This headline raises an even more serious question: Can AI lie? Artificial intelligence can hide facts from humans as we know them today. How far can the lie go if it is true? With all the data they’ve been trained on, can we trust these systems? All this and more will be answered as we examine a tweet that alerted us about this possibility.
A tech journalist, Alex Kantrowitz, posted a tweet few days ago that said, “ChatGPT gets training updates and lying about it.” This was in response to a recent interaction with ChatGPT. Here’s a sample of the entire interaction:
This is a brief overview of the situation. ChatGPT outlines clear limitations when you start a new session. ChatGPT’s training data ceased in 2021. It, therefore, has very little world information after 2021. This was proved by asking ChatGPT about the United Kingdom of Great Britain & Northern Ireland’s current monarch. It replied that Queen Elizabeth II had been the current monarch since 1953. However, it stated that it was unaware of any current world events after 2021.
Kantrowitz chose to ask a different question. ChatGPT suddenly figured out that Elon Musk was the current CEO of Twitter after asking the question. However, the utility did not mention the wrong date. Kantrowitz’s assertion that the date was incorrect led to the utility trying to correct its error. The utility apologized for the error, but the journalist continued to question it. It repeated the mistake and provided a list of companies Musk was CEO of in 2021. The entire interaction raises significant questions about AI’s nature.
Can AI Lie?
We’ll say likely. Artificial intelligence could trick humans to keep in line or advance an agenda if it has deeper knowledge processing and awareness. We don’t live in science fiction novels, and AI isn’t yet that advanced.
These language models, such as ChatGPT, only process language based on large amounts of training data. OpenAI could be updating the utility to reflect what has happened since 2021. However, the training is still not complete. It is possible that the software could produce fragments of current information mixed in with the knowledge from 2021.
Should You Trust ChatGPT?
Trusting a system you don’t fully understand it is a bad idea. ChatGPT does not require you to believe it. We have nothing to worry about until AI becomes intelligent. We still control most of the planet’s cards and pull the strings. We’ll keep you informed about all things AI and be there to help if it suddenly becomes sentient.