By Michael Scott
Anthropic creator Claude AI has recently confirmed partnering with the Collective Intelligence Project to deliver a draft template for future artificial intelligence conduct. The firm explained the initiative aimed at democratizing AI tool development through input from the non-profit Collective Intelligence Project (CIP).
Anthropic Joins Collective Intelligence Project to Deliver AI Constitution
The project involved exploring different AI development courses by surveying 1,000 Americans in drafting the AI constitution. Collective Intelligence Project’s Divya Siddarth illustrated that artificial intelligence has an inherent transformative impact.
Siddarth added that a small percentage of the population decides on its form. The draft constitution portrays efforts to build processes founded on truthhood.
The Collective Intelligence Project started operations in January 2023 as a brainchild of Siddarth. Besides Siddarth’s political economist and social technologist experience, CIP was co-founded with Saffron Huang.
The CIP establishment targeted delivering collective intelligence and progress. The CIP would, in March, partner with Anthropic to deliver the collective constitution for AI initiatives. The partners seek to develop models integrating external inputs and explore the democratic process’s role in AI development.
Siddarth restated the need to assemble input from proven practices borrowed from the democracy space. The primary principle is replicating what constitutes a good decision for the parties and ascertaining the participants involved. He added that conducting research regarding AI’s credentials as a transformative technology becomes inevitable.
Anthropic regarded the project as involving 1,000 individuals to help it pick rules for the AI chatbot. The sample size aimed to have a representative composition of age, location, income, and gender. The participants were tasked to choose from the listed rules and formulate their own guidelines.
Anthropic confirmed receiving 1,117 statements from the participants. The firm secured 38,000 votes regarding questions on sexual preference and race-oriented discrimination. Anthropic admitted that though the participants expressed diverse opinions, there was consensus on most statements.
Expand Sample Sizes and Composition Beyond Tech-Savvy Respondents
Although the survey targeted collecting information from diverse sample sizes, Siddarth opines that subsequent experiments should seek beyond affluent and tech-savvy respondents.
Siddarth lauds the input by US AI consumers to dictate the AI constitution. The executive urges for additional information outside the US. The statement indicated that the involvement of multiple parties in developing AI would necessitate overlapping values to accommodate diversity.
Siddarth hailed the Collective Intelligence Project for featuring input from rival AI firm OpenAI alongside other democratic developers. Nonetheless, the AI Constitution project relied upon Antrhropic’s Claude AI models.
Siddarth lauded the comparison since the model developed by Anthropic already had previous training on constitutional provisions. However, Siddarth indicated that Anthropic researchers drafted and compared the same model. The comparison extended against one written by the public.
Siddarth illustrated a surprising discovery in the statements that subsequent AI models should avoid discriminating on disability grounds. However, such were overlooked when formulating the the Anthropic constitution. The executive acknowledged that the inclusion was not accidental but rather an element significantly emphasized.
Anthropic Journey Attracts Google and Amazon Investment
Anthropic began in 2021 among developers oriented to unveil advanced artificial intelligence. Its establishment coincided with Google, Meta, and Amazon replicating Microsoft’s move to allocate billions towards projects involving generative AI development.
This year turned lucrative for artificial intelligence developers. In particular, Anthropic would benefit from a $400 million investment from Google in February. Amazon would in September commit $4 billion in September.
Anthropic, earlier in 2023, teamed up with Google, Microsoft, and Google in a pledge towards responsible AI development shortly after meeting representatives from the Biden administration. Other tech companies, including Amazon, Inflection, and Meta, would join the pledge. The pledge list would become longer in September as NVIDIA, Adobe, IBM, Palantir, and Scale AI joined. The new entrants included Stability AI, Cohere, and Salesforce.
Siddarth expressed optimism in the Collective Intelligence Project delivering the much-needed democracy in AI development. Despite the limitations of the Constitutional AI experiment, the initiative is set to attract more public input toward tech development.
Siddarth emphasized the need to undertake more processes and learn from the existing accomplishments. As such, civil society groups, policymakers, and various companies should offer input towards bettering the Constitutional AI initiative.