We the Chatbots …
Above, three “ Founding Mothers” gather in the year 2087 to discuss revisions to the dated 2030 version of the AI Constitution. Generated with Canva’s Magic Media in Magic Studio.
In America, the US Constitution is regarded as the foundation for the law of the land. As such, federal law requires public school children to learn about the US Constitution each year on September 17, Constitution Day. So, when I read about Anthropic AI’s integration of something called constitutional AI, a technique which amplifies ethics in their large language model (LLM), Claude, I immediately began to wonder what an AI Constitution might look like in a century from now.
As I reminisced about my own early education and learning the US Constitution, I instinctively hummed “the Preamble song,” a tune made popular by the 1973 American educational cartoon, Schoolhouse Rock. “We the People of the United States, in order to form a more perfect union…” Quickly realizing that I forgot the rest of the lyrics, and the following lines of the preamble, I returned to the top of the song, slipping out the following, “We the Chatbots… of the Internet, in order to form a more perfect union…” Then it hit me. Not only is the US Constitution dated, it simply does not suit today’s digital world of competing large language models (LLMs).
When I asked Google’s Gemini what the US Constitution does, it said “The United States Constitution is the supreme law of the land, outlining the framework of the federal government and safeguarding citizens' rights.” The words, “law of the land” stuck with me. Why did Gemini choose these words? Did it mean “law of the land” literally? Does the Constitution prioritize “the land” over the people? And if so, does the Constitution even count since the land the US Constitution is supposed to protect was stolen from the start? How does this translate over to a world with AI? Will “law of the land” even matter as the emergence of the LLMs invites exploration into establishing the “law of the digital.” I was bursting with queries when I realized these must have been similar questions and concerns that spurred the existence of public benefit corporation, Anthropic AI and their safety-forward LLM, “Claude.”
In the Atlantic’s 2023 magazine, Dialogues: On AI, society, and what comes next, Anthropic AI President and Co-Founder, Daniela Amodei discusses safety and responsibility around LLMs for the public and society overall. In order to ensure protection and tackle malefactors, Claude has been trained with a technique known as “constitutional AI.” Using a written constitution instead of reinforcement learning from human feedback (RLHF), Claude is trained by a set of values to be “helpful, harmless, and honest.” Claude’s responses are generated based on an ever-changing constitution that comes from a variety of documents including the UN Universal Declaration of Human Rights and Apple’s terms of service.
Any history buff might suggest Anthropic’s constitutional Claude might be setting the stage for the 21st Century’s Constitutional Convention as Daniela Amodei thinks beyond centralized power, and instead unification of LLMs (or powers) for the greater good of humanity. “External engagement on these issues is central to our work. We think developing AI safely is a much broader project than Anthropic can—or should—tackle alone,” said the co-founder in the Dialogues interview. Amodei’s stance is clear, the development of constitutional AI values should not be singular or held solely in the power of one corporate entity. Instead, constitutional AI should be shaped by a variety of contributors, and as noted by Anthropic’s Co-Founder and Amodei’s brother, Dario Amodei, Thus, an early decentralization of corporate power in this space may help avoid competition driven “corporate propaganda,” a phrase used by Anthropic’s Co-Founder and Amodei’s brother, Dario Amodei in the inaugural Time100 AI list released in 2023 when dubiously addressing utopian hypotheticals for AI.
Unlike the US Constitutional Convention of 1787, one hopes a more diverse group of founding people might be invited to contribute to such powerful and integral documents that will undoubtedly shape the future of Artificial Intelligence in the coming years. In the 1700s Founding Fathers were concerned with the King of England’s power, resulting in events like the American Revolution, drafting the Declaration of Independence and writing the US Constitution. In today’s public hysteria and growing fear of unknown technology, we can only hope “Founding Mothers” like Daniela Amodei are able to steer an AI driven society towards a peaceful, innovative, and just future.
References:
Dario and Daniela Amodei: The 100 Most Influential People in AI 2023, Time Magazine, written by Will Henshall and Billy Perrigo, 7 Sept. 2023
Dialogues: On AI, society, and what comes next, Google + Atlantic Re:think, 2023
The Constitution,The White House. Accessed March 2024
Generative AI Tools:
Gemini, 2024
Canva, Magic Studio: Magic Media