US researchers used AI for phages; computation is not wisdom
By Caroline Ashford · Reporting from Richmond, Virginia ·
The latest breathless reports from Silicon Valley’s biological offshoots suggest we have reached a new zenith in the mastery of creation.
The Language of Life, Predicted by Code
The latest breathless reports from Silicon Valley’s biological offshoots suggest we have reached a new zenith in the mastery of creation. Apparently, artificial intelligence has moved beyond mere prediction—it is now generating entire functional genomes. As reported by the BBC, US researchers utilized generative AI to design 16 novel bacteriophages that can replicate and infect E. coli. The technology works similarly to large language models like ChatGPT, but instead of words, it predicts the "language of life." This move, which astrobiology.com details using frontier models like Evo 1 and Evo 2 trained on trillions of letters of DNA, is heralded by some as a “very significant turning point.” We are told that for the first time in history, we are beginning to design biology on a computer.
When Computation Replaces Observation
The sheer scope of this achievement—synthesizing 302 promising AI designs and finding 16 viable phages—is undeniable. Prof Marc Güell noted that while previous work allowed us to read or rewrite DNA, AI allows us to generate new synthetic content. Yet, I find myself troubled by the underlying assumption: that this computational ability equates to wisdom. The reports from sciencemediacentre.org are replete with technical milestones—the low sequence identity of generated proteins, the high pLDDT scores predicted by ESMFold—but they gloss over what is at stake. We are being asked to trust a statistical pattern recognition engine with the fundamental building blocks of life.
The Switch Does Not Replace the Circuit Board
To understand this leap, one must recall the transistor. It was an invention that allowed us to amplify and switch electrical signals, enabling all modern electronics. Its significance lay not just in its existence, but in how it functioned: a small input controlling a massive output through precise physical principles. Generative AI functions similarly; it is a powerful switch—taking a prompt (a voltage) and controlling the flow of information (the current). But this analogy breaks down when we consider the nature of what is being switched. A transistor operates within established, predictable laws of physics. Life, even at the level of a bacteriophage genome, is not merely an algorithm; it is the accumulated history of billions of years of struggle and adaptation.
The promise that AI can simply "design" away the messy work of evolution—that we can generate viable life by feeding enough data into a model—is profound hubris. We are so eager to build our next great machine that we forget what makes a truly robust system: the slow, imperfect process of local resilience and human understanding. The fact that half of the functional phages found in one study had acquired mutations suggests that even these advanced models require natural evolution to "polish" their designs.
We must resist the temptation to view this as an end unto itself. We are not merely building a faster calculator; we are touching the deep, mysterious architecture of existence. The true genius lies not in the capacity to generate novel sequences, but in the human capacity—the understanding, the ethical framework, and the rigorous caution—to know when and why such power should be wielded.