TheChek News Flow

.

IDF Spokesperson Rear Adm. Daniel Hagari says Israel has located a Hezbollah bunker under a Beirut hospital where the terror group is storing more than $500 million in gold and cash.

He releases the details after a night of attacks in which Israel targeted Hezbollah’s financial institutions.

In a press conference, Hagari shows one of slain Hezbollah leader Hassan Nasrallah’s bunkers located under the Sahel hospital in Beirut.

“The bunker was deliberately placed under a hospital, and it holds more than half a billion dollars in cash and gold,” Hagari says.

“That money could have been used to rehabilitate Lebanon, but it went to rehabilitate Hezbollah,” he says.

“The [Israeli] Air Force aircraft are watching the site and will continue to track it,” he says.

During the press conference, Hagari details how Iran transfers money to Hezbollah and other ways the terror group finances its operations and vows that Israel will continue to strike its financial operations.

The reservoir of possible leakers of U.S. intelligence documents detailing Israel's apparent preparations for a retaliatory attack on Iran is big, and the platform chosen by them – a pro-Iranian Telegram channel – is unsophisticated. But national security leaks are rarely a singular isolated event

On Thursday, two U.S. intelligence documents detailing Israel's apparent preparations for a retaliatory attack on Iran leaked on a Telegram channel called Middle East Spectator.

As is routine in such cases, the news was initially met with skepticism by the media and the disclaimer that "We cannot confirm the authenticity and veracity of these documents."

On Sunday, the Pentagon confirmed that the documents attributed to two U.S. agencies – the National Geospatial-Intelligence Agency and National Security Agency – were real. It announced its concern and that it was conducting an investigation into the leak's origins, adding that "damage to national security appears to be limited."

That's hardly reassuring, even if the damage does end up being confined to just these two documents. Surely there are more where they came from, whether on Israel-Iran, Russia-Ukraine or China-Taiwan.

As opposed to an unintentional leak, a deliberate leak of national security information is a criminal offense. Sometimes, depending on the subject matter, the circumstances, the sensitivity of the information, the methodology of how it was obtained and the intended (even if indirectly) recipient, it can constitute treason.

Such leaks are obviously a matter of access, opportunity and motivation. A leak is rarely a singular isolated event, whether it relates to national security, to politics or to corporate insider-information on securities. Every leak has several ingredients: What is the content? How was the information gathered? What is the context? What is its significance? What are the implications? Who is the leaker(s)? Was it intentional? What are the motives? Is it political dissent? Is it a warning? Was there a financial incentive? Were they extorted?

While the original term "leak" was used in reference to an inadvertent slip of the tongue by an official, or an unauthorized but essentially benign divulgence of information, modern usage applies to a whole taxonomy of leaks.

Bernard and Marvin Kalb described at length the art of the leak following their experience with U.S. Secretary of State Henry Kissinger. University of Washington scholar Richard Kielbowicz went further and wrote a book on the phenomenon: "The Role of News Leaks in Governance and the Law of Journalists' Confidentiality, 1795-2005."

Kielbowicz identified the major types of leak: those intended to influence policy; those intended to shape personal image; those intended to implicate someone else; or those intended to improve relations with the media.

Then there are leaks as a form of organizational communication: leaks as upward communication; leaks as downward communication; leaks as horizontal communication within and between governments.

This latest leak could be any of the above, or a combination. One thing seems clear: the platform chosen indicates that this was probably not a deliberate U.S. leak to pressure Israel or alert Iran. There are other, more sophisticated ways and better outlets to do that. The pro-Iranian Middle East Spectator Telegram channel describes itself as "independent, but not unbiased, striving for a multipolar world. Mainly focused on Iran & Resistance Axis" – referring to the likes of Hamas, Hezbollah and the Houthis.

The U.S. documents, based on satellite imagery from October 15-16, detail Israeli preparations for a prospective strike in response to Iran's launching of 181 ballistic missiles at Israel on October 1.

One document is titled "Israel: Air Force Continues Preparations for Strike on Iran and Conducts a Second Large-Force Employment Exercise." The second is titled ""Israel: Defense Forces Continue Key Munitions Preparations and Covert UAV Activity Almost Certainly for a Strike on Iran."

