Recently I ran across a discussion of Artificial Intelligence (AI) that invoked Galileo! The idea was that an AI could never have figured out that the heliocentric theory had something going for it. The discussion referenced the paper “Theory Is All You Need: AI, Human Cognition, and Causal Reasoning”, by Teppo Felin of Utah State University and the University of Oxford, and Matthias Holweg of the University of Oxford. In their abstract they write,
Scholars argue that artificial intelligence (AI) can generate genuine novelty and new knowledge and, in turn, that AI and computational models of cognition will replace human decision making under uncertainty. We disagree. We argue that AI’s data-based prediction is different from human theory-based causal logic and reasoning. We highlight problems with the decades-old analogy between computers and minds as input–output devices, using large language models [LLMs] as an example.
The idea is that, since an LLM like ChatGPT can only sift words and make statistical associations which it produces as output, it does not function in any way like a human mind. Thus an LLM cannot be expected to ever be able to really do what the human mind, like Galileo’s human mind, can do. And in particular, an LLM can’t do anything new. It can’t make a leap like Galileo did to embrace a new idea. Having recently written here about why I think we can’t equate human minds and computers, I found their discussion most interesting.
The authors write:
To illustrate the problem of generating something novel — such as new knowledge — with an LLM…. [i]magine an LLM in the year 1633, where the LLM’s training data incorporates all the scientific and other texts published by humans to that point in history. If the LLM were asked about Galileo’s heliocentric view, how would it respond? Because the LLM would probabilistically sample from the association and correlation-based word structure of its vast training data — again, everything that has so far been written (including all the scientific writings about the structure of the cosmos) — it would only restate, represent, and mirror the accumulated scientific consensus. The training dataset for the LLM would overwhelmingly feature texts with word structures supporting a geocentric view, in the form of the work of Aristotle, Ptolemy, and many others…. The evidence — as inferred from the repeated word associations found in the training data — would overwhelmingly be against Galileo. LLMs do not have any way of accessing truth (for example, through experimentation or counterfactuals) beyond mirroring and restating what is found in the text.
The authors state that even if the works of Copernicus and various Copernicans were included in the training data, that material “would be dwarfed by all the texts and materials that supported the predominant geocentric paradigm”. Two thousand years of geocentric texts “would vastly outweigh Galileo’s view, or anything supporting it.”
They continue:
An LLM’s model of truth or knowledge is solely statistical, relying on frequency and probability. Outputs are influenced by the frequency with which an idea is mentioned in the training data, as reflected by associated word structures. For example, the frequency with which the geocentric view has been mentioned, summarized, and discussed in the training data necessarily imprints itself onto the output of the LLM as truth. As the LLM has no actual grounding in truth, beyond the statistical relationships between words, it would say that Galileo’s view and belief is delusional and in no way grounded in science.
Hmmm… “delusional and in no way grounded in science”… does this sound familiar? Like “foolish and absurd in philosophy“?
A neural network like an LLM might, in fact, include any number of delusional beliefs, including beliefs that turned out to eventually be correct (such as Galileo’s), but also beliefs that objectively were (and still are) delusional. Ex ante, there is no way for an LLM to arbitrate between the two.
Felin and Holweg then mention Tycho Brahe and his astrological works. This is a good point. Lots of astronomical writings prior to 1633 would have included an astrological component. They continue:
A hypothetical LLM (in 1633) would have no way of arbitrating between Galileo’s (seeming) delusions about heliocentrism [and] Brahe’s (actual) delusions about astrology. Our hypothetical LLM would be far more likely to have claimed that Brahe’s astrological claims are true than that Galileo’s argument about heliocentrism is true. The LLM can only represent and mirror the predominant and existing conceptions — in this case, the support for the geocentric view of the universe — it finds in the frequencies and statistical association of words in its training data.
In sum, it is important to recognize that the way an LLM gets at truth and knowledge is via a statistical exercise of finding more frequent mentions of (hopefully) a true claim (in the form of statistical associations between words) and less frequent mentions of a false claim. LLM outputs are probabilistically drawn from the statistical associations of words it has encountered while being trained. When an LLM makes truthful claims, these are an epiphenomenon of the fact that true claims happened to have been made more frequently. There is no other way for the LLM to assess truth, or to reason. Truth — if it happens to emerge — is a byproduct of statistical patterns and frequencies rather than from the LLM developing an intrinsic understanding of — or ability to bootstrap or reason — what is true or false in reality.
Thus, human beings can discern truth, and can discern new truths. AIs can’t. That is a pretty interesting discussion.
So… Why did the Inquisition put Galileo on trial? Because that is what their AIs said was the right thing to do!
Click here for other AI-related posts. (The image at top is another AI product!)
