Yes, AI has reached human levels

AGI, or artificial general intelligence, has been the holy grail of AI for many years. Ever since genAI was introduced, there has been increased excitement, with commentators galore asking the question: has AI reached AGI, that is, truly human levels of communication? Well, I can state with certainty that it has indeed reached human levels in the area of scholarly discourse. But you might not be entirely satisfied with the results.

Sitting on the fence

As you no doubt know, genAI draws on a corpus of knowledge, which means that, for scholarly purposes, it accesses all available articles, books, theses, on a topic, and uses these to provide conclusions for the reader. Typically, genAI has been trained to sit on the fence when asked a big question, for example, “Was the Renaissance a religious or a secular age?”

Here is a Copilot answer to another Renaissance-related question, a topic that scholars have been arguing over for more than 50 years:

Now, consider a typical academic print book on the same subject – here is Gene Brucker, in the often reprinted classic, Renaissance Florence, discussing religion in Renaissance Florence:

Religion is the least explored aspect of Florentine Renaissance history … Studies of Florentine religion tend to fall into two categories. One interpretation … emphasizes the continuity of religious institutions and traditions, and the vitality of Florentine spirituality … At the other extreme are those historians like Jakob Burckhardt, who minimize the important of Christianity in Renaissance Italy. For Burckhardt, the most significant feature of Italian religious history in the Renaissance is the secularization of institutions and beliefs. [Brucker, Renaissance Florence, chapter 5]

Doesn’t this look exactly the way that genAI answers the question? Does Mr Brucker have an opinion one way or the other?

Both perspectives are flawed, the one because it denies change, the other because it posits a revolutionary break with the past … The degree and direction of change … are the … particular focus of this chapter.

Brucker then spends the next 40 pages giving examples and counterexamples of the two views. Brucker’s book was published well before the launch of genAI. His writing is a typical example of what undergraduate historians are trained to do: to demonstrate they have read the secondary sources, that they are aware of and state the major arguments of the best-regarded academics, and that’s probably, in most cases, sufficient. This is exactly what genAI mimics:

The AI matches the academic style perfectly. Pick a disputed subject: Some say A, some say B. So yes, AI has reached human levels. It may not give us the answers we are looking for, but neither does a lot of academic discourse. Pity the poor academics who want to distinguish AI results from human responses in exams.

Of course, you could argue that AI has no knowledge of the real world, since what it knows has been gleaned entirely from a corpus. But for most subjects, the academic is in the same boat. Their knowledge is gained from other books. But commentators are already accepting knowledge from a corpus as equivalent to knowledge of the real world. Here is Christopher Summerfield, Professor of Cognitive Neuroscience at the University of Oxford, no less:

We should subject LLMs to the so-called Duck Test: if something swims like a duck and quacks like a duck, then we should assume that it probably is a duck [These Strange New Minds: How AI learned to talk and what it means]

You might argue that there is more to academic study than reading other books. A lucky few (very few) get to examine the primary sources. As it happens, Gene Brucker spent many years combing the vast archives of Florence, so if anyone knows, he does. He edited and published some of the most exquisite autobiographies of Florentine merchants, for example. He gives wonderfully vivid examples both for and against every argument. But his conclusions about Renaissance Florence seem to be very much in line with the genAI conclusions. Is he afraid to state his convictions, or has AI caught up with the majority of scholars? Perhaps the mark of distinction of a scholar is to say “I don’t care what the consensus is, my conclusions are different!” – which is why we still read Burckhardt on the Renaissance, and probably will do when Brucker has been forgotten. He might be wrong, but he’s so entertaining to read. For the most part, however, genAI could take over many of the academic roles in the humanities in universities, and I don’t imagine many people would notice.