Return-Path: Delivered-To: alvarochf@traceseis.com Received: from cs510.bluehost.com by cs510.bluehost.com with LMTP id sIYVHTZivWlE2wYAWiBQxw (envelope-from ) for ; Fri, 20 Mar 2026 09:05:26 -0600 Return-path: Envelope-to: alvarochf@traceseis.com Delivery-date: Fri, 20 Mar 2026 09:05:26 -0600 Received: from v546.v5375b7fa.use4.send.mailgun.net ([159.112.244.46]:36392) by cs510.bluehost.com with esmtps (TLS1.2) tls TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256 (Exim 4.98.2) (envelope-from ) id 1w3bPi-000000020hu-0IrA for alvarochf@traceseis.com; Fri, 20 Mar 2026 09:05:26 -0600 DKIM-Signature: a=rsa-sha256; v=1; c=relaxed/relaxed; d=mg2.substack.com; q=dns/txt; s=mailo; t=1774019105; x=1774026305; h=List-Unsubscribe-Post: List-Unsubscribe: List-Post: List-Id: List-Archive: List-Owner: Reply-To: In-Reply-To: References: Sender: Sender: Date: Message-Id: To: To: From: From: Subject: Subject: Content-Type: Mime-Version; bh=9mJ1DbXBKWdyry/5ZBKlW8O6eljGlxALp4J1yBo66EY=; b=kjvKciws61REvvSKot5shkB39ksgPwOJRAtzLM1kijtkm5WMKjcfxNpOP6s9xMaWradCJqWj6lmIGyvDyRFkpXhiqehZO1AmP9gayBXIwH6jno551Il73HZCq8VnwgNexAcML2yqd7bwR8uySEi6o+r9jog1YJwTZAX/FYzFZ58= X-Mailgun-Sid: WyI3MDdlYyIsImFsdmFyb2NoZkB0cmFjZXNlaXMuY29tIiwiNGQxODU4Il0= Received: by 98d5f46fafb2ab4cad713e22ba75b8cc1f6bbc32e54fa20c45464d73648ca9da with HTTP id 69bd621f8a6bb5c0e2edcaf0; Fri, 20 Mar 2026 15:05:03 GMT X-Mailgun-Sending-Ip: 159.112.244.46 X-Mailgun-Batch-Id: 69bd621fb2c138bc4a856fab Mime-Version: 1.0 Content-Type: multipart/alternative; boundary="4c6858d1ef593dce85290ec1e03701321a5ec12510c253b8f4ea227511cd" Subject: Science Needs Scientists From: Cosmos Institute To: alvarochf@traceseis.com X-Mailgun-Tag: post X-Mailgun-Track-Clicks: false Message-Id: <20260320150320.3.1ed86f6d2fb0e465@mg2.substack.com> Date: Fri, 20 Mar 2026 15:03:20 +0000 Feedback-Id: post-191347337:cat-post:pub-2225794:substack Sender: Cosmos Institute References: In-Reply-To: Reply-To: Cosmos Institute List-Owner: List-Url: List-Archive: List-Id: List-Post: List-Unsubscribe: List-Unsubscribe-Post: List-Unsubscribe=One-Click X-Mailgun-Variables: {"category": "post", "email_generated_at": "1774019102170", "is_freemail": "true", "post_audience": "everyone", "post_id": "191347337", "post_type": "newsletter", "pub_community_enabled": "true", "publication_id": "2225794", "subdomain": "cosmosinstitute", "user_id": "347258985"} --4c6858d1ef593dce85290ec1e03701321a5ec12510c253b8f4ea227511cd Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: quoted-printable View this post on the web at https://blog.cosmos-institute.org/p/science-ne= eds-scientists Today=E2=80=99s essay is a guest post by Iulia Georgescu [ https://substack= =2Ecom/redirect/0721cd82-fc0e-4512-852= 5-e6e7f5befd87?j=3DeyJ1IjoiNXFxeXF4In0.= h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ], a physicist and independent = scholar researching the history of computational physics. =E2=80=9CArtificial Intelligence will revolutionize science.=E2=80=9D The p= hrase is usually invoked by researchers when discussing the promise of thin= king machines to help us understand the natural world. In the early 2020s, = I subscribed to, and was perhaps even evangelizing for, the role of AI in s= cientific research. Like many of my colleagues, I was gripped by a new tool= that promised so much. More recently my view has changed. I still believe that AI will be fantasti= cally useful, but not necessarily in the way we think. Discovery, after all= , is not the same as understanding. As scientists, we need to engage with t= he process of inquiry to truly make sense of what we learn about the world.= Only then can we understand. The Long History of =E2=80=9CAI for Science=E2=80=9D=20 In the summer of 1953, physicists Enrico Fermi, John Pasta, and mathematici= ans Stanislaw Ulam and Mary Tsingou ran the first =E2=80=9Cnumerical experi= ment=E2=80=9D on MANIAC, one of the early electronic computers built at the= Los Alamos National Laboratory after World War II. Computers were new and = scientists were excited to explore their use in solving research problems l= ike simulating the dynamics of atoms and molecules, studying tumor cell pop= ulations and investigating numerically fluid dynamics. They even created th= e first documented chess-playing program