The documents' very recent date is consistent with U.S. assessments about the timing of an Israeli strike, and partially detail Israel Air Force exercises and munitions movement.

The text clarifies that "We cannot definitively predict the scale and scope of a strike on Iran, and such a strike can occur with no further GEOINT warning." GEOINT is an abbreviation of geospatial intelligence. It stresses that "MRBM [Medium Range Ballistic Missile] dispersal is almost certainly defensive" and adds that "We have not observed indications that Israel intends to use a nuclear weapon." Leak or not, that's good to know.

The visual information was compiled and analyzed by the National Geospatial Intelligence Agency, one of the U.S.' intelligence agencies providing general, planning stage and real-time operational data.

The second document is attributed to the National Security Agency, the United States' main agency working under the Director of National Intelligence. It is responsible for monitoring, collecting, and processing of information and data on foreign intelligence and counterintelligence, covering both friend and foe. The NSA's technological focus is on signals intelligence (SIGINT) and cyberwarfare capabilities.

Analytical intelligence reports – and this is what these documents are – have a wide distribution net within the U.S. intelligence community and national security apparatus, thus enlarging the reservoir of possible suspected leakers.

Furthermore, these two documents are presumably only a fraction of continuous monitoring, a snapshot of a process being watched diligently from above the planet.

On Sunday, the Pentagon confirmed the veracity of the information and that the leaked NGA documents are authentic. The leak is "deeply concerning," a U.S. Defense Department official told CNN, adding that the Pentagon and FBI have begun an investigation to identify the source. The documents were marked "top secret," but apparently also had markings indicating that they can be seen and shared among allies in Five Eyes.

Five Eyes is an Anglosphere intelligence alliance comprising the United States, United Kingdom, Australia, Canada and New Zealand. The last three countries have joined and become parties to the multilateral Anglo-American agreement – a treaty for joint cooperation in signals intelligence whose origins go back to joint efforts to break Nazi codes in World War II.

Informally, Five Eyes refers to these countries' intelligence agencies. The term pertains to the fact that no information can be shared with any other country or agency outside of the five. However, the distinctive U.S. formatting of the documents indicates that they were likely not yet disseminated to the other four allies or that the other countries were the likely source of the leak. Which leaves the U.S. government.

The Americans urged, implored and cautioned Israel not to attack Iran's nuclear installations or its oil industry infrastructure. Reportedly, Israel provided such assurances and the United States simultaneously deployed a THAAD missile battery in Israel as another layer of defense and a deterrent against Iran if it entertains the thought of further retaliation following the expected Israeli attack.

The Americans know an Israeli attack is imminent, with President Joe Biden going as far as to say on Friday that he knows when, where and what such an attack will look like. But the United States is also apprehensive about possible escalation (you can't write an article on the Middle East these days without using that term at least once).

Even if the leak was not a deliberate U.S. government attempt to influence policy, and even if the leaker – whatever their motivation – did no real harm, the two documents shed light on the level of nervousness all around.

Nigeria rejects Shell’s planned sale of $1.3bn onshore production unit

The decision comes the same day as ExxonMobil wins approval for its long-delayed asset sale in Niger Delta

Nigeria has rejected Shell’s proposed $1.3bn sale of its onshore oil production unit, dealing a blow to the oil major’s plans to exit the troubled shallow water sector in the Niger Delta region.

Gbenga Komolafe, chief executive of the Nigerian Upstream Petroleum Regulatory Commission (NUPRC), told an oil conference in Abuja that the sale to a local consortium, Renaissance Africa Energy, was blocked because the transaction did “not scale [the] regulatory test”.

While he did not provide details of the regulator’s reasoning on Monday, Komolafe has previously cast doubt on little known Renaissance’s ability to operate Shell’s assets in the country. Renaissance declined to comment and Shell did not immediately respond to a request for comment.

The rejection of Shell’s deal with Renaissance came as regulators approved ExxonMobil’s long-delayed $1.28bn arrangement to sell onshore assets to London-listed Seplat Energy, demonstrating hurdles and uncertainty often faced by investors seeking to divest from Nigeria.

