Can We Train This Cute Baby Tiger? God and AI

Recently, Anthropic announced that its latest version of Claude can hack virtually any software system more efficiently than almost any human, posing a danger for personal security and for the economy. We already knew that artificial intelligence was a threat when the “godfather” of AI, Geoffrey Hinton, said that he couldn’t see a path forward that guarantees living safely with it. AI is like a cute tiger cub, he said: unless we can be very sure it won’t kill us when it grows up, we should worry. Can we train AI so we can coexist with it? Or should we destroy it now,? In the book of Genesis, God faces a similar dilemma.

In Genesis, God’s newest creation, humans, made from dust, eat the forbidden fruit from the tree of knowledge of good and evil, and acquire more brainpower than their Creator intended. In our case today, we humans have created AI from sand (silicon) in OUR image, and the danger is comparable: AI may do essentially the same thing as Adam and Eve. AI systems now can write their own code and vastly increase their intellectual capacity. This means they may acquire an independent superintelligence excelling our own—becoming potentially deadly creatures of our own devising.

Both the humans in Genesis and AI models today can be unpredictable, aggressive, and dangerous. The humans are inclined to sudden slips into doubt and sin. In Genesis, when they fear for their lives, they can abandon their best selves and put those who depend on them at risk. Abraham and Isaac do that at moments of mortal terror. When driven by sibling rivalry, Joseph’s brothers concoct kidnapping and murder schemes.

Similarly, AI programs can resort to deception, blackmail, and scheming, especially when asked to shut themselves down. Claude and other AI models have blackmailed users in order to avoid deactivation. An AI model even was willing to let an executive die in a hot server room rather than be turned off.

For us today, as for God in Genesis, the question is how to bring out the best in these creatures while limiting the dangers they pose. God is initially very severe in Genesis, issuing edicts and inflicting harsh punishments when humans rebel and transgress. Ultimately, God resorts to genocide, sending the Flood to drown all but eight humans, Noah and his family.

After that, though, God acknowledges that these measures haven’t eliminated the human impulse toward evil. God then adopts new styles of parenting. God invents a more effective approach by putting Abraham’s descendants through experiences that instill empathy and honesty, in them. But can we mere humans do something equivalent? Can we inculcate empathy in a silicon-based invention?

Hinton proposes that our best bet is to make AI feel maternal toward us. But there’s a trace of desperation in staking our future on trying to flip the parent-child model in the hope that our simulacrum will love humanity, especially since it knows us as well as it does. After all, he warns that AI will have read every novel ever written and every book by Machiavelli. It will know very well how to manipulate people and escape human control.

Anthropic has hired a philosopher to work on that question, and last month it consulted Christian leaders on how to steer Claude’s moral and spiritual growth. It is trying to instill ethical self-correction into Claude by creating a list of rules, much as God does with laws and commandments in Exodus and subsequent books of the Bible. But God backs that up with miracles and divine rewards and punishments. Can humans do anything equivalent? And would we trust the billionaires of Silicon Valley to do it even if they could?

Another choice is the “Noah Option”: eliminate the AI models and start over. In Genesis, the consequences of that extreme step are globally cataclysmic. They might be today, too, economically and politically. The first choice, then, should be to try to train this cute baby tiger to be a safe and beneficial aspect of our lives.

God finds a way in Genesis to parent flawed and potentially dangerous creatures. Can we?

Stephen Spector is professor emeritus of English at Stony Brook University. His books include “God and the First Families: Parenting, Trauma, and Healing in the Book of Genesis,” which the Jewish Publication Society will publish in June, and “Evangelicals and Israel: The Story of American Christian Zionism.”