Constructed using the free availlable software RunWayML an open source AI solution which allows the use of a slew of different AI trained sets and provides means to use them on your own data
Using its text to image GAN (Generative Adversarial Neural Network) and selecting several sentences made of the abstract concepts: Television and Radio combined with some adjectives and nouns the following images were produced
The images themselves evoke a kind of eerie painterly atmosphere, which probably is caused by the un-understanding of the AI of the particular meaning of these concepts.
The dataset used appears to have the visual cognition of a 5 to 6 year old, it is interesting to see how this translates to the resulting outcome
Also interesting to note is that adding one or several spaces will change the output significantly. It is as if "nothing" is very meaningful for our AI!
Input sentence: "Tv statiuc with radio phone ", AA 2020 (Input exactly as is, including misspelling and spaces)
Television and radio I, Input sentence: "Television and radio", (with diffent arrangements of spaces in between and at the end), AA2020
Input sentence: "Television and radiophone in concrete wall with nail", AA2020
Television and radio II (with diffent arrangements of spaces in between and at the end), AA2020
Television and radio III (with diffent arrangements of spaces in between and at the end), AA2020