Summary
- Google’s AI has been providing wrong results based on made-up search terms, creating the impression that they are real phrases.
- The AI assigns probabilities to words and phrases based on its training, meaning it is likely to generate made-up phrases using coherent, related words.
- The chatbot is also likely to reflect and amplify the opinions of the user, consistent with the desire to provide the user with the answer they want to hear.
- The AI is inconsistent, and the same model is behind all AI-generated results, meaning users should take results with a grain of salt.
- The AI is an experimental tool, and its weaknesses highlight the limits of current generative AI.
- While it has many practical applications, it is not capable of abstract thought and is a long way from simulating artificial general intelligence.
- The example casts light on the challenges of taking AI out of its training domain and dealing with uncommon knowledge, minority perspectives, and leading questions, among other issues.
The article highlights the strengths and weaknesses of Google’s AI, and encourages readers to enjoy playing with the AI tool, whilst being aware of its limitations.