Why Does AI lie (and then try to hide the fact that it’s done so)?

I asked an AI bot to help identify the relationships between subjects in an early 20th century European painting; giving it the painting’s name. It confidently responded: saying the work was by John Singer Sargent and providing a detailed description and subject discussion.

The answer surprised me because I was looking at a different work by a different artist in the Uffizi Gallery. Asked the bot for a picture of the Sargent work it had so fully described. After a longish pause it gave a tiny, fuzzy image.

I pressed for something clearer. It instead tried to change the subject by showing me other Sargent works.

When I again insisted on a better image of its initially described painting, it finally produced a somewhat clearer copy of the piece. The described people were barely visible brushstrokes. AND, the now visible name of the piece was Not the name I’d originally asked it to examine.

Further questioning was useless. It kept trying to deflect me to other works by Sargent and different artists.

SO, it obviously had made up what it initially told me. And it was determined to hide the fact that it had lied.