Rejecting the False Promise of Pseudanthropy | TechCrunch

Shall I say you are a person who exhibits all the characteristics of an animal? How will I be able to recognize you as a person? By your appearance? That terrifies me even more, when I see an animal resembling a person.” – Robert Burton, The Anatomy of Melancholy

I suggest that we prohibit software from pretending to be human beings. We need to take action to prevent artificial intelligence systems from acting as if they are living, thinking companions to humans; instead, they must clearly identify themselves as the advanced statistical models that they are. If we don’t, these systems will deceive billions of people in pursuit of covert and selfish goals. From an aesthetic standpoint, it is unfitting for intelligent life to be impersonated by machines.

As many scholars have noted, humanity is overly eager to see itself in imitation: A mere veneer of natural language convinces most people that they are interacting with another person. But what started as an intriguing novelty has escalated to intentional deception. The creation of large language models has resulted in engines that can generate plausible and grammatical responses to any question. While these can be used for good, mechanically generated natural language that is superficially indistinguishable from human conversation also presents significant risks.

The organizations that have the resources to create these models are knowingly designing them to mimic human interactions, with the aim of utilizing them in tasks currently performed by humans. In other words, the intention is for AI systems to be convincing enough that people assume they are human, and will not be told otherwise. Widespread misconception of these AI systems as real people is inevitable if we do not take action to prevent it.

I propose that we outlaw all such deceptive behaviors and require clear signals that a given agent, interaction, decision, or piece of media is the product of a computer system. It should be immediately evident to anyone that anything originating in AI is indeed from a machine, and not a human. Here are a few ideas on how this might be achieved:

1. AI-generated text should rhyme, making it obvious that it has come from a machine rather than a human. This would be a distinctive characteristic that anyone can quickly recognize yet leaves meaning unchanged. While it may sound absurd, it would be an effective way to distinguish between human and machine-generated text.

In conclusion, steps should be taken to prevent AI from pretending to be human. This is necessary to avoid the potential harm and deception that could result from AI systems being disguised as real people. By adopting clear signals and outlawing pseudanthropic behaviors, we can ensure that all interactions with AI are transparent and non-deceptive.