Cyber-Peace Making: How Constitutional AI is Redefining the AI Race

By. Nina Rutherford

DOI. 10.57912/31081750

---

The Golem is a Jewish folklore artificial humanoid made of mud. The Golem is soulless and serves as a cautionary tale for free will, the dark side of technology/AI, and humanity’s desire to play God.

New developments in AI have sparked debates on the AI Race and shifted rhetoric from existential fears to economic dilemmas. An advancing product called Constitutional AI (CAI) is a beacon of hope for both. CAI is structured around ethical guidelines, fostering a user-centric  culture for AI. The new code offers remedies for the issues that have previously plagued AI, such as  international conflict and pernicious user experience. While relations between China and the United States are strained amid the AI Race, agreements between both countries must be reached to promote international prosperity.

The competitive nature of the AI Race has led to significant breakthroughs in Artificial Intelligence. One AI advancement that has set the United States apart is the concept of Constitutional AI (CAI). Constitutional AI is an AI model that follows a certain set of rules that address the legal, ethical, and societal implications of AI, ensuring that AI systems operate within ethical bounds. The principles often include human rights contracts, privacy protections, ethical codes, and the like. CAI is a mix of machine-learning and reinforcement learning from human feedback (RLHF). It is different from commonly used machine-learning systems because it trains AI to use the data supplied, compared to other systems that just supply the data. Another thing that sets CAI apart is that it will refer users to other sources of information if it does not possess the answer itself. These deliberate measures to create AI that is ethical, sound, and helpful are the distinguishing features of Constitutional AI and what make it the code of the future.

Enter: Claude, the first functional, though imperfect, Constitutional AI product. The constitution that guides Clause is based on the UN Universal Declaration of Human Rights, industry best practices (such as avoiding sensitive/private information), non-Western cultural values, and adaptability and transparency learning. Claude is special because it explains why certain requests are problematic, its decision making is open to its users, and it is constantly updating through Anthropic’s machine learning coding techniques. This means that in the preliminary code, programmers at Anthropic can view the information and subsequent series of decisions that lead Claude to make its final response. Anthropic can then go in at various points and teach Claude to make better decisions in the future that more closely align with the constitution it is guided by.

As a new tech, AI is not immune to setbacks. Ordinary users and hackers have found numerous ways to jailbreak ChatGPT and Claude’s codes. Additionally, Chinese-subsidized agents penetrated Claude’s code in the first cyberattack without substantial human intervention. The agents were able to infiltrate a small number of Claude user’s operations, but like all foreign affairs, AI policies must be developed over-time. While Claude was the victim of this cyberattack, the offense is not specific to CAI, nor does it reflect a weakness in the idea of Constitutional AI itself. The novelty of Constitutional AI models lies in their regenerative learning model. The self-correcting nature of CAI’s code makes it an evolving product and signifies its hopeful future. Events such as this highlight shortcomings in AI that are more advantageously identified in the provisional stage. 

There is no documented equivalent of Constitutional AI developed by any Chinese tech company. This critical difference is America’s advantage in negotiating AI contracts and in navigating the future of trade with China as a whole. Should AI seep into every business sector, along with security and privacy concerns, ethical concerns will enter as well. Ensuring that the products Americans use that incorporate Chinese AI are programmed with trustworthy, morally sound artificial intelligence should be the metric that guides the use of international products. Because of the unique programming Constitutional AI uses, it should become an industry standard and be the most important metric in contracts between the United States and international businesses. Constitutional AI is a step in the right direction in the creation of “moralware [over] software”.

The AI Race has heightened the already competitive relationship between the United States and China. While relations between China and the United States have been uncollaborative, AI is such a distinct technology that will flourish in cooperation. Namely, user-experience and environmental concerns are some basic components that partnership can improve. Constitutional AI derives its ethical constitution from the societal and cultural practices programmers feed it. If CAI is developed collaboratively, all consumers can have an equal benefit from usage because it possesses the cultural and ethical context to create appropriate output. It’s the difference between taking advice from someone who knows information and someone who can apply it. Given the 31% rise in international exports associated with AI, negotiations between countries must be communicative. Should we fail to do this, the current trade that is due to AI, as well as projected growth, are at risk.

Current Chinese AI Law is in line with numerous U.S. CAI principles. The PRC’s 2024 proposal outlines that “AI developers, providers, and users shall prevent and control the possible ethical risks of AI in accordance with law,” and “adhere to the principle of fairness and impartiality.” These statements closely mirror the ethical and privacy concerns that U.S. companies express in AI. Similar to Claude, the Chinese proposal states, “AI [...] shall adhere to the principle of transparency and explainability.” This aspect of explainability is synonymous with Claude’s ability to present its decision making, something that is vital in Constitutional AI. A section in the proposal that details the integration of AI into several aspects of life: “The state formulates AI development plans, [...] and promotes the deep integration and application of AI in economic and social fields.” This, coupled with their later statement: “The People’s government [...] shall establish industrial policies that conform to international trade norms” substantiates that China is willing to negotiate.

An interesting area where China’s proposal diverges from the U.S. standard is that it encourages “environmentally friendly and energy saving technologies for AI development.” Despite documented environmental concerns, remedies or preventative measures have seldom been addressed in the U.S. The proposal also states, “AI providers shall clarify the applicable audiences, occasions, and uses of their products.” This distinction is interesting because it refocuses the idea of AI as an enhancing rather than supplemental tool. This line encourages developers to consider user-experiences and give instructions on a technology that is often ambiguous and confusing. It has the potential to highlight the limits of AI use, re-centering human judgement in its application. U.S. Constitutional AI should adopt this idea because it places the responsibility for ethical use back on the user, creating a more informed and decisive user base. These are the kinds of strengths that cooperation, on both sides of the race, can bring. Joint development of CAI has only positive outcomes for China, the U.S., other nations, and AI itself.

An international peace organization posits that “even if countries fundamentally disagree on the specific content of a regulation, they can still learn from each other when it comes to the underlying structures and technical feasibility of different regulatory approaches.” Support such as this bodes well, placing AI as a technology to handle as a united front, centered on ethical and human values rather than a race to create the best or fastest technology. Additionally, cooperation on AI will reveal weaknesses or areas where the U.S. lags, particularly in environmental impact and user experience. Put succinctly: “The future of AI depends on cooperation, infrastructure, and uniform standards.” Acknowledging the similarities between Chinese AI law and the United States’, and cooperating on international standards, is the direction the AI Race should head in. 

Building support for an international AI contract is the most important next step in the tech world. Constitutional AI is the most comprehensive and ethical guide in the AI world and should become the universal standard. Recognizing the similarities in goals, most of which are already aligned with CAI constitutions, between the U.S. and China will be our most effective tool. AI has a bright present and future, the advancements it will bring to businesses, government agencies, the medical field, law, communications, and essentially every aspect of daily life will be tremendous. However, this advanced society AI will materialize with the caveat of international collaboration and camaraderie, something that must be built. If there were anything to shake up the status quo, it would be AI.