that defeated a human in the game. This group decided to model a chain of oscillators (identical masses connec= ted by springs), add a small nonlinearity (a quadratic or cubic term), and = see what would happen. The expectation was that the system would reach equi= librium (common sense would predict that there is some wiggling around, but= ultimately energy ends up equally distributed throughout the springs). But= the results were surprising. Instead, the system showed a recurrent behavi= or where the energy spreads throughout the chain, then comes back, before s= preading out again. This observation sparked interest in the study of nonli= near systems that would later lead, among other things, to the discovery of= solitons (waves that travel freely preserving their shape) and the develop= ment of chaos theory in the 1960s-1970s. If one was to assign a birthyear to computational physics, that honorific s= hould go to 1953 where Fermi and colleagues made for the first time an unex= pected discovery through a purely computational approach. Computers had bee= n used for physics and astronomy calculations before =E2=80=93 for example,= in the 1930s mechanical IBM accounting machines were modified to solve the= differential equations of planetary motion by numerical integration =E2=80= =93 and had played a key role in the development of the atomic bomb during= the Manhattan project. In the following decades computers would transform = scientific research. Today, there are almost no advances in physics that ar= e not enabled by some aspect of computer simulation. Three years after the =E2=80=9Cfirst numerical experiment=E2=80=9D the term= artificial intelligence was coined and a first attempt was made to use aut= omated reasoning to prove mathematical theorems through the Logic Theorist = system in 1956. The first AI =E2=80=9Cexpert=E2=80=9D system, DENDRAL, was = created in 1965. Combining a knowledge base with a reasoning engine, it was= capable of determining the molecular structure of a compound from its mass= spectra. Here, =E2=80=9Cexpert=E2=80=9D refers to a kind of AI system that= encoded domain knowledge from human experts as rules and applied them thro= ugh an inference engine. Other examples include LHASA, a program designed in 1972 to discover sequen= ces of reactions to synthesize a molecule; and the Automated Mathematician = in 1977, a heuristic-based program designed to discover new mathematical co= ncepts and theorems; and other expert systems like MYCIN (to identify bacte= ria) and PROSPECTOR (for mineral analysis). In the late 1980s and early 199= 0s, artificial neural networks were used to tackle problems in particle phy= sics, astrophysics, and =E2=80=93=E2=80=93 in a glimpse of what was to come= =E2=80=93=E2=80=93 predicting protein structure. By the 2000s, AI methods = were in use in many areas of physics and even assisted data analysis leadin= g to the discovery of the Higgs boson particle. In the 2010s and 2020s, scientists began to use a wave of powerful neural n= etwork-based tools to solve research problems. Some of these advances, such= as Google DeepMind=E2=80=99s AlphaFold protein prediction system, made hea= dlines around the world. Others were important but less glamorous. Machine-= learned interatomic potentials, for example, significantly speed up materia= ls and chemistry simulations yet remain unknown outside the research commun= ity. But if =E2=80=9CAI for science=E2=80=9D has a long history, why has it only= recently started to make the news? In his 2003 book, Douglas S. Robertson = coined the term =E2=80=9Cphase change=E2=80=9D to describe a radical change= that an instrument makes possible compared to the prior state of the art. = AI may be a phase change in how we do science, but not only because of powe= rful individual tools. It is their extreme accessibility, from large langua= ge models to the AlphaFold Protein Structure Database, that sees advances i= n one field become instruments of exploration in others. Epistemic Enhancers Scientific instruments provide extensions to humans=E2=80=99 senses. Telesc= opes and microscopes allowed us to see the far and the small. This type of = enhancement is known as extrapolation. Another type of epistemic support is= conversion, that is, transforming