Why FAITH?

FAITH Lab stands for the Foundation of AI Lab at TsingHua. The acronym is deliberate. Beyond describing what we study and where we are, the word faith expresses how I understand scientific research and what I hope every student in the lab will develop.

Science is often described as a purely rational process: collect data, derive a theory, then test its predictions. This account captures how science examines a claim, but not how inquiry begins. At the research frontier, the available evidence is incomplete. More than one explanation may remain compatible with what is known. The importance of a question cannot be demonstrated in advance, nor can the value of a new direction be calculated with certainty. If researchers acted only after the evidence had already determined the path, research could only follow established roads.

Scientific inquiry therefore requires a commitment that comes before certainty.

Faith is the commitment that makes inquiry possible before proof is available.

Where evidence ends

Evidence constrains scientific theories, but it does not mechanically produce them. The same observations can remain compatible with different explanations, a problem known as the underdetermination of theory by evidence (Stanford, 2017). This does not make theory choice arbitrary. It shows why scientific judgment remains indispensable.

A researcher has to decide which anomaly deserves attention, which abstraction reveals the essential structure of a problem, and which possibility is worth years of effort. Logical consistency matters. Explanatory power matters. Simplicity can matter. None of them, however, guarantees that a research direction will succeed. At some point, the researcher has to move beyond what the existing evidence already establishes.

Faith enters at precisely this point. It is the willingness to pursue a possibility before the evidence is complete.

The leap of faith

The history of science contains many such leaps.

• Copernicus did not arrive at heliocentrism by reading an unavoidable conclusion from astronomical observations. The evidence available in his time did not uniquely compel a moving Earth, and his system did not immediately achieve decisive predictive superiority. Its appeal also came from a conviction that celestial motion could be understood through a more coherent order. Harmony was not a substitute for calculation. It guided the search for a structure that calculation could later examine (Rabin, 2022).
• Einstein’s work offers another example. Relativity responded to real tensions in physics, yet its fundamental concepts were not distilled automatically from experiments. Einstein described the conceptual foundations of physics as free inventions: they are created rather than induced, then judged through their contact with the empirical world (Einstein, 1936).

These cases were not departures from reason. They reveal a limit of deduction. Evidence can expose a problem. It can eliminate inadequate answers. It cannot always supply the conceptual form of the answer that should replace them. Scientific discovery needs an imaginative step from what is observed to what might explain it.

From faith to knowledge

The leap is followed by a return to evidence. An initial commitment becomes scientific when it can be articulated clearly, examined by others, and tested against reality. Reason develops its consequences. Experiments put them to the test. Evidence reveals what should be retained, refined, or rejected.

Validation does not mean that every initial belief will be confirmed. It means that the belief is given a form in which reality can answer it. Confirmation, revision, and rejection can all advance inquiry. Through this process, a personal commitment becomes public and revisable knowledge.

What FAITH means to us

At FAITH Lab, faith begins with a shared foundation. I hope that members of the lab will broadly recognize three commitments:

• Faith in the intelligibility of the world. We believe reality contains structure that can be discovered. In our field, this means that machine intelligence is not merely a black box to be used through trial and error. It has principles, mechanisms, and boundaries that can be studied and understood.
• Faith in our research mission. We believe artificial intelligence can extend human intelligence beyond the limited reach of any individual. It can make accumulated human knowledge and experience useful to people far beyond their original context. This is what we mean by scaling human intelligence.
• Faith in the research process. Important questions rarely prove in advance that they can be solved. A researcher chooses a direction under uncertainty, then transforms an initial conviction into testable knowledge through sustained work.

These commitments provide the lab with a common orientation, but the concrete meaning of faith can only be developed by each student. Every student should gradually form an understanding of their own: which questions matter, which direction deserves sustained effort, and what standards should guide their work. This understanding emerges through study, experiment, failure, and reflection.

I believe that forming such a conviction gives each student the direction and resilience to continue making progress through uncertainty.

Faith determines which uncertain future is worth pursuing.