective on AI and exponential technologies

Hi, I’m Azeem Azhar. In this week’s edition, we explore the emerging AI-curated knowledge ecosystem, and how trust and explainability will be necessary to safeguard its health.

And in the rest of today’s issue: 

1. Need to know: Reversible computing
   As the energy toll of AI and computation expands, the science of reversible computing is more important than ever. Read here 

2. Today in data: Electric Ethiopia
   Ethiopia reached its 2030 goal of 148,000 electric vehicles on the roads in just two years.

3. Opinion: Governments (and citizens) can thrive in the exponential era
   I outline seven policies as we look ahead at the UK’s general election. Read here

---

💥 Today’s edition is brought to you by Sana: Unlocking knowledge with AI.

---

Sunday chart: A new information age is dawning

Google debuted its “AI Overviews”, a box which summarises some users’ search results using genAI. It is less than reliable so far, as dozens of examples shared on social media have shown. For example, its AI overviews have recommended adding glue to a pizza to stop cheese from sliding off, suggested that Google’s AI dataset contains child sexual abuse material, and pointed out that a depressed person could jump off the Golden Gate Bridge. 

If and when AI Overviews become reliable, they’ll change the web experience. Users will spend less time clicking and visiting websites as Google search summaries improve and become useful enough. The few remaining clicks may increasingly go to the highest bidder, as ads start to appear in the summaries.

Companies including OpenAI are on a quest for licensing deals with publishers, some of which have already succeeded (Axel Springer SE, Newscorp), but as journalist Hamilton Nolan points out, this is akin to “selling your own house for firewood”. A leaked deck from OpenAI’s “Preferred Publisher Program” promises that members of the program will get priority placement and “richer brand expression” in chat conversations. Results from AI systems may be influenced by sponsored deals even more covertly.

Set against this trend is the sad disappearance of the Web. One estimate shows that 38% of web pages from 2013 are no longer accessible.

There are many moving pieces, but it is clear that our online knowledge ecosystem is likely to undergo a radical shift. More and more of our information will become written and curated by AI, whose data provenance is unclear. Worse, the webs heritage may be lost as pages die, never to reach search indexes of language models.

Two things are needed in this emergent knowledge ecosystem. First, a better understanding of how LLMs transform sources of information into structured outputs in an auditable way. Just this week, AI safety and research company Anthropic managed a breakthrough in this, more specifically what is called mechanistic interpretability. The team found that the model contained a map of related concepts that could be adjusted manually, pointing to a promising avenue of safety research. Second, there needs to be an active effort to maintain and preserve the Web’s rich information heritage.

---

🚀 Today’s edition is supported by Sana.

Last week, Sana hosted its annual AI Summit in Stockholm, with 500 guests exploring AI’s impact across various disciplines. The event recordings are now available for free, including:

• Geoffrey Hinton’s thesis: Hinton and Joel Hellermark discuss AI creativity and emotions.

• Joel Hellermark on building Sana AI: Insights and plans for Sana’s AI knowledge assistant.

• Daphne Koller on reimagining medicine: Using machine learning to revolutionize drug discovery.

• Max Tegmark’s thesis: AI safety, consciousness, and meaning.

---

Key reads

Shadow of the law. The EU AI Act (whose provisions will soon come into force), an international AI treaty from the Council of Europe and commitments from the Seoul AI Summit have all been given the final green light this week. To make sense of these efforts, there are two important considerations. First, the texts are often vague and their effectiveness is undermined by exemptions, such as those for national security, and the influence of powerful nations and tech lobbies, leading to less transparent and enforceable rules. The rules are the result of a power play by influential nations and lobbies, rather than the result of careful and transparent negotiations. Second, as we have written in the past, many of the ways in which the safety of AI is discussed are unhelpful at best. Take the 132-page Safety of Advanced AI report accompanying the AI Seoul Summit, which misses the mark on how to make sure we have a safe and positive AI future. Analyst Brian Williamson puts its well:

We justifiably worry about AI bias, yet the report demonstrates human bias, a glass half-empty view and a failure to consider the potential upside in areas identified as risks; a failure to think through the potential equilibrium impact of more intelligence. This report is not a good place to start on the journey to AI Safety.