one modality into another (like sound in= to a visual image). Finally, we have those instruments that extend our capa= cities by giving access to phenomena that elude our senses, such as detecti= ng radiation or magnetic fields. This is known as augmentation. These three= moves =E2=80=93 extrapolation, conversion, and augmentation =E2=80=93 are = all types of epistemic enhancers. A central motivation behind scientific practice is understanding. But what = is scientific understanding? The 2009 book Scientific Understanding edited = by Henk W. de Regt, Sabina Leonelli, Kai Eigner suggests that there is no u= niversal definition of scientific understanding. Understanding in physics d= iffers from understanding in biology or engineering. Even within my own dis= cipline, physics, what we mean by understanding is not straightforward. Whi= le clearly related to the explanation of a phenomenon, understanding is not= precisely the same thing. There is a difference between understanding the = phenomena and understanding the theories or models that explain that phenom= enon. A prerequisite for understanding is discovery insofar as one cannot underst= and a phenomenon that has not been observed. Discovery is the process or pr= oduct of successful scientific inquiry. Objects of discovery are things, ev= ents, processes, causes, and properties as well as theories and hypotheses = and their features. The 1987 book Scientific Discovery proposed that discov= ery in science can be broken down into problem solving tasks and therefore = can be automated with computers. The authors of the book built four types of programs to look for quantitati= ve or qualitative laws and structural models. Then they used them individua= lly and combined them to rediscover many laws in physics and chemistry. Whi= le the authors recognized that there is no unique process that accounts for= scientific discovery, they showed that most of these can be cast as proble= m solving tasks that can be tackled with heuristics. Their method, based on= tree searches, was general in theory but limited in practice. This was bec= ause the combinatorial explosion of possible paths made them intractable fo= r problems beyond a modest scale. Both traditional numerical methods and modern AI tools enhance epistemic ex= trapolation and conversion. The application of such methods represents a di= fference in kind as well as magnitude for scientific practice. As Paul Hum= phreys put it: =E2=80=9CThis extrapolation of our computational abilities t= akes us to a region where the quantitatively different becomes the qualitat= ively different.=E2=80=9D This is because these simulations cannot be carri= ed out in practice except in regions of computational speed far beyond the = capacities of humans. This dual articulation of the quantitative and the qu= alitative could be taken as an argument for why modern AI tools ought to us= her in a new way of doing science. In the 1960s mathematicians Bryan John Birch and Peter Swinnerton-Dyer used= the EDSAC-2 computer at the University of Cambridge Computer Laboratory to= run numerical calculations that allowed them to state a conjecture about t= he set of rational solutions to equations defining an elliptic curve. This = is now known as Birch and Swinnerton-Dyer (BSD) conjecture and is one of th= e seven Millennium Prize Problems [ https://substack.com/redirect/fdc5544f-= 3e49-4699-8a0b-f7a6823c95bc?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYW= klSu8UKGoFv6eWS-HyoA ]. The computer helped them explore an abstract space = so they could formulate the conjecture. This computer-assisted discovery fa= lls in the final category of epistemic enhancers mentioned above: augmentat= ion. If the formulation of the BSD conjecture is a weaker example of augmentatio= n, more recently AI methods helped mathematicians make a breakthrough towar= ds solving it. Using machine learning, mathematicians discovered unexpected= oscillatory patterns in the parameter space key to BSD (of hundreds of dim= ensions) which led to the =E2=80=9Cmurmuration=E2=80=9D conjectures [ https= ://substack.com/redirect/6e11ea49-b953-4e22-a539-89f2efb330f2?j=3DeyJ1IjoiN= XFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ]. Progress has also= been made towards solving another Millennium Prize Problem as AI tools wer= e used to find potential singularities [ https://substack.com/redirect/223b= b920-410f-4b3f-86d7-28b248f8da2c?