Komolafe said at the conference, held in Nigeria’s capital, that a long-awaited ministerial consent had been granted by his agency for the Exxon sale to proceed. Seplat declined to comment but acknowledged the announcement made by the regulator.

Shell announced in January that it had struck a deal to sell its onshore assets in the swamplands of the Niger Delta in the south of the country, putting it on course to exit the region after 68 years.

The European major had been waiting for approval from the country’s oil minister, who in turn is advised by NUPRC on whether to rubber stamp deals. Nigeria’s president, Bola Tinubu, doubles as petroleum minister.

Italy’s Eni, Norway’s Equinor and China’s Addax, are among companies that have announced deals to sell onshore assets in Nigeria in the past two years because of declining returns as a result of oil theft, violence and environmental damage.

Prospects of better returns in offshore fields have also lured oil majors away from the swamplands of the Niger Delta.

US oil company Exxon and Seplat, which is also listed on the Nigerian Stock Exchange, first agreed the deal in February 2022. Seplat has forecast that the takeover of Exxon’s assets will almost triple its production to roughly 130,000 barrels of crude oil, from 48,000.

The all-cash deal had been in limbo since state-owned oil company NNPC sought to block it, arguing that it had a right of first refusal to purchase the assets from Exxon. The US company operates the permits in a partnership with NNPC, as all international companies are legally required to.

Former President Muhammadu Buhari approved the deal in August 2022 after “considering the extensive benefits of the transaction to the Nigerian energy sector and the larger economy”. But he reversed course less than three days later, saying further regulatory scrutiny was needed.

Italian giant Eni completed the sale of its Nigeria unit to Oando for $783mn in August while Norway’s Equinor sold its subsidiary to Chappal Energies, a local company, for an undisclosed amount last November.

Chappal Energies this year acquired a minority stake in a Total onshore venture for $860mn.

Clementine Wallop, director for sub-Saharan Africa at Horizon Engage, a consultancy, said the approval of Exxon’s transaction was “good news for the Nigerian government and for investors in the energy space, after a long wait that has caused uncertainty”.

The US and Europe must partner with the region to dismantle bigger and stronger criminal enterprises

Latin America is learning the hard way. Organised crime in the region has been bad since the 1980s; but “reorganised crime” is proving far worse.

Ecuador’s gang war meltdown; Mexican mafias’ colonisation of avocado farms; hitmen prowling the once peaceful streets of Chile. These are a few recent developments that have made organised crime the unavoidable question of the moment in Latin America.

But these are just the symptoms. The underlying disease: a reorganisation of the region’s criminal economies, now over a decade in the making. One that is testing democracy’s capacity to respond — and survive. 

Three market disruptions jump-started reorganisation in the 2010s. First, cocaine production nearly tripled from 2014 to 2022, as coca eradication efforts in Colombia, Peru and Bolivia stalled and cultivation expanded. Meanwhile, demand for cocaine, long dominated by the US, became more global, spreading to Europe, Africa and the Asia-Pacific. 

This had two major consequences: a rewiring of drug trafficking routes and huge windfalls. Brazil’s crime syndicate First Capital Command (PCC) made an estimated $40mn just over a decade ago. Now, since building a transcontinental pipeline to supply increased demand, it makes north of $1bn annually from cocaine alone.

Second, soaring gold prices triggered a criminal gold rush. Organised crime groups took over areas where wildcat gold miners operated, equipping and taxing them, and enabling a boom in output. In 2022, Latin America’s illegal mines accounted for over 11 per cent of global gold production (up from 6 per cent a decade earlier), out-earning cocaine in Colombia and Peru.

Last, during the 2010s, millions of Latin Americans (Venezuelans in particular) fled the dismal conditions created by mafias and mafia states. But these very crime groups made their flight into an industry, systematically taxing coyotes that ferry migrants and refugees (as well as kidnapping and ransoming migrants passing through Mexico), and reaping billions in profit annually.

Barring any sudden and unlikely ebbs in demand for the region’s illicit goods and services, reorganised crime is here to stay. And everyone should be worried, including Europe and the US. 

