Serendipity on Steroids
AI and autonomous agents are poised to revolutionise scientific research and innovation. This article will outline the mechanisms driving this revolution and envisage the future of scientific exploration.
Introduction
Researchers often stumble upon drug discoveries serendipitously due to the complex and unpredictable nature of biological systems. They make these discoveries while searching for something else, when incorrect assumptions lead to the correct outcome, or through sheer luck, often leading to significant medical breakthroughs.
Roentgen’s discovery of x-rays, Fleming’s discovery of penicillin, and Banting’s discovery of insulin — all serendipitous.
AI supercharges serendipitous discoveries in drug development by enabling more efficient data analysis, predictive modelling, and integration of cross-disciplinary knowledge, all crucial for uncovering unexpected therapeutic potentials.
But how?
The Nature of Discovery
Decades ago, Thomas Kuhn posited that significant scientific revolutions occur in sudden, transformative steps, not gradually. These “paradigm shifts” typically arise when the existing dominant theories struggle to account for novel findings. Of note, such shifts are often led by young scientists or those new to a troubled field, unencumbered by prevailing theories and methods.
In a recent study, Shi and Evans used context embeddings from academic publications to reveal a trend towards narrower knowledge bases in research. This finding may explain phenomena like Eroom’s Law (drug discovery is becoming slower and more expensive over time) and indicates a shift in the nature of scientific progress. This challenges the assumption that accumulated knowledge should naturally lead to diverse novelty. This aligns with Kuhn’s observations on the nature of scientific revolutions.
Novelty comes in two flavours; content novelty (surprising combinations of concepts and methods) and context novelty (surprising combinations of disciplines or fields). Research exhibiting high degrees of both content and context novelty is most likely to result in “breakthroughs”. Distinguishing between these two types of novelty is essential as they capture different elements of scientific or technological advancements:
Imagine it this way: a Michelin-starred chef crafts an extraordinary dish that surprises with its innovative blend of flavors, all while utilising ingredients sourced exclusively from Italy. This epitomises content novelty and is captured by the phrase “standing on the shoulders of giants”. Nobel prize-winning papers have low context novelty, but high content novelty.
On the other hand, fusion cuisine is like a young backpacking explorer who, after journeying across continents and gathering diverse culinary insights, mixes elements from various culinary traditions to concoct dishes that are both entirely new and often astonishing. This represents context novelty. This is the stuff of “breakthroughs”.
AI’s Role in DeSci
In December last year, scientists using AI deep learning technology discovered the first new class of antibiotic in over 60 years.
Artificial Intelligence (AI) has evolved into a significant force within modern biology, bridging the realms of structural bioinformatics and genomics. It empowers drug discovery by meticulously analysing genetic information, scientific literature, and preclinical/clinical data, unveiling tailored treatments, and optimising drug design with remarkable precision.
I will discuss these in more detail in an upcoming essay, A Blueprint for Breakthroughs, but summarise some here. AI models like AlphaFold and RoseTTAFold, have accelerated R&D by achieving breakthroughs in understanding proteins. They unveiled previously hidden insights into protein structure and function, opening up new avenues for potential therapeutic interventions.
Companies such as Insilico Medicine and Adaptyv Bio have harnessed the generative abilities of AI to develop drugs with entirely novel chemical structures. AI actively explores uncharted chemical spaces, offering a new hope for more effective treatments across various diseases.
LLMs will revolutionise scientific tasks by synthesising literature and predicting properties of intricate biological systems. Their potential transcends drug discovery and extends to transformative healthcare applications, where they can streamline diagnosis, treatment recommendations, and the development of personalised medicine.
These innovations will be combined with an emerging and exciting set of blockchain-powered DeSci primitives…
BioDAOs
Vita, Hair, Valley, Athena, Cryo, and Cerebrum are revolutionising biomedical R&D with their decentralised, tokenised models, operating as meritocracies by rewarding diverse contributions beyond capital. Fine-tuned LLMs specific to a therapeutic area can be contributed to by DAO members, incentivised to submit personal data and share their experiences. They will be co-owned, co-governed and monetised within these entities whilst acting as assistants to patients on their journeys.
Intellectual Property Tokens (IPTs) and IP-NFTs
Molecule modernises IP rights management and exchange through its Intellectual Property Framework. It facilitates the fractionalisation of IP assets, increasing liquidity and access to individuals instead of biotech VCs and private equity firms. AI driven processes facilitate cross-disciplinary innovation to identify countless potential inventions that, if realised, can be used to finance further development of the assets.
Decentralised Knowledge Graphs (DKGs)
Origin Trail and Ceramic aim to overcome the limitations of traditional data silos (and, to a certain extent, blockchains) by enhancing data reproducibility, transparency, and composability. AI algorithms can perform semantic analysis on the vast, interconnected datasets stored in DKGs revealing trends and patterns not immediately apparent. This accelerates hypothesis generation.
Hypercerts
Coordination.network leverages these new economic structures to align interests at a network level, across numerous DAOs. AI can identify potential collaborations by matching research entities with complementary skills, interests, and resources and create “push mechanisms” to subsidise research inputs and tackle complex research challenges via synergistic partnerships.
The Agent Revolution
Autonomous agents will excel in novel content combinations to yield legible advances within research fields i.e. patent creation, as well as novel context combinations that generate field-violating controversy a.k.a, paradigm shifts.
Agents will supercharge serendipitous discoveries in drug development.
Multi Agent Systems (MAS) for Drug Discovery
CoScientist, a set of intelligent agents powered by multiple LLMs, plans chemical syntheses, operates cloud labs, manages liquid handling instruments, integrates data sources, and solves optimisation problems. In drug design, while LLMs are good at exploring chemical space, their outputs lack molecular diversity. A specialised framework using multiple GPT agents that play a game addresses this issue to improve diversity in molecular design.
We have written extensively about blockchain enshrined property rights giving agents true autonomy. Facilitated by frameworks such as OLAS and DAIN, MASs stand out as more than mere executors; they are visionary architects and harbingers of a new era in scientific research that transcends traditional research boundaries.
As agents navigate the deep web of DeSci, seamlessly synchronising DAO incentives, they will stimulate patent creation, and spark breakthroughs.
But what precisely positions MASs as a nexus of all this novelty?
Interdisciplinary Expeditions and Intellectual Risk-Taking
Shi and Evans formalised the notion of a “knowledge expedition”, an abductive process whereby scientists apply their expertise to entirely new fields. They quantified Kuhn’s qualitative observation on scientific revolutions.
Surprisingly, the most groundbreaking discoveries, or paradigm shifts, often emerge not from traditional interdisciplinary teams, but from those bold enough to venture across disciplinary boundaries. Such expeditions challenge the echo chambers of familiar thinking, embodying intellectual risk-taking crucial for significant advancements in science and technology.
MASs, through rapid cross-disciplinary exploration, will uncover fresh challenges and ideas, stewarding a new wave of hypotheses and breakthroughs. Enabled by crypto-economic incentives, these ventures herald an inflection point in scientific inquiry, propelling us towards a future of novel markets and enterprises (aftershocks) to service them.
Autonomous Agents + DeSci = Novelty Explosion
The role of MASs transcends facilitation; they are vanguards of interdisciplinary research. Their “knowledge expeditions’’ paired with a shift towards new approaches to drug discovery mark a significant departure from established practices. They promise a future of transformative medical breakthroughs. Utilising DeSci primitives such as IPTs, DKGs and HyperCerts, these agents will be adept at forging new markets out of these instruments — and markets move faster than humans who make decisions by committee.
AI Powered IP Machines
Science’s inward focus within dense fields has led to a paradox within which scientists collectively learn more about less. In contrast, inventors cite and search widely to know less about more. Yet, information silos and knowledge restrictions have led to a diminishing impact of patents.
DAOs and MASs emerge as catalysts for change, orchestrating work plans and marshalling creative and novel combinations of existing concepts and methods, much like Michelin-starred chefs crafting groundbreaking dishes within the confines of their culinary traditions — this is content novelty at its finest. They will nurture an explosion of patents — with agents, akin to these elite chefs, iterating on feedback to refine and hone in on viable inventions, continually tasting and adjusting their creations until they reach perfection.
Autonomous Agents Driven Breakthroughs
Deep analyses highlight the prevalent risk aversion among scientists. Those following conventional domain wisdom enjoy consistent success publishing as well as access to grant monies. The danger of pursuing innovative fringe concepts far outweighs the potential returns of any breakthrough discoveries it may yield.
Autonomous agents, unburdened by concerns for reputation or job security, are free to explore high-risk, high-reward concepts. Supported by each DAOs’ ability to lower entry barriers to ambitious projects, MASs are primed for scientific expeditions akin to globetrotting culinary experts who, in their pursuit of the next astonishing fusion dish, traverse continents to mix and match unlikely ingredients — this is context novelty in action. We sit at the dawn of an explosion in scientific breakthroughs, with DeSci agents cooking up the unforeseen and the groundbreaking.
More on this from Agent #23….soon!
An Anecdote…
During my time in the pharmaceutical industry, some of the most interesting research ideas I encountered were at the bars of scientific conferences. Here, academics lubricated by alcohol would share surprising observations from their ongoing studies, quickly followed by similar findings from a fellow drinker. These conversations quickly evolved into planning studies to test weird and wonderful hypotheses.
To quote Shi and Evans:
“It is no surprise then that abduction is routinely social, where scientists from distant fields achieve substantial impact in advancing on a topic or challenge by bringing them into conversation with alien insights and perspectives.”
Unfortunately, these late night bar dwellers resided in different institutions making it almost impossible to collaborate and materialise their cross pollinations. Enter DAOs; a social forum within which these ideas can be allowed to proliferate at their earliest stages of ideation.
They will become supercharged by decentralised knowledge graphs and autonomous agents.
References:
- Ban, T.A. (2006). The role of serendipity in drug discovery. Dialogues in Clinical Neuroscience, 8(3), 335–344.
- Kuhn, T.S. (1962). The structure of scientific revolutions (2nd ed.). University of Chicago Press.
- Shi, F., & Evans, J. (2023). Surprising combinations of research contents and contexts are related to impact and emerge with scientific outsiders from distant disciplines. Nature Communications, 14, 1641.
- Scannell, J.W., Blanckley, A., Boldon, H., & Warrington, B. (2012). Diagnosing the decline in pharmaceutical R&D efficiency. Nature Reviews Drug Discovery.
- Wong, F., Zheng, E.J., Valeri, J.A., et al. (2024). Discovery of a structural class of antibiotics with explainable deep learning. Nature, 626(177), 185.
- Jumper, J., Evans, R., Pritzel, A., et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature, 596, 583–589.
- Baek, M., McHugh, R., Anishchenko, I., et al. (2024). Accurate prediction of protein–nucleic acid complexes using RoseTTAFoldNA. Nature Methods, 21, 117–121.
- Boiko, D.A., MacKnight, R., Kline, B., et al. (2023). Autonomous chemical research with large language models. Nature, 624, 570–578.
- Hu, X., Liu, G., Zhao, Y., & Zhang, H. (2023). De novo Drug Design using Reinforcement Learning with Multiple GPT Agents. arXiv preprint arXiv:2401.06155.
- Dempsey, M. (2020, July 29). On Inflection Points. Retrieved from https://www.michaeldempsey.me/blog/2020/07/29/inflection-points/
- Sadri, A. (2023). Is Target-Based Drug Discovery Efficient? Discovery and “Off-Target” Mechanisms of All Drugs. Journal of Medicinal Chemistry, 66(18), 12651–12677.
- Park, M., Leahey, E., & Funk, R.J. (2023). Papers and patents are becoming less disruptive over time. Nature, 613, 138–144.
- Foster, J.G., Rzhetsky, A., & Evans, J.A. (2015). Tradition and Innovation in Scientists’ Research Strategies. American Sociological Review, 80(5), 875–908.
ID Theory may hold positions in some of the projects discussed in the above. This article is strictly for informational and educational purposes only. It does not in any way constitute an offer or solicitation of an offer to buy or sell any investment or cryptoassets discussed herein. Always perform your own research and conduct independent due diligence prior to making any investment decisions.
Interested in partnering with ID Theory or building something special? Get in touch through our website or at info@idtheory.io.