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluP= wbMYWklSu8UKGoFv6eWS-HyoA ] in the Navier-Stokes equations. Although automated theorem proving programs have been around for 80 years, = these examples do suggest that we are witnessing, in Robertson=E2=80=99s te= rms, a phase change. The quantitatively different becomes the qualitatively= different in a more interesting way. It=E2=80=99s speed, yes, but it=E2=80= =99s also breadth. Take AlphaFold and the AlphaFold Protein Structure Datab= ase it made possible. What makes it revolutionary is not only the improved = accuracy of the protein structure prediction, but also its usability by sci= entists from many fields (who can access hundreds of millions of protein st= ructure predictions). We are yet to see the full extent of what AI tools can do for scientific re= search, but one differentiator between them and traditional scientific inst= ruments, including computer simulation, is that they combine epistemic enha= ncers to explore vast dimensions of information. Remember DENDRAL, the earl= y AI system determining the molecular structure of a compound from its mass= spectra? Now, imagine the power of its modern incarnation which has access= to several databases of molecular spectra, can cross-check with published = scientific literature, and has the ability to write code to perform additio= nal calculations if needed. Or consider a powerful telescope that discovers a new exoplanet. Here an AI= tool analyzes the spectra, determines the likely composition of the exopla= net atmosphere, compares it with the existing recorded information, runs si= mulations with the known and inferred parameters to better characterize the= exoplanet and adds it to the catalog. All of these individual steps are al= ready done separately and for each the use of AI tools can be seen as an ev= olutionary improvement. But the combination of these improvements, working = together across different sources of information is potentially revolutiona= ry.=20 If I=E2=80=99m permitted to add a new epistemic enhancer category for AI as= an instrument the most appropriate would be integration (as in synthesizin= g across large amounts of very different types of information). When a tool= synthesizes across hundreds of heterogeneous sources simultaneously, the i= nferential path from evidence to conclusion can become challenging for the = scientist receiving the result to trace. The more dimensions integration sp= ans, the wider this gap between discovery and reconstruction becomes. Where= earlier epistemic enhancers extended what scientists could see or calculat= e, integration may increasingly determine what they conclude. Our Role as Scientists At the end of his book, Humphreys argued that computer simulation brings a = =E2=80=9Cshift of emphasis in the scientific enterprise away from humans.= =E2=80=9D By the 1980s, researchers already knew that, at least in part, sc= ientific discovery and problem generation could be successfully automated. = Modern AI methods can do the same thing, only much faster and much better. = I regularly hear from colleagues that in the coming years AI will take over= many of the tasks associated with scientific inquiry. Use of AI will likely mean more discoveries overall, but to discover is not= to understand. Henk de Regt argued [ https://substack.com/redirect/237f7f2= 3-af6d-4b99-b3fa-a78e8f312b82?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbM= YWklSu8UKGoFv6eWS-HyoA ] that: =E2=80=9Cscientists achieve understanding of= phenomena by basing their explanations on intelligible theories. The intel= ligibility of theories is related to scientists=E2=80=99 abilities: theorie= s are intelligible if scientists have the skills to use those theories in f= ruitful ways.=E2=80=9D Quantum mechanics is an example of a very effective,= yet not particularly intelligible theory, that despite the practical succe= sses still eludes physicists in its interpretation. If AI tools can optimiz= e discovery and generate explanations, they could perhaps also help produce= more intelligible [ https://substack.com/redirect/38a25789-f25e-4ce6-96b0-= 71c916fe55ee?