Is AI lying to you?

A hand is holding a wooden block with the word fake on it.

Is AI lying to you? It may be. When asked about airplane crashes google Bard returned with this:

There have been many airplane crashes throughout history. Here are some of the most notable ones:

  • On February 3, 1953, a British Overseas Airways Corporation (BOAC) de Havilland Comet 1 jet airliner broke up in mid-air over the Mediterranean Sea, killing all 35 people on board. This was the first fatal accident involving a commercial jet airliner.
  • On April 15, 1960, a Cuban Douglas DC-8 crashed near Havana, Cuba, killing all 73 people on board. This was the deadliest aviation accident in Cuban history.
  • On March 27, 1977, two Boeing 747s collided on a runway in Tenerife, Spain, killing 583 people. This was the deadliest aviation accident in history.
  • On January 28, 1986, the Space Shuttle Challenger exploded shortly after launch, killing all seven crew members on board. This was the first fatal accident involving a U.S. space shuttle.
  • On July 17, 1996, a TWA Boeing 747 exploded over Long Island, New York, killing all 230 people on board. This was the deadliest aviation accident in U.S. history.

Most of this list is correct according to a google search, except the 1960 Cuban Douglas DC-8 Crash.

Google is fact checking it’s AI

Hand flip wooden cube change the word from fake to fact

Now with the click of a button I am now able to fact check the information that I am receiving. Google Bard has the ability to fact-check itself using a variety of methods. One method is to compare its responses to Google Search results. When you tap the “G” icon in Bard, it will cross-reference its response against Google Search results to determine if there is web content that supports the provided answer. If the statement is confirmed, you can click on the highlighted phrases to access additional information through Google Search.

Another method that Bard uses to fact-check itself is to rely on its knowledge of the world. Bard is trained on a massive dataset of text and code, which gives it a broad understanding of the world. This allows Bard to identify and flag potential inaccuracies in its responses.

Bard also uses a variety of other methods to fact-check itself, such as:

  • Checking for consistency with its own prior responses
  • Checking for consistency with the context of the conversation
  • Checking for consistency with known facts

It is important to note that Bard is still under development, and its fact-checking ability is not perfect. However, Bard’s fact-checking ability is constantly being improved, and it is already one of the most accurate fact-checking AI models in the world.

Skip to content