{"id":197,"date":"2019-09-30T10:00:54","date_gmt":"2019-09-30T15:00:54","guid":{"rendered":"http:\/\/capertongillett.com\/blog\/?p=197"},"modified":"2019-09-30T10:00:54","modified_gmt":"2019-09-30T15:00:54","slug":"chase-got-an-ai-copywriter-how-screwed-are-we","status":"publish","type":"post","link":"https:\/\/capertongillett.com\/blog\/chase-got-an-ai-copywriter-how-screwed-are-we\/","title":{"rendered":"Chase got an AI copywriter. How screwed are we?"},"content":{"rendered":"\n
\"\"
I, for one, welcome our new robotic creative overlords. (Credit Matthew Hurst<\/a>)<\/em><\/figcaption><\/figure><\/div>\n\n\n\n

In July, JPMorgan Chase made an announcement that struck fear into the heart of many a creative: They\u2019ve signed a five-year deal with Persado<\/a>, a company \u201creinventing marketing creative by applying mathematical certainty to words.\u201d<\/p>\n\n\n\n

Do what, now?<\/p>\n\n\n\n

It does my job, but with artificial intelligence.<\/p>\n\n\n\n

How this all started<\/h2>\n\n\n\n

Chase started with a three-year pilot program testing the effectiveness of Persado\u2019s AI writing digital copy for their card and mortgage services. The resulting creative has garnered a higher percentage of clicks than comparable human creative \u2014 sometimes twice as much. Compare a human-written digital ad:<\/p>\n\n\n\n

Access cash from the equity in your home.<\/p><\/blockquote>\n\n\n\n

to the AI\u2019s version.<\/p>\n\n\n\n

It\u2019s true \u2014 you can unlock cash from the equity in your home. <\/p><\/blockquote>\n\n\n\n

The scary thing is? The computer\u2019s version is actually kind of better.<\/p>\n\n\n\n

Chase has been so satisfied with the results that it\u2019ll be applying Persado\u2019s AI across all its service lines. And Persado says it\u2019s already working with 250 brands across the country, including some pretty big names, to expand the use of AI in marketing creative.<\/p>\n\n\n\n

How machine learning works<\/h2>\n\n\n\n

So how would an AI copywriter work, anyway?<\/a><\/p>\n\n\n\n

A vastly oversimplified and probably not entirely accurate overview, because I am not by any means a software engineer: Artificial intelligence works via a computer algorithm that\u2019s fed by data and constantly refined by its own successes and failures. This could be through supervised learning \u2014 No, algorithm, bad<\/em> or Good job, algorithm! That\u2019s exactly right<\/em> \u2014 (which doesn\u2019t necessarily need human supervision) or through unsupervised learning, in which the algorithm finds hidden patterns and structures on its own. It\u2019s trial and error at four billion cycles a second, and it doesn\u2019t need your help, thank you very much, it\u2019s got this.<\/p>\n\n\n\n

(When Google asks you to prove you\u2019re not a robot by selecting all the images that have a storefront in them, you\u2019re actually helping to \u201ctrain\u201d an algorithm by showing it what a storefront looks like and doesn\u2019t look like, so it can better identify storefronts in the future. That\u2