AI bias might not be a threat and here’s why

OpenAI, developer of the ChatGPT language model, is the best-funded and largest AI platform company with over $10 billion in funding at a valuation of nearly $30 billion. Microsoft uses OpenAI, but Google, Meta, Apple and Amazon have their own AI platforms and there are hundreds of other AI startups in Silicon Valley. Will industry forces drive one of these to become a monopoly?  

When Google started its search business, there were already a dozen existing search platforms such as Yahoo, AltaVista, Excite and InfoSeek. Many observers asked if we need Google as yet another search engine? Yet Google became an online monopoly due to its superior performance and networking effects.  

Networking effects are powerful and Amazon is a perfect example with nearly 60% share of the online commerce market. Buyers go there to find the largest selection of product suppliers, and sellers go there to find the largest audience of customers. Even if another site performed better than Amazon, it remains nearly impossible to overcome these networking effects.  

PENNSYLVANIA STATE GOVERNMENT OPERATIONS TO START USING ARTIFICIAL INTELLIGENCE

Google introduced a new page-ranking system that users immediately recognized as superior to its predecessors. Users flocked to Google with its rapidly expanding catalog of online sites and more accurate search results; and advertisers gravitated to Google to reach this massive audience of users.  

Bias in AI or artificial intelligence might not end up being an issue because it's not likely any one AI will dominate the market. (iStock)

Earlier search platforms quickly withered and died. Facebook became a monopoly with the same networking effects as users left existing platforms like Friendster and MySpace to join Facebook, and it became the biggest platform where users could more easily find their online friends.  

AI platforms don’t appear to exhibit the same networking effects as the search and social media platforms. AI platforms are more similar to online publishers like the New York Times or Fox News online sites – where they gather existing available information and create new content based on this information and intelligence.  

There are no networking effects driving the New York Times, Fox News and other news sites to coalesce around a dominant news publisher monopoly. It appears that the multitude of AI platforms will operate as publishers and are not likely to become eventually dominated by one monopoly company.  

If the AI companies operate as publishers rather than as "computer services providers" as defined by the Section 230 law, they do not gain the benefit of Section 230 legal liability protections.  

Rather than simply sharing third-party content, these platforms receive and analyze such content in order to publish new content and operate in a manner similar to other publishers such as the Times and Fox News.  

The information used as inputs for AI platforms is called training data which is scraped from all sorts of online news sites and social media platforms like Twitter. If an AI platform limits its sources of training data to mainly right-wing news sites then the resulting AI content will reflect this right-wing bias. Similarly, if an AI platform relies primarily on liberal news media then the resulting content will reflect a left-wing bias.  

CLICK HERE FOR MORE FOX NEWS OPINION

AI platforms only introduce issues of censorship if they hide the biases of their training data, or if they introduce blunt restrictions on their AI content outputs – for example by allowing certain types of content for leaders of one political party and disallowing similar content for the other political party.  

Some science fiction movies like "The Matrix" and "Bladerunner" illustrate how AI content could be misused to reflect only the current consensus or government narrative – and this is why it is imperative for AI platforms to be required to show transparency in publishing the specific sources/sites of their training data.  

Earlier search platforms quickly withered and died. Facebook became a monopoly with the same networking effects as users left existing platforms like Friendster and MySpace to join Facebook, and it became the biggest platform where users could more easily find their online friends.  

There will likely be many AI platforms and if their training data sources/sites and biases are required to be transparently published, then users can select which platforms they prefer.   

Since there are dozens of AI platforms with little likelihood that one becomes a monopoly, more attention should focus on the existing monopoly platforms for search, online video and social media platforms that control far more of the content and act as gatekeepers for public and private discussions of new and sometimes controversial topics.  

The four guardrails of online content moderation – safety, neutrality, transparency and accountability apply to both AI platforms and the online search and social media platforms. However, the issue of neutrality is more urgently focused on the existing monopoly platforms that are currently able to operate freely to censor content based on viewpoint or the current consensus opinion of their favored political organizations.  

Michael Matthys is co-founder of the Institute for a Better Internet.