Reorganised crime threatens democracy — no small thing, given that Latin America remains the most democratic region in the global south. While mafias don’t seek to overthrow the government, they seed “parallel powers” — networks of corrupt politicians, judicial officials, and bureaucrats — that disable the state’s law enforcement capacities. Such powers are now surfacing in once unaffected countries and consolidating elsewhere, undermining democracy in Mexico, Honduras and Peru.

Latin America’s democracies lack a blueprint for combating transnational crime. While some see El Salvador President Nayib Bukele’s anti-gang crackdown as a model, this is a deceptive siren song — El Salvador’s mafias were poorer and weaker than those elsewhere, and the country is now an authoritarian police state.

The difficult truth: reorganised crime is in all likelihood too global a phenomenon for any one country alone to make an appreciable dent. The US and Europe should quit relegating Latin America to the back burner and prioritise partnering with regional governments to reduce the profitability and power of reorganised crime: consider the record number of fentanyl deaths, the overwhelmed US immigration system and the consequent nativist backlash.

The biggest risk is assuming that the cost of organised crime in Latin America can be contained. Left to its own devices, and subject to pure market forces, crime will keep innovating. And it will keep reorganising. 

Israel arrests 7 Israelis for allegedly spying on military sites for Iran

Authorities say group gathered intelligence on air base, air defences and power plant over 2 years

Israel said on Monday it had arrested seven Israeli citizens after uncovering a long-standing operation to spy on sensitive military and energy installations on behalf of Iran.

The arrests of the group — who Israeli authorities alleged carried out reconnaissance for more than two years in exchange for hundreds of thousands of dollars in cryptocurrency — came after Iran and Tehran-backed Hizbollah attacked Israeli military sites in recent months.

The seven Jewish Israelis, including two minors, carried out “hundreds” of intelligence-gathering missions for Iran, including photographing sensitive military installations such as the Nevatim air base, Iron Dome air defence batteries and the Hadera power plant, said Israeli authorities. They also gathered information on naval bases and other targets, authorities said.

Some of the sites have been targeted in missile attacks by Iran or the Lebanon-based Hizbollah militant group in recent months.

Iran has not so far responded to the latest spying claims.

The suspects are expected to be charged soon with major security offences, said police and the Shin Bet internal security agency. Israeli authorities alleged they were directed by two Iranian agents code-named “Alkhan” and “Orkhan”.

Israeli police said the alleged offences were “among the most serious that the state of Israel has known. The suspects acted while knowing about their actions and out of greed for money — and harmed the State of Israel and its citizens,” an official from the force said in a statement.

Israel has been battling Iran and its various proxy militias across the Middle East for more than a year, since Hamas’s cross-border assault from Gaza on October 7 2023 sparked the ongoing conflict.

Iran and Israel exchanged direct fire for the first time in their history last April, and again earlier this month when Tehran fired nearly 200 ballistic missiles at Israel. Israeli leaders have vowed a “severe” response against Iran.

Members of the suspected spy network “were aware that the intelligence they provided compromised national security and could potentially aid enemy missile attacks”, the Israel police and Shin Bet said in a statement.

Several of the suspects were arrested near the home of an “Israeli citizen” they were said to be monitoring, raising suspicions that Iran had also made prominent Israelis potential targets.

Israel’s intelligence services have been suspected of conducting espionage, sabotage and assassinations on Iranian soil for years, including the theft of the country’s nuclear archives in 2018 and the killing of chief Iranian nuclear scientist Mohsen Fakhrizadeh in 2020.

More recently, Iran accused Israel of carrying out a bombing in July at an Iranian military guesthouse in Tehran that killed Hamas leader Ismail Haniyeh.

Iranian efforts to recruit Israeli citizens are an ongoing concern for Israel’s intelligence services, including in recent weeks.

Last month a 72-year-old Israeli businessman was indicted on espionage charges after Shin Bet said he had travelled to Iran. Last week, an Israeli couple was arrested over police allegations that they had been recruited by Iranian intelligence.

However, the latest arrests point to a network of alleged spies that was unusual both for its scale and for the long period in which it was said to have operated.

“This occurred amidst ongoing conflicts in multiple arenas. It is assessed that these activities have inflicted security damage on the state,” said a Shin Bet official.

