#53: AI's Struggle with Facts: The Case of the Strawberries and the Moon Cats

Cats on the Moon?

Since Google revamped its search page to include AI-generated summaries at the top of search results, there have been several instances of erroneous and misleading information. For example, an AI-generated response falsely claimed that astronauts have encountered cats on the moon. This feature, which is intended to provide quick answers, has instead spread misinformation and raised concerns among experts.

One notable error was when Google’s AI stated that the U.S. had a Muslim president, referring to a debunked conspiracy theory about Barack Obama. This misinformation, backed by a misinterpreted academic citation, underscores the AI's inability to correctly validate sources. Melanie Mitchell, an AI researcher, criticized the AI's unreliability and called for the feature to be taken offline.

Google responded by stating they are making "swift action" to correct errors and that the majority of AI-generated summaries provide accurate information. However, the randomness and potential for errors inherent in AI language models remain problematic, especially in emergency situations where incorrect information can have serious consequences.

Experts also express concerns about the AI’s potential to perpetuate biases and misinformation, thus impacting users' perceptions and confirming their biases. Additionally, the new feature threatens the traditional search model, which directs traffic to various websites and online forums that rely on Google for visibility and revenue.

Overall, while Google aims to enhance its search capabilities with AI, the rollout has been criticized for its haste and the resulting inaccuracies.

Lost in the Strawberry Fields

We had first visited Hallucinations in a very early edition of the newsletter and had seen how early versions of LLMs had pronounced me dead!

And looks like even as they have got much better they still have a propensity to get basic things wrong like:

How many R's in Strawberry?

GPT-4o

GPT-4

GPT-3.5

Gemini

Well, rather than being just funny, it becomes a bit scary too when we realize that we are getting ready to hand the keys of the kingdom to a technology which can't even make up its mind about how many R's in Strawberry. The fact that it tries to make up logic to justify a wrong answer or to rule out a perfectly right answer make it even more scarier.

Just a reminder that building the right guardrails and checks & balances for this technology is Vital (not just Essential or Desirable)

Google's AI Search Leaves Publishers Scrambling

Coming back to Google's recent overhaul of its search engine. As mentioned, it now includes AI-generated summaries,which has left publishers worried about the impact on their business models. These AI-generated paragraphs, appearing at the top of search results, often reduce traffic to the original news sites, causing concern among media executives.

Frank Pine, executive editor of Media News Group and Tribune Publishing, noted that the AI Overviews could

"choke off the original creators of the content."

Media executives are in a tough spot: they need Google's traffic but are frustrated by how their content is used in AI summaries. Blocking Google's web crawler could reduce their visibility, while refusing to be indexed could be catastrophic for their business.

Google claims its AI summaries improve traffic to websites and has announced plans to refine the system after some high-profile errors. However, media executives argue that these AI Overviews are harmful, particularly to smaller publishers and businesses, and have urged for government intervention.

The conflict is part of a larger tension between tech companies and publishers over the use of news content for AI training. Some publishers have sued companies like OpenAI and Microsoft for copyright infringement, while others, like The Atlantic, have negotiated deals for content access.

As publishers adapt to these changes, many are focusing on building direct relationships with readers through subscriptions, newsletters, and other engagement strategies. Despite these efforts, the industry's reliance on search engines remains a critical issue, with calls for legislative action to protect content creators and ensure fair compensation.

Signing Off

Why did the AI struggle with understanding a strawberry?

Because it couldn't find any "byte"-sized pieces of information!

Keep an eye on our upcoming editions for in-depth discussions on specific AI trends, expert insights, and answers to your most pressing AI questions!

Stay connected for more updates and insights in the dynamic world of AI.

For any feedback or topics you'd like us to cover, feel free to contact me via LinkedIn.

DEEPakAI: AI Demystifed Demystifying AI, one newsletter at a time!

p.s. - The newsletter includes smart prompt based LLM generated content. The views and opinions expressed in the newsletter are my personal views and opinions.