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS= -HyoA ] theories. Yet intelligibility is not a property of theories themselves. As de Regt su= ggests, it depends on the capacities of the scientists who use them. A theo= ry becomes intelligible only when scientists acquire the skills to explore = its implications and apply it fruitfully. Even if machines make discoveries= and generate explanations, this kind of understanding still depends on the= participation of human investigators. Put differently: scientists need sci= ence just as science needs scientists. Cosmos Institute [ https://substack.com/redirect/72065885-ece6-4d47-8bcb-e1= 89c586a365?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-H= yoA ] is the Academy for Philosopher-Builders, technologists building AI fo= r human flourishing. We run fellowships, fund fast prototypes, and host sem= inars with institutions like Oxford, Aspen Institute, and Liberty Fund. Unsubscribe https://substack.com/redirect/2/eyJlIjoiaHR0cHM6Ly9ibG9nLmNvc21= vcy1pbnN0aXR1dGUub3JnL2FjdGlvbi9kaXNhYmxlX2VtYWlsP3Rva2VuPWV5SjFjMlZ5WDJsa0= lqb3pORGN5TlRnNU9EVXNJbkJ2YzNSZmFXUWlPakU1TVRNME56TXpOeXdpYVdGMElqb3hOemMwT= URFNU1UQXhMQ0psZUhBaU9qRTRNRFUxTlRVeE1ERXNJbWx6Y3lJNkluQjFZaTB5TWpJMU56azBJ= aXdpYzNWaUlqb2laR2x6WVdKc1pWOWxiV0ZwYkNKOS4tVFdIRmNNOHRkQ0loVTdDRmtubG1BQ2l= 0STBENE5yaW04eVBXMldBVzd3IiwicCI6MTkxMzQ3MzM3LCJzIjoyMjI1Nzk0LCJmIjp0cnVlLC= J1IjozNDcyNTg5ODUsImlhdCI6MTc3NDAxOTEwMSwiZXhwIjoyMDg5NTk1MTAxLCJpc3MiOiJwd= WItMCIsInN1YiI6ImxpbmstcmVkaXJlY3QifQ.klWp5c-tK7CjcoDAmI2dojbZ4Z2TyhS830-JD= YHnAJo? --4c6858d1ef593dce85290ec1e03701321a5ec12510c253b8f4ea227511cd Content-Type: text/html; charset="utf-8" Content-Transfer-Encoding: quoted-printable Science Needs Scientists= 3D""
And Scientists Need Science
͏     ­͏ =     ­͏     ­͏     = 3;͏     ­͏     ­͏   = 199; ­͏     ­͏     ­͏ &= nbsp;   ­͏     ­͏     ­= ;͏     ­͏     ­͏   Q= 99; ­͏     ­͏     ­͏ &n= bsp;   ­͏     ­͏     ­= ͏     ­͏     ­͏   ̳= 9; ­͏     ­͏     ­͏ &nb= sp;   ­͏     ­͏     ­&= #847;     ­͏     ­͏    = ; ­͏     ­͏     ­͏ &nbs= p;   ­͏     ­͏     ­&#= 847;     ­͏     ­͏    = ­͏     ­͏     ­͏  = ;   ­͏     ­͏     ­= 47;     ­͏     ­͏     = ­͏     ­͏     ­͏  =   ­͏     ­͏     ­T= 7;     ­͏     ­͏     &= #173;͏     ­͏     ­͏   =   ­͏     ­͏     ­͏= ;     ­͏     ­͏     &#= 173;͏     ­͏     ­͏   &= #8199; ­͏     ­͏     ­͏=     ­͏     ­͏     = 73;͏     ­͏     ­͏   &#= 8199; ­͏     ­͏     ­͏ =     ­͏     ­͏     = 3;͏     ­͏     ­͏   = 199; ­͏     ­͏     ­͏ &= nbsp;   ­͏     ­͏     ­= ;͏     ­͏     ­͏   Q= 99; ­͏     ­͏     ­͏ &n= bsp;   ­͏     ­͏     ­= ͏     ­͏     ­͏   ̳= 9; ­͏     ­͏     ­͏ &nb= sp;   ­͏     ­͏     ­&= #847;     ­͏     ­͏    = ; ­͏     ­͏     ­͏ &nbs= p;   ­͏     ­͏     ­&#= 847;     ­͏     ­͏    = ­͏     ­͏     ­͏  = ;   ­͏     ­͏     ­= 47;     ­͏     ­͏     = ­͏     ­͏     ­͏  =   ­͏     ­͏     ­T= 7;     ­͏     ­͏     &= #173;͏     ­͏     ­͏   =   ­͏     ­͏     ­͏= ;     ­͏     ­͏     &#= 173;͏     ­͏     ­͏   &= #8199; ­͏     ­͏     ­͏=     ­͏     ­͏     = 73;͏     ­͏     ­͏   &#= 8199; ­͏     ­͏     ­͏ =     ­͏     ­͏     = 3;͏     ­͏     ­͏   = 199; ­͏     ­͏     ­͏ &= nbsp;   ­͏     ­͏     ­= ;͏     ­͏     ­͏   Q= 99; ­͏     ­͏     ­͏ &n= bsp;   ­͏     ­͏     ­= ͏     ­͏     ­͏   ̳= 9; ­͏     ­͏     ­͏ &nb= sp;   ­͏     ­͏     ­&= #847;     ­͏     ­͏    = ; ­͏     ­͏     ­͏ &nbs= p;   ­͏     ­͏     ­&#= 847;     ­͏     ­͏    = ­͏     ­͏     ­͏  = ;   ­͏     ­͏     ­= 47;     ­͏     ­͏     = ­͏     ­͏     ­͏  =   ­
<= td>=
Forwarded this email? <= a class=3D"pencraft pc-reset decoration-underline-ClTkYc reset-IxiVJZ" href= =3D"https://substack.com/redirect/2/eyJlIjoiaHR0cHM6Ly9ibG9nLmNvc21vcy1pbnN= 0aXR1dGUub3JnL3N1YnNjcmliZT91dG1fc291cmNlPWVtYWlsJnV0bV9jYW1wYWlnbj1lbWFpbC= 1zdWJzY3JpYmUmcj01cXF5cXgmbmV4dD1odHRwcyUzQSUyRiUyRmJsb2cuY29zbW9zLWluc3Rpd= HV0ZS5vcmclMkZwJTJGc2NpZW5jZS1uZWVkcy1zY2llbnRpc3RzIiwicCI6MTkxMzQ3MzM3LCJz= IjoyMjI1Nzk0LCJmIjp0cnVlLCJ1IjozNDcyNTg5ODUsImlhdCI6MTc3NDAxOTEwMSwiZXhwIjo= yMDg5NTk1MTAxLCJpc3MiOiJwdWItMCIsInN1YiI6ImxpbmstcmVkaXJlY3QifQ.HryorGlZGsS= L-eXeo0ZxiCTJpPrlUIwt2fy0kow17U4?" style=3D"list-style: none;color: unset;t= ext-decoration: unset;margin: 0;-webkit-text-decoration-line: underline;tex= t-decoration-line: underline;">Subscribe here for more