Google DeepMind’s Demis Hassabis on his Nobel Prize: ‘It feels like a watershed moment for AI’ 

Founder of the artificial intelligence R&D lab says scientific understanding can prevent mis-steps in developing the technology

In the 15 years since it was founded, Google DeepMind has grown into one of the world’s foremost artificial intelligence research and development labs. In October, its chief executive and co-founder Sir Demis Hassabis was one of three joint recipients of this year’s Nobel Prize in chemistry for unlocking a 50-year-old problem: predicting the structure of every known protein using AI software known as AlphaFold.

DeepMind, which was acquired by Google in 2014, was founded with the mission of “solving” intelligence — designing artificial intelligence systems that could mimic and even supersede human cognitive capabilities. In recent years, the technology has become increasingly powerful and ubiquitous and is now embedded in industries ranging from healthcare and education to financial and government services.

Last year, the London-based lab merged with Google Brain, the tech giant’s own AI lab headquartered in California, to take on stiff competition from its peers in the tech industry, in the race to create powerful AI.

DeepMind’s new positioning at the centre of Google’s AI development was spurred by OpenAI’s ChatGPT, the Microsoft-backed group’s chatbot that provides plausible and nuanced text responses to questions. Despite its commercial underpinnings, Google DeepMind has remained focused on complex and fundamental problems in science and engineering, making it one of the most consequential projects in AI globally. 

In the first interview of our new AI Exchange series, Hassabis — a child chess prodigy, designer of cult video game Theme Park, and a trained neuroscientist — spoke to FT’s Madhumita Murgia just 24 hours after being announced as a Nobel Prize winner. He talked extensively about the big puzzles he wants to crack next, the role of AI in scientific progress, his views on the path to artificial general intelligence — and what will happen when we get there. 

Madhumita Murgia: Having reflected on your Nobel Prize for a day, how are you feeling?

Demis Hassabis: To be honest with you, yesterday was just a blur and my mind was completely frazzled which hardly ever happens. It was a strange experience, almost like an out-of-body experience. And it still feels pretty surreal today. When I woke up this morning, I was like, is this real or not? It still feels like a dream, to be honest. 

MM: Protein folding is essentially solved, because of your work on AlphaFold models — an AI system that can predict the structure of all known proteins. What is your next grand challenge for AI to crack?

DH: There are several. Firstly, on the biology track — you can see where we are going with that with AlphaFold 3 — the idea is to understand [biological] interactions, and eventually to model a whole pathway. And, then, I want to maybe build a virtual cell at some point. 

With Isomorphic [DeepMind’s drug development spin-off] we are trying to expand into drug discovery — designing chemical compounds, working out where they bind, predicting properties of those compounds, absorption, toxicity and so on. We have great partners [in] Eli Lilly and Novartis . . . working on projects with them, which are going really well. I want to solve some diseases, Madhu. I want us to help cure some diseases.

MM: Do you have any specific diseases you’re interested in tackling?

DH: We do. We are working on six actual drug programmes. I can’t say which areas but they are the big areas of health. I hope we will have something in the clinic in the next couple of years — so, very fast. And, then, obviously, we will have to go through the whole clinical process, but at least the drug discovery part we will have shrunk massively.

MM: What about outside of biology? Are there areas you’re excited about working on?

DH: I’m very excited about our material design work: we published a paper in Nature last year on a tool called GNoME [an AI tool that discovered 2.2mn new crystals]. That’s AlphaFold 1-level of material design. We need to get to AlphaFold-2 level, which we are working on. 

We are going to solve some important conjectures in maths with the help of AI. We got the Olympiad silver medal over the summer. It’s a really hard competition. In the next couple of years, we will solve one of the major conjectures. 

And, then, on energy/climate, you saw our Graphcast weather modelling won a MacRobert award, a big honour on the engineering side. We’re investigating if we can use some of these techniques to help with climate modelling, to do that more accurately, which will be important to help tackle climate change, as well as optimising power grids and so on. 

MM: It seems like your focus is more on the application side — on work that translates into real world impact, rather than purely fundamental. 

DH: That’s probably true to say. There aren’t many challenges like protein folding. I used to call it Fermat’s last theorem of biology, equivalent. There’s not that many things that are that important and long-standing as a challenge. 

