New applied metascience lab - Future House announced + some thoughts on UK applied metascience

Some very good news amidst the madness…….

During my UCL metascience talk and a couple of other posts on this blog where I explored the need for new kinds of labs with radically different incentives, career paths, and funding approaches, I alluded to something very exciting that would be being announced in the US. 

It’s announced today, funded by Eric Schmidt and is called Future House. It's the work of a team seeded around Sam Rodriques.  

In this blog I’m a) trying to draw attention to it amongst UK people and b) I’ve also put together some loose thoughts on it for a UK context. Eric Schmidt asked me in december 2022 to be a reviewer on the Future House grant proposal, as we had discussed the Lovelace vision document that year and he knew what we had been trying to do in Number Ten. So I’ve been having to keep this exciting secret for a while…..

Some background for those who don’t know him: Sam Rodriques is a very entrepreneurial scientist and technologist  - I recommend following him on twitter at @sgrodriques on X. He’s focussed on building new things to make the world better. He came over to the UK from MIT, starting a group at the Crick institute a little over two years ago at age 29. Sam’s recruitment to the Crick Institute is a great example of how creating a strong talent centered offer with long-term funding support, freeing researchers from some of the ‘grant treadmill’ issue, can attract genuinely first rate talent from top international institutions. The UK civil service actually brought Sam as one of three examples of the best people in their fields to speak with the Prime Minister’s Council for Science and Technology (without me nudging them). 

Sam was also a key figure in the Focussed Research Organisation effort, first outlining it in his PhD thesis. 

Sam is now moving to San Francisco to start this new Future House endeavour. Future House is a setup inspired in part of Xerox PARC, where Schmidt had worked. Its the topic of a Bloomberg exclusive, which I’ll put some excerpts from:

“The nonprofit, Future House, plans to develop AI tools that can analyze and summarize research papers as well as respond to scientific questions using large language models — the same technology that supports popular AI chatbots.”

Much of what Future House is aiming to do is accelerate the process of science itself - much of it can be accelerated in just the same way that Licklider saw that computational devices could automate things like copy-paste, organisation of files, search functions etc. 

AI promises to extend this to many new areas, freeing humans from much rote drudgery like designing plasmids for gene expression, summarising scientific data, and more. This would greatly improve the efficacy and efficiency of scientists and thus productivity. As they say on their launch blog:

“Biology research today is set to scale. New techniques allow us to test tens or hundreds of thousands of hypotheses in a single experiment. New tools allow us to design thousands of proteins computationally in parallel. However, the fundamental bottleneck in biology today is not just data or computational power, but human effort, too: no individual scientist has time to design tens of thousands of individual hypotheses, or to read the thousands of biology papers that are published each day.

At Future House, we aim to remove this effort bottleneck by building AI systems -- AI Scientists -- that can reason scientifically on their own. Our AI Scientists will augment human intelligence. In 10 years, we believe that the AI Scientists will allow science to scale both vertically, allowing every human scientist to perform 10x or 100x more experiments and analyses than they can today; and laterally, by democratizing access to science and to all disciplines that are derived from it.

But Future House also has a longer term ambition to create an ‘AI scientist’, that can actually do the thinking of scientists, such as looking through the literature to develop hypothesis.

“The “AI scientist,” as Future House refers to it, will one day be able to sift through thousands of scientific papers and independently compose hypotheses at greater speed and scale than humans”

It’s very interesting from an applied metascience perspective. Its explicitly inspired by places like Xerox PARC which was very different to modern academia. Schmidt describes this inspiration:

“[PARC] was a place where you got these people in their late 20s and early 30s, gave them independence and all the resources they needed, and they would invent things at a pace that you didn't get anywhere else,” Schmidt said. “What I really want is to create new environments like what PARC used to be, where outstanding young researchers can pursue their best ideas.”

Bingo. When I explained to Schmidt last year what we had been trying to do in Number Ten, I said almost exactly that. And his response was “that's what we are trying to do to at Schmidt Futures”. I failed at that enterprise but I did at least try very hard to create environments focussed on junior entrepreneurial scientists. Rodriques + co are going full steam ahead.

As we said in the Lovelace doc:

There is a key global need for junior researchers to pursue long-term research programs, to self-organise and collaboratively pursue visions of a better future through invention as they could in the postwar period that created our technological world. To achieve this… [Lovelace] design features are inspired by the striking commonalities between the early Cambridge Laboratory for Molecular Biology, Bell labs, and Xerox PARC: three model institutes which birthed new fields of research and industries in the postwar period

On the Future House launch blog they explicitly say that they want to reinvent the organisation of early stage research:

“In pursuit of our mission, we are pioneering a new approach to scientific research. Future House is an independent, non-profit research organization, headquartered in San Francisco, with the freedom to remain laser-focused on our long-term mission even if it takes us away from shorter-term, value-generating targets like drugs or industrial catalysts. We are fiercely committed to a flat structure, team science and individual contributions. We believe that the way to enable discoveries is to enable small, integrated teams of outstanding biologists and AI researchers to iterate rapidly towards big-if-true ideas. Seniority, tenure and even specific disciplines may blur as our teams forge a path forward together. As a non-profit, we also have a unique ability to prioritize responsible use of AI, which will be critical to ensure that our AI Scientists accelerate science without sacrificing safety or empowering bad actors.”

Bingo 2! Future House is directly targeted the basic organisation of researchers in a lab environment, moving away from academic structures.

Future House’s have zeroed in on a key thing: talent, the quality of it, and its utilisation is the core and most important feature of research. The system should be built around that. Currently the system prioritises quantity over quality, and has a very homogeneous approach to talent that doesn’t recognise the diversity of skills required. Phrases like “we produce X PhD students”, and “we need to increase the R&D workforce by X% by Y date” are recurrent themes in government/think tank docs but they miss the point in my view. As showed in our analysis on ‘is the UK a world leader in science?’, some much smaller countries like Switzerland, with much lower R&D funding investment in absolute terms, outperform the UK comfortably in key areas. Quality and organisational structures to empower quality is key. In my experience the UK pays very little attention to whether or not it is attracting and retaining outlier talent like Rodriques.

At the start of the year I came very close to trying to be a part of Future House and move to San Francisco, but I thought something else was on the table which didn’t work out and I prioritised that as it would have kept me in the UK. Obviously I might come to strongly regret not going, as I think what Sam and his team are doing is incredibly exciting……..

Some broader background/digressions, why didn’t the UK do similar…..

It is great for the world that more of these experiments are happening.

But what about for the UK? 

It's very helpful for people in the UK trying to do new things to have this precedent. And, in my view, Sam’s new initiative should be higher on the priority list than FROs (which I’m also a supporter of). 

But it should also cause some concern in the UK.

First, both Adam Marblestone and Sam Rodriques, two key people in driving applied metascience, were working in London as of 2 years ago. Both have now left. It's a wider trend.

More broadly it points to the UK’s failure to orient to new opportunities, though the research ventures catalyst is a move in the right direction if it is run properly. The UK remains almost entirely depending on an academic career path that is collapsing globally.

Our team tried and tried to get the UK to look to do things like FROs and Future House. Once FROs came about, people start wanting to create a UK FRO. Hopefully the same thing happens now for this. I know for certain Future House is not a one off - other major philanthropists are exploring doing these kind of nonacademic, early stage labs in other fields stateside, in addition to institutes like Arc and Arcadia. The UK system is structurally incapable of being ahead of the curve, yet routinely babbles about being world leading, being ‘strategic’, and finding areas of competitive advantage. 

There is nothing like these organisations in the UK - the independent early stage institutes we have are cushier versions of university departments with the same career structure, fixed labs. 

Notably, the only thing in the UK that would likely be described as part of the applied metascience effort by the people in the US is ARIA, and to do that a bunch of renegade scientists literally had to take over Number Ten Downing Street. That's a lot of friction.

Why are experimental labs particularly hard to create via government?

There are many reasons. But a key one is that fundamentally these kinds of new initiatives are deeply unpopular with the powers that be in the scientific establishment, and hard for them to understand. The academic career path is extremely deeply ingrained in the UK, more so than any other country, and people within it sometimes do not even recognise core assumptions they are making about how research is organised. Further, you can’t spin new labs as being useful to the interests of the status quo, as its in competition with their funding, whereas ARIA could be expected to give money to the status quo characters and so less resistance arises.

The tech philanthropists generally recognise how broken the standard academic model now is, and know something better exists, because they themselves have worked in it. Schmidt, for example, worked at both PARC and Bell labs. By contrast, our attempts to learn from those examples were derided at the time by very senior figures as ‘simplistic’ and undermined quietly. 

Notably, Sam is 32 years old. Almost all institution/lab directors in the UK are in their fifties at the earliest. There is a fundamentally different attitude and culture in the UK public R&D system. The leadership team founding OpenAI was also in their early thirties at the time. In the UK setting up an AI lab, the Turing, involved recruiting a set of senior professors, practically none of whom had any notable experience in neural nets technology, despite the reason AI was becoming such a key topic warranting a national lab was because neural nets had been shown to be far more effective! The senior figures simply had the right ornaments and kitemarks on their resume to mark them as senior, and in they went. This isn’t a criticism of them personally - it is a structural problem with the way the British state approaches R&D: it sharply overvalues volume of academic learning and prestige signals, and undervalues entrepreneurship and creativity. Then we wonder why the UK doesn’t produce the next amazon or google!

The strong impression I got in Number Ten is that R&D policy lobbying on the academic side is essentially made in the self-interest of the actors senior enough in the system. Factors like consultations, stakeholder engagement, reviews etc tend to reinforce this, not resist it. Whilst many in senior circles do advise in the national interest, because they each advise in somewhat different directions, the self interested status quo thrives. Government has to instead be a counterweight to this tendency for an ossified elite to control national policy.

I’ll probably stay in UK for at least another 2 years and see if some things can get going, otherwise its increasingly hard to watch such great stuff going on in the US whilst the senior UK science establishment works away defending its own taxpayer funded buffet.

PS - Schmidt on an emerging AI niche policy folks shd attend to!

In the bloomberg piece Schmidt also made an interesting comment that resonates with what we said in the New National Purpose AI report - that there is a gap emerging for earlier stage AI research as the tech companies orient directly to commercialisable LLMs (and, a pool of talent that wants to carry on doing it, who are actually world class researchers w/experience at frontier):

“With his financial backing, Schmidt believes Future House will be able to prioritize research rather than racing to make money. “I think getting the incentives right is especially important right now, when there’s a very high expectation that progress in AI will lead to products in the short term, which is leading a lot of the big AI research centers to focus very much on commercialization over research,” Schmidt said.”