Eyes on the prize. Ahead of my speech at the ITU AI for Good Summit next week, I’ve been thinking about how AI can help achieve the UN’s Sustainable Development Goals. And something struck me: many of the most promising applications of AI that aim to solve global challenges are also the least hyped. The technology can help make sense of large amounts of complicated information, which helps accelerate scientific research – for example, to improving health outcomes (SDG3) by searching for new malaria vaccines. It can help us fight climate change (SDG13) by designing new, more sustainable materials thanks to the LLM chemistry agents such as ChemCrow. Advanced systems help track electricity access in Kenya (SDG7) and its impact on livelihood outcomes. Better information leads to making better decisions and optimising how we use critical resources. For example, forecasting renewable energy reduces the amount wasted through intermittency. Similarly, precision agriculture utilises data to ensure that only the right amount of precious resources are used.

Dymaxion projection. Bloomberg columnist Lionel Laurent makes the case that Europe’s economic power is becoming multipolar, with an increasing number of cities outside of the traditional hubs of London, Berlin, Paris and Brussels wielding newfound influence. And indeed, key exponential technologies are impacting traditional power relations. For example, the Netherlands houses ASML, a company that designs and manufactures the lithography machines essential to microchip making. The Swedish city of Kiruna is extracting important rare earth elements for the green transition. I’ve written about this before: a deglobalisation movement is unfolding, and this can be seen even in the already highly decentralised European continent. Countries are focusing on their domestic industries and interests, and supply chains are transforming.

---

Data

• The UK was the first country to reach peak carbon emissions way back in 1971. 

• India’s solar additions grew five-fold in the first quarter of this year.

• Revenue from ChatGPT’s mobile application increased by 22% on the day GPT-4o was announced. Also, it looks like 22% of people are using ChatGPT as an alternative to Google.

• Ethiopia reached its 2030 goal of 148,000 electric vehicles on the roads in just two years.

• France has tripled the price of parking for SUVs. Parking for six hours will now cost €225, up from €75.

• Nvidia’s quarterly revenue soared to $26 billion, marking a remarkable 262% year-on-year increase, once again exceeding analysts’ expectations.

• Half of Americans surveyed wrongly believe that unemployment is at a 50-year high.

• Daily marijuana use is outpacing alcohol consumption for the first time in the US.

---

Short morsels to appear smart at dinner parties

🤔 Chat Xi PT: a chatbot trained on Xi’s political philosophy as China tries to balance AI development and control over its citizens. The FT writes about it here.

🍟 A new study links increased consumption of ultra-processed foods to higher risks of cognitive impairment and stroke. 

🎢  How liberalism evolved into a comprehensive worldview: an interview with Prof. Alexandre Lefebvre. Not a short read, but it’s an interesting one. 

🧠 Mental disorder diagnoses spread through young people’s social networks (school groups for instance).

☕ Ants learn faster on caffeine. And eagles changed their routes to avoid the war in Ukraine.

🗯️ X’s crowdsourced fact-checking may just be working. Bloomberg covers it here.

---

End note

I planned to write about Microsoft’s launch of PCs infused with AI CoPilots but got derailed by the political news in the UK. Two quick observations:

• Joanna Stern’s discussion with Satya about CoPilots is worth watching. I agree with Satya that we should think of AI as tools not people.

• Kevin Scott, Microsoft’s CTO, argues that we are not seeing the end of the scaling laws that have improved LLMs over the past few years. 

I’ll be travelling to Geneva next week to speak at the ITU’s AI for Good event. This will be streamed if you aren’t attending in-person.

Cheers,

A

---

What you’re up to — community updates

• Celia Francis’ Ponterra is one of the developers of the biggest fully-financed nature-based carbon removal projects in South America, providing Microsoft with 1.6 million tonnes of removal credits over 30 years.

• Jacob Taylor writes about collective intelligence and AI.

Share your updates with EV readers by telling us what you’re up to here.