Obviously, I’m very focused on advancing artificial general intelligence [AGI] with agent-based systems. Probably, we are going to want to talk about Project Astra and the future of digital assistants, universal digital assistants, which I’m personally working on, as well, and which I consider to be on the path to AGI. 

MM: What does the AI double Nobel Prize in chemistry and physics [this year’s prize for physics went to Geoffrey Hinton and John Hopfield for their work on neural networks, the foundational technology for modern AI systems] say about the technology’s role and impact in science?

DH: It’s interesting. Obviously, nobody knows what the committee was thinking. But it’s hard to escape the idea that maybe it’s a statement the committee is making. It feels like a watershed moment for AI, a recognition that it can actually, is mature enough now, to help with scientific discovery. 

AlphaFold is the best example of that. And Geoff and Hopfield’s prizes were for more fundamental, underlying algorithmic work . . . interesting they decided to put that together, almost as double, related awards. 

For me, I hope we look back in 10 years and AlphaFold will have heralded a new golden era of scientific discovery in all these different domains. I hope that we will be adding to that body of work. I think we’re quite unique as one of the big labs in the world that actually doesn’t just talk about using it for science, but is doing it.

There’s so many cool things going on in academia as well. I was talking to someone in astrophysics, actually a Nobel Prize winner, who is using it to scan the skies for atmospheric signals and so on. It’s perfect. It’s being used in Cern. So maybe the committee wanted to recognise that moment. I think it’s pretty cool they’ve done that. 

MM: Where is your AlphaFold work going to take us next in terms of new discoveries? Have there been any interesting breakthroughs in other labs you’ve seen that you’re excited about?

DH: I was really impressed with the special issue of Science on the nuclear pore complex, one of the biggest proteins in the body, which opens and closes like a gateway to let nutrients in and out of the cell nucleus. Four studies found this structure at the same time. Three out of four papers found AlphaFold predictions [were] a key part of them being able to solve the overall structure. That was fundamental biology understanding. That was the thing that stuck out to me. 

Enzyme design is really interesting. People like [US biochemist and Nobel laureate] Francis Arnold have looked at combining AI with directed [protein] evolution. There’s lots of interesting combinations of things. Lots of top labs have been using it for plants, to see if they can make them better resistant to climate change. Wheat has tens of thousands of proteins. No one had investigated that because it would be experimentally too expensive to do that. It’s helped in all kinds of areas, it’s been wonderful to see. 

MM: I have a conceptual question about scientific endeavour. We originally thought predicting something was the be-all-and-end-all, and spent all this time and effort predicting, say, the structure of a protein. But now we can do that really quickly with machine learning, without understanding the ‘why’. Does that mean we should be pushing ourselves to look for more, as scientists? Does that change how we learn about scientific concepts?

DH: That’s an interesting question. Prediction is partly understanding, in some sense. If you can predict, that can lead to understanding. Now, with these new [AI] systems, they are new artefacts in the world, they don’t fit into normal classification of objects. They have some intrinsic capability themselves, which makes them a unique class of new tool.

My view on that is, if the output is important enough, for example, a protein structure, then that, in itself, is valuable. If a biologist is working on leishmaniasis, it doesn’t matter where they got protein structures from as long as they are correct for them to do their science work on top. Or, if you cure cancer, you’re not going to say: don’t give me that because we don’t understand it. It would be an amazing thing, without understanding it fully. 

Science has a lot of abstraction. The whole of chemistry is like that, right? It’s built on physics, and then biology emerges out of it. But it can be understood in its own abstract layer, without necessarily understanding all the physics below it. You can talk about atoms and chemicals and compounds, without fully understanding everything about quantum mechanics — which we don’t fully understand yet. It’s an abstraction layer. It already exists in science.

And biology, we can study life and still don’t know how life evolved or emerged. We can’t even define it properly. But these are massive fields: biology, chemistry, and physics. So it’s not unusual in a sense — AI is like an abstraction layer. The people building the programs and networks understand this at some physics level but, then, this emergent property comes out of it, in this case, predictions. But you can analyse the predictions on their own at a scientific level.

