TGITM collaborated with Google as well as artists Es Devlin and Ross Goodwin on “Poem Portraits,” an “experimental, collective artwork, woven at the intersection of AI and human creativity.”

The custom algorithms developed by TGITM were trained on over 20 million words of 19th century poetry. Users are prompted to input a word, and the algorithm consequently generates an original two line poem. The poem is then projected onto either an image of the user’s face, or in the cases of the multiple physical installations of the work, onto major sculptural artworks or landmarks.

It operates much like predictive text: instead of replicating or modifying existing phrases, it uses its training data to develop a detailed statistical model. This allows the algorithm to create original phrases that reflect the style of the material it was trained on.

The poems it produces can be remarkably human and touching, and sometimes amusingly silly.

On the cutting edge of human and machine-learning collaboration, as well as image transposition and selfie “filters,” this artwork leveraged state-of-the-art AI capabilities to turn users into custom “poem portraits” in installations with global acclaim.