Science and computers
Since the dawn of the computer age back in the 50s, computation has been a fundamental tool for scientists in every field. Aiding researchers to analyse, verify and test scientific data or hypotheses through mathematical means. It has greatly augmented our capacity to deal with complex problems and reduce the time needed to work out their solutions.
Pablo Picasso once famously said: “computers are useless, they can only give you answers”.
That statement reveals some truth about a duality within the scientific method. On one side there is human formulation of a question: a hypothesis; a possibility. On the other side there is its testing or actualisation through computational means.
This sets the conventional boundaries between the scientist, who is the creator of potential novelty, and the computer, which hardens that potential into fact. With the advent of Artificial Intelligence, these boundaries are slowly being blurred, leading to a shift in how discoveries happen in science.
The transformation of computation
Very subtly, the nature of computation itself is changing.
Traditionally, scientists make observations, create a theory based on those observations, and then use that theory to predict new observations.
AI, in the form of neural networks, has taken the value of observations and data to the next level. By computing immense amounts of raw data, it can infer patterns within it and produce accurate predictions bypassing the need for “middleman” first-principles, laws or assumptions given by humans. From data to prediction only through statistics. Correlations instead of cause.
All we have left is a “black box”, a network that provides predictions without revealing or needing any overarching principles.
What this all means is that computation, in this form, represents altogether a new mode by which we can access and generate knowledge. Removing human interpretation from the loop. Uplifting in theory above the limitations of human intellect. How will we see this in the real world?
What it means for scientific discovery
This Deus-ex-Machina feature is what established AI’s success from online recommendations to the centuries-old game of GO. Most recently, it entered the public scientific domain thanks to the AI software alfafold’s success at a major protein-folding competition, in which it successfully predicted a protein’s 3D structure from its amino-acid sequence.
It is in science in fact, that AI can reach its full potential in the future. Signals of it are already here.
The field of chemistry has seen in the last years the emergence of AI-designed molecules. The idea behind is to set the features of a desired molecule in the system, and then have an AI run simulations and combinations through a chemical space of potential molecules until it finds the one suitable. The dream is to be able to “dial up” molecules at will.
This can lead to the discovery of unexpected building blocks for new and powerful medicines, new molecules for chemistry that are sustainable and don’t rely on fossil sources, new materials for solar panels or batteries that dramatically improve their efficiency.
Today’s reality: AI in practice
Using AI to aid molecule discovery and fabrication is an objective of departments at Imperial College London such as ROAR (Centre for Rapid Online Analysis of Reactions) led by Professor Mimi Hii and DigiFAB ( The institute for Digital Molecular Design and Fabrication), co-lead by Professor Claire Adjiman. Their data centric approach aims at finding new molecules and chemicals and at ways to optimise their synthesis processes.
This leads to a future where trial and error phases in experimentation is dramatically reduced and new molecules can be produced in shorter time frames, with less expanse of energy and more sustainable processes.
This has a tremendous impact from the industry perspective, from reducing the wastage in product design through to quicker development of market-ready solutions.
The discovery of new materials is not the only area where AI can help, scientists have recently used neural networks to reveal connections among fields in mathematics based on patterns revealed from data. Some researchers have used AI to mix building blocks from standard quantum experiments to generate new ones.
Another example is the AI “Copernicus “, a machine-learning algorithm (ML) inspired by the brain that worked out how the Sun resides at the centre of the Solar System. Teaching itself the laws of physics based on how movements of the Sun and Mars appear from Earth. The feat is one the first tests of a technique that researchers hope they can use to discover new laws of physics, maybe reformulate quantum mechanics.
Looking at the future the signals we’re receiving
In the future, we might be able to use AI systems to navigate the space of present knowledge and theories. From this we may infer higher order levels of synthesis and novel unexpected connections among fields. Even predict new technologies. This could lead to an exponential acceleration of scientific discovery.
What if the speed of scientific discoveries outpaces our capacity to understand them?
Post-scientific research might inhabit a knowable but unexplainable reality, where engineers are guided by AI to create novel technologies without being able to fully understand how and why they work.
This creates a possible future scenario in which our capacity to functionally know and harness natural phenomena through AI will greatly surpass our capacity to grasp nature itself.
- How might a world look like, in which algorithms model and predict physics better than humans can understand?
- What might happen if our theoretical frameworks were not anymore able to explain and conceptualise AI-driven results?
- And could an AI win a Nobel Prize? We’ll leave that up to the Committee to decide…
As pointed out by researchers, the goal of any scientist is prediction. For which human interpretable laws might not be needed. It’s arguable that by doing so scientists would understand less about nature, but from a practical point of view there wouldn’t be any loss.
It’s also interesting to notice, that most of us already use in our daily life technology we don’t have the means to understand. Who can explain how a smartphone works? We can only explain what it can do, in terms of function.
Speculations aside, the way we perform science and the traditional infrastructures and modes of research, will evolve tremendously differently than the ones we know and apply today. Driven by “smarter-than-us” software and global common challenges that affect both industry and academia – such as sustainability, food security or health.
What does this mean for human ventures?
The value of science without application is questionable. Automated scientific discovery, if not prompted by research, is prompted by challenges to the planet and humanity.
Transitioning to a zero-carbon economy begs for new materials and reactions to replace those in place and to solve problems like carbon sequestration. This will include rethinking the materials used in goods, transportation, and consumer living.
Similarly, maintaining the quality of life under a recovering environment and protecting biodiversity will call for not only lower-impact materials and processes but circular consumption pathways.
Beyond the challenge of a zero-carbon economy and recovery are our intentions to take humanity to new ventures and domains. Space exploration, asteroid harvesting and interplanetary travel are a provocation for new materials.
Want to know more?
Are you interested in this future development and what it means to you and your organisation?
Check-out “Automation”, one of our recently released scenarios for 2041 in collaboration with Imperial College London academics.
Imperial Tech Foresight is foresight backed with the scientific community of Imperial College London. Get in touch to learn more about the possibilities, challenges, and opportunities ahead with such emerging technologies.