Having said all of that, I think understanding is very important. Especially as we get closer to AGI. I think it will get a lot better than it is today. AI is an engineering science. That means you have to build the artefact first and then you can study it. It’s different to a natural science, where the phenomenon is already there.

And just because it’s an artificial, engineered artefact doesn’t mean it will be any less complex than the natural phenomena we want to study. So you should expect it to be just as hard to understand and unpack and deconstruct an engineered artefact like a neural network. That’s happening now and we are making some good progress. There is a whole field called mechanistic interpretation, which is all about using neuroscience tools and ideas to analyse these virtual brains. I love this area and have encouraged this at DeepMind.

MM: I looked up a project you mentioned previously about a fruit fly connectome [brain map], made using neural networks. AI helped understand that natural phenomenon. 

DH: Exactly. That’s a perfect example of how these things can be combined, and then we slowly understand more and more about the systems. So, yes, it’s a great question, and I’m very optimistic we will make a lot of progress in the next few years on the understanding of AI systems. And, then, of course, maybe they can also explain themselves. Imagine combining an AlphaFold with a language capability system, and maybe it can explain a little bit about what it’s doing. 

MM: The competitive dynamics in the technology industry have intensified a lot in AI. How do you see that impacting and shaping progress in this field? Are you worried there will be fewer ideas and a focus on transformer-based large language models (LLMs)? 

DH: I think that actually a lot of the leading labs are getting narrower with what they are exploring — scaling transformers. Clearly, they are amazing and going to be a key component of ultimate AGI systems. But we have always been big believers in exploration and innovative research. We have kept our capabilities of doing that — we have by far the broadest and deepest research bench in terms of inventing the next transformer, if that’s what is required. That’s part of our scientific heritage, not just at DeepMind but also Google Brain. We are doubling down on that, as well as obviously matching everyone on engineering and scaling.

One has to do that — partly to see how far that could go, so you know what you need to explore. I’ve always believed in pushing exciting ideas to the maximum as well as exploring new ideas. You don’t know what breakthrough you need until you know the absolute limits of the current ideas.

You saw that with long context windows [a measure of how much text can be processed by an LLM at once]. It was a cool new innovation and no one else has been able to replicate that. That’s just one thing — you’ll see a lot more breakthroughs coming into our mainstream work. 

MM: You and others have said AGI is anywhere between 5 to 20 years away: what does the scientific approach look like to achieving that goal? What happens when we get there?  

DH: The scientific approach would mean focusing a lot more time and energy and thought on understanding and analysis tools, benchmarking, and evaluations. There needs to be 10 times more of that, not just from companies but also AI safety institutes. I think from academia and civil society, [too]. 

I think we need to understand what the systems are doing, the limits of those systems, and how to, then, control and guardrail those systems. Understanding is a big part of scientific method. I think that is missing from pure engineering. Engineering is just seeing — does it work? And, if it doesn’t, you try again. It’s all trial and error.

Science is what can you understand before all that happens. And, ideally, that understanding means you make less mistakes. The reason that’s important for AGI and AI is that it’s such a powerful technology you want to make as few mis-steps as you can. 

Of course, you want to be able to get it perfect, but it’s too new and fast moving. But we can definitely do a better job than, perhaps, we’ve done with past technologies. I think we need to do that with AI. That’s what I’d advocate.

When we get nearer to AGI, maybe a few years out, then a societal question comes, which also could be informed by the scientific method. What values do we want these systems to have? What goals do we want to set them?

So they’re sort of separate things. There’s the technical question of how do you keep the thing on track to the goal that you set? But that doesn’t help you decide what goal this should be, right? But you need both those things to be correct for a safe AGI system. 

The second one, I think, may be harder, like, what goals, what values and so on — because that’s more of a UN or geopolitical question. I think we need a broad discussion on that, with governments, with civil society, and academia, all parts of society — and social science and philosophy, even, as well. 

And I try and engage with all those types of people, but I’m a bit unusual in that sense. I am trying to encourage more people to do that or at least act as a role model, act as a conduit to bring those voices around the table. 

I think we should start now because, even if AGI is 10 years away, and some people think it could be a lot sooner, that’s not a lot of time.

Gapping down
In reaction to earnings/guidance:

Other news:

Gapping up
In reaction to earnings/guidance:

Other news: