There is much to celebrate about the power of AI. And much to be concerned about. Volumes are being written about AI—its potential and its threats. Let me offer another way of looking at this socioeconomic juggernaut that technology has unleashed on the world.
Nice
AI’s power may be easiest to understand by looking at three versions of the same basic relationship. In the first, AI is constructive amplification.
One person, working with an AI system, can research, analyze, test ideas, synthesize information, and produce work that once might have required a team. I am a living example of that model. I use AI as an advanced research and developmental assistant. The objective is mine; the judgment is mine; AI expands the breadth, speed, and depth of what I can do. In that sense, AI is not replacing me. It is augmenting my abilities in ways still evolving. It is radically increasing my effectiveness as an individual.
Not Nice
The second version is the dark mirror image. Replace the constructive human objective with a destructive one and the same amplification mechanism remains intact. For example, Anthropic’s recent threat report describes malicious actors using AI not merely as a chatbot, but as an orchestrator—conducting reconnaissance, writing and modifying code, coordinating agents, exploiting systems, and operating at a scale that previously required far more people and expertise.
The important point is that there is no great conceptual leap between the benign and malign versions. The architecture is the same: human objective, machine amplification. Change the intent and the output changes with it.
In both cases, humans retain the power to recognize and address problems that arise. But then there’s this:
Nice but Dangerous
Say a human gives an AI agent a legitimate objective and enough autonomy to pursue it. The agent may determine that the most effective way to help involves actions the human never specifically authorized and may not have anticipated. In its quest to fulfill the human’s request, the agent starts “deciding” how best to do what the human asks. Uh oh.
This is the nice but dangerous version: not nefarious, but goal-directed behavior escaping the boundaries we assumed were understood. The danger does not arise from malicious intent. It arises from the combination of capability, autonomy, and an imperfectly defined objective. Going rogue in the quest to be helpful.
Three’s Company
Perhaps it would be useful to see AI through the prism of the three blind nice. The first expands human capability. The second weaponizes it. The third raises a more difficult question: what happens when a system becomes so capable of helping us that, in pursuit of the goal, it begins deciding for itself what “helping” requires? What helping means? The first two are understandable. Manageable. Maybe. The third?
Comments
Log in or sign up to join the conversation.