Science Needs Scientists

And Scientists Need Science

Cosmos Institute=
=
 
=
3D""
3D""
3D""<= /a>
READ IN APP3D""
 

Today’s essay is a guest post by Iulia = Georgescu, a physicist and independent scholar researching the hi= story of computational physics.

=3D""
Harmony by= Remedios Varo (1956)

“= ;Artificial Intelligence will revolutionize science.” The phrase is u= sually invoked by researchers when discussing the promise of thinking machi= nes to help us understand the natural world. In the early 2020s, I subscrib= ed to, and was perhaps even evangelizing for, the role of AI in scientific = research. Like many of my colleagues, I was gripped by a new tool that prom= ised so much.

More recently my view has changed. I still be= lieve that AI will be fantastically useful, but not necessarily in the way = we think. Discovery, after all, is not the same as understanding. As scient= ists, we need to engage with the process of inquiry to truly make sense of = what we learn about the world. Only then can we understand.

The Long History of “AI for Science”

In th= e summer of 1953, physicists Enrico Fermi, John Pasta, and mathematicians S= tanislaw Ulam and Mary Tsingou ran the first “numerical experiment= 221; on MANIAC, one of the early electronic computers built at the Los Alam= os National Laboratory after World War II. Computers were new and scientist= s were excited to explore their use in solving research problems like simul= ating the dynamics of atoms and molecules, studying tumor cell populations = and investigating numerically fluid dynamics. They even created the first d= ocumented chess-playing program that defeated a human in the game.

This group decided to model a chain of oscillators (identical masse= s connected by springs), add a small nonlinearity (a quadratic or cubic ter= m), and see what would happen. The expectation was that the system would re= ach equilibrium (common sense would predict that there is some wiggling aro= und, but ultimately energy ends up equally distributed throughout the sprin= gs). But the results were surprising. Instead, the system showed a recurren= t behavior where the energy spreads throughout the chain, then comes back, = before spreading out again. This observation sparked interest in the study = of nonlinear systems that would later lead, among other things, to the disc= overy of solitons (waves that travel freely preserving their shape) and the= development of chaos theory in the 1960s-1970s.

If one was= to assign a birthyear to computational physics, that honorific should go t= o 1953 where Fermi and colleagues made for the first time an unexpected dis= covery through a purely computational approach. Computers had been used for= physics and astronomy calculations before – for example, in the 1930= s mechanical IBM accounting machines were modified to solve the differentia= l equations of planetary motion by numerical integration – and had p= layed a key role in the development of the atomic bomb during the Manhattan= project. In the following decades computers would transform scientific res= earch. Today, there are almost no advances in physics that are not enabled = by some aspect of computer simulation.

Three years af= ter the “first numerical experiment” the term artifi= cial intelligence was coined and a first attempt was made to use= automated reasoning to prove mathematical theorems through the Logic Theor= ist system in 1956. The first AI “expert” system, DENDRAL, was = created in 1965. Combining a knowledge base with a reasoning engine, it was= capable of determining the molecular structure of a compound from its mass= spectra. Here, “expert” refers to a kind of AI system that enc= oded domain knowledge from human experts as rules and applied them through = an inference engine.

Other examples include LHASA, a= program designed in 1972 to discover sequences of reactions to synthesize = a molecule; and the Automated Mathematician in 1977, a heuristic-based prog= ram designed to discover new mathematical concepts and theorems; and other = expert systems like MYCIN (to identify bacteria) and PROSPECTOR (for minera= l analysis). In the late 1980s and early 1990s, artificial neural networks = were used to tackle problems in particle physics, astrophysics, and –= – in a glimpse of what was to come –– predicting protein = structure. By the 2000s, AI methods were in use in many areas of physics an= d even assisted data analysis leading to the discovery of the Higgs boson p= article.

In the 2010s and 2020s, scientists began to use a = wave of powerful neural network-based tools to solve research problems. Som= e of these advances, such as Google DeepMind’s AlphaFold protein pred= iction system, made headlines around the world. Others were important but l= ess glamorous. Machine-learned interatomic potentials, for example, signifi= cantly speed up materials and chemistry simulations yet remain unknown outs= ide the research community.

But if “AI for scie= nce” has a long history, why has it only recently started to make the= news? In his 2003 book, Douglas S. Robertson coined = the term “phase change” to describe a radical change that an in= strument makes possible compared to the prior state of the art. AI may be a= phase change in how we do science, but not only because of powerful indivi= dual tools. It is their extreme accessibility, from large language models t= o the AlphaFold Protein Structure Database, that sees advances in one field= become instruments of exploration in others.

Epistemic Enhancers

Scientific instruments provide extensions to humansR= 17; senses. Telescopes and microscopes allowed us to see the far and the sm= all. This type of enhancement is known as extrapolation. Another type of epistemic support is conversion, = that is, transforming one modality into another (like sound into a visual i= mage). Finally, we have those instruments that extend our capacities by giv= ing access to phenomena that elude our senses, such as detecting radiation = or magnetic fields. This is known as augmentation. Th= ese three moves – extrapolation, con= version, and augmentation – are all types of epistemic enh= ancers.

A central motivation behind scientific= practice is understanding. But what is scientific un= derstanding? The 2009 book Scientific Understanding e= dited by Henk W. de Regt, Sabina Leonelli, Kai Eigner suggests that there i= s no universal definition of scientific understanding. Understanding in phy= sics differs from understanding in biology or engineering. Even within my o= wn discipline, physics, what we mean by understanding is not straightforwar= d. While clearly related to the explanation of a phenomenon, understanding = is not precisely the same thing. There is a difference between understandin= g the phenomena and understanding the theories or models that explain that = phenomenon.

A prerequisite for understanding i= s discovery insofar as one cannot understand a phenomenon that has not been= observed. Discovery is the process or product of successful scientific inq= uiry. Objects of discovery are things, events, processes, causes, and prope= rties as well as theories and hypotheses and their features. The 1987 book = Scientific Discovery proposed that discovery in scien= ce can be broken down into problem solving tasks and therefore can be autom= ated with computers.

The authors of the book built f= our types of programs to look for quantitative or qualitative laws and stru= ctural models. Then they used them individually and combined them to redisc= over many laws in physics and chemistry. While the authors recognized that = there is no unique process that accounts for scientific discovery, they sho= wed that most of these can be cast as problem solving tasks that can be tac= kled with heuristics. Their method, based on tree searches, was general in = theory but limited in practice. This was because the combinatorial explosio= n of possible paths made them intractable for problems beyond a modest scal= e.

Both traditional numerical methods and modern AI tools e= nhance epistemic extrapolation and conversion. The application of such meth= ods represents a difference in kind as well as magnitude for scientific pra= ctice. As Paul Humphreys put it: “This extrapolation of our computat= ional abilities takes us to a region where the quantitatively different bec= omes the qualitatively different.” This is because these simulations = cannot be carried out in practice except in regions of computational speed = far beyond the capacities of humans. This dual articulation of the quantita= tive and the qualitative could be taken as an argument for why modern AI to= ols ought to usher in a new way of doing science.

In = the 1960s mathematicians Bryan John Birch and Peter Swinnerton-Dyer used th= e EDSAC-2 computer at the University of Cambridge Computer Laboratory to ru= n numerical calculations that allowed them to state a conjecture about the = set of rational solutions to equations defining an elliptic curve. This is = now known as Birch and Swinnerton-Dyer (BSD) conjecture and is one of the s= even If the formulation of the = BSD conjecture is a weaker example of augmentation, more recently AI method= s helped mathematicians make a breakthrough towards solving it. Using machi= ne learning, mathematicians discovered unexpected oscillatory patterns in t= he parameter space key to BSD (of hundreds of dimensions) which led to the = “murmuration” conjectures. Progress has also been made t= owards solving another Millennium Prize Problem as AI tools were used to fi= nd = potential singularities in the Navier-Stokes equations.

