The Legality of Perplexity

Legal and ethical considerations on the controversy regarding Perplexity, AI search engines vs Section 230 protection, the web crawler problem, and the hallucination problem.

Perplexity 

At the beginning of this year, Perplexity was hailed as a potential Google disruptor. On the back of an investment of $73.6 million in Series B funding from Jeff Bezos’ family fund, Nvidia, Databricks,  among 56 investors, the stage was set for Perplexity to disrupt web search and make its mark on the information ecosystem in 2024.

CEO Aravind Srinivas explained to Forbes, that Perplexity was "almost like ChatGPT and Wikipedia had a kid". It's a free search engine that combines a stand-alone large language model (LLM) with internet access and integration to other LLMs from Anthropic, OpenAI, and Meta. It delivers fast AI-generated answers to search queries and always cites the original web pages in its answers. 

Srinivas had built solid research experience as an intern at first OpenAI, then DeepMind, and finally Google, before pivoting back to OpenAI once more as a research scientist. Straight hereafter, Srinivas founded Perplexity with Johnny Ho, Denis Yarats, and Andrew Konwinski in August 2022. 

When asked in interviews, Srinivas insists that Perplexity will be a positive for writers, creators, and freelancers who rely on search traffic to put food on the table. He claims that AI-infused web search will lead to higher quality visitors and marketing leads to the web pages cited by Perplexity, and thus it will be a net benefit to the industry. 

Many people, however, myself included, are skeptical of such claims - especially when checking the general temperature among tech wizards and executives at major AI companies when it comes to their value-judgment of original work as a concept. To this small but powerful group of people, any data on the internet is regarded as just content meaning fresh meat for insatiable AI models. 

For example, the always honest and outspoken OpenAI CTO, Mira Murati, said in a recent talk at Dartmouth's School of Engineering:

“Some creative jobs maybe will go away, but maybe they shouldn’t have been there in the first place — if the content that comes out of it is not high quality.”

In a similar fashion, Yann LeCun, Chief AI scientist Meta AI, posted on Twitter on the first day of 2024:

“Only a small number of book authors make significant money from book sales. This seems to suggest that most books should be freely available for download. The lost revenue for authors would be small, and the benefits to society large by comparison.”

Statements like these are worth having in the back of mind as we look further into the business model of AI companies, here with Perplexity as a prime example. 

From Platform to Publisher

The business model of generative AI companies is to crawl through vast amounts of original content on the internet and feed it back to users in a revised form. 

Google search is doing something similar but with one important caveat. While traditional search engines are presenting original content "as is", AI search engines like Perplexity are mashing various sources together to present something "new". Technically, this – I think - makes Perplexity a publisher of information rather than a more neutral indexing platform such as Google.

Section 230 (c)(1), Title 47 of the United States Code that was enacted as part of the Communications Decency Act of 1996 is also known as the “26 words that made the internet.” It states: 

“No provider or user of an interactive computer service shall be treated as the publisher or speaker of any information provided by another information content provider. “

American social media platforms have shielded themselves from liability for user-generated content because of Section 230. If Facebook was exposed to the threat of a lawsuit each time a user posted offensive/harmful/violent/disturbing content, it could never have scaled into the billion-user network it has been for years. 

Last year in May, the boundaries of Section 230 were challenged in two Supreme Court cases - Gonzalez v. Google LLC  and Twitter v. Taamneh. 

In the former case, plaintiff was the family of Nohemi Gonzalez, a 23-year-old design student who became victim of a terrorist attack in Paris in November 2015. The family of Gonzales claimed that YouTube's recommendation algorithm had played a crucial part in radicalizing the gunmen behind the attacks by exposing them and continuously nudging them toward ISIS propaganda and recruitment material. 

In the latter case, the family of Nawras Alassaf, who was killed in a terrorist attack in an Istanbul nightclub called Reina Club on New Year's Eve, 2017, sued Twitter, Google, and Facebook for failing to police against ISIS accounts, posts, and videos leading up to the attack, and for recommending terrorist-related content to users. 

Both cases were ruled in favor of the BigTech companies.  Justice Clarence Thomas wrote in an opinion in Twitter v. Taamneh:

“Given the lack of any concrete nexus between defendants’ services and the Reina attack, plaintiffs’ claims would necessarily hold defendants liable as having aided and abetted each and every ISIS terrorist act committed anywhere in the world.” 

The Supreme Court did explicitly not comment on the scope of Section 230’s liability shield in either case. However, we can likely infer from the decisions that – in general – social media companies cannot be held liable for content promoted by their recommendation algorithms. But what about generative AI?

Can we say that AI search engines should be afforded the same Section 230 protection that allowed social media companies to grow abnormally in the past decade-and-a half? It’s doubtful.

We will circle back to the Section 230-question later. First, let’s look at the recent controversy surrounding Perplexity and identify two critical legal issues.