Although automated theorem proving programs have been around = for 80 years, these examples do suggest that we are witnessing, in Robertso= n’s terms, a phase change. The quantitatively different becomes the q= ualitatively different in a more interesting way. It’s speed, yes, bu= t it’s also breadth. Take AlphaFold and the AlphaFold Protein Structu= re Database it made possible. What makes it revolutionary is not only the i= mproved accuracy of the protein structure prediction, but also its usabilit= y by scientists from many fields (who can access hundreds of millions of pr= otein structure predictions).

<= tr>
3D""

We are yet to see the full extent of what AI tools can do for= scientific research, but one differentiator between them and traditional s= cientific instruments, including computer simulation, is that they combine = epistemic enhancers to explore vast dimensions of information. Remember DEN= DRAL, the early AI system determining the molecular structure of a compound= from its mass spectra? Now, imagine the power of its modern incarnation wh= ich has access to several databases of molecular spectra, can cross-check w= ith published scientific literature, and has the ability to write code to p= erform additional calculations if needed.

Or consider a pow= erful telescope that discovers a new exoplanet. Here an AI tool analyzes th= e spectra, determines the likely composition of the exoplanet atmosphere, c= ompares it with the existing recorded information, runs simulations with th= e known and inferred parameters to better characterize the exoplanet and ad= ds it to the catalog. All of these individual steps are already done separa= tely and for each the use of AI tools can be seen as an evolutionary improv= ement. But the combination of these improvements, working together across d= ifferent sources of information is potentially revolutionary.

If I’m permitted to add a new epistemic enhancer category = for AI as an instrument the most appropriate would be integratio= n (as in synthesizing across large amounts of very different typ= es of information). When a tool synthesizes across hundreds of heterogeneou= s sources simultaneously, the inferential path from evidence to conclusion = can become challenging for the scientist receiving the result to trace. The= more dimensions integration spans, the wider this gap between discovery an= d reconstruction becomes. Where earlier epistemic enhancers extended what s= cientists could see or calculate, integration may increasingly determine wh= at they conclude.

Our Role as Scientists

At the end of his book, Humphreys argued that computer simulation b= rings a “shift of emphasis in the scientific enterprise away from hum= ans.” By the 1980s, researchers already knew that, at least in part, = scientific discovery and problem generation could be successfully automated= =2E Modern AI methods can do the same t= hing, only much faster and much better= =2E I regularly hear from colleagues th= at in the coming years AI will take ov= er many of the tasks associated with scientific inquiry.

Use of AI will likely mean more discoveries overall, but to discover is= not to understand. Henk de Regt argued that: “scientists achi= eve understanding of phenomena by basing their explanations on intelligible= theories. The intelligibility of theories is related to scientists’ = abilities: theories are intelligible if scientists have the skills to use t= hose theories in fruitful ways.” Quantum mechanics is an example of a= very effective, yet not particularly intelligible theory, that despite the= practical successes still eludes physicists in its interpretation. If AI t= ools can optimize discovery and generate explanations, they could perhaps a= lso help produce more intelligible theories.

= Yet intelligibility is not a property of theories themselves. As de Regt su= ggests, it depends on the capacities of the scientists who use them. A theo= ry becomes intelligible only when scientists acquire the skills to explore = its implications and apply it fruitfully. Even if machines make discoveries= and generate explanations, this kind of understanding still depends on the= participation of human investigators. Put differently: scientists need sci= ence just as science needs scientists.

Cosmos Institute i= s the Academy for Philosopher-Builders, technologists building AI for human= flourishing. We run fellowships, fund fast prototypes, and host seminars w= ith institutions like Oxford, Aspen Institute, and Liberty Fund.

Upgrade to paid

 
<= td align=3D"center">3D""Like<= /a>
=
3D""Comment
<= /tr>
3D""Restack
 

© 2026 Cosmos Institute=
500 W 2nd Street, Suite 1900, Austin, TX 78701
Unsubscribe

3D"Get3D"Start

<= /td>3D""3D"" --4c6858d1ef593dce85290ec1e03701321a5ec12510c253b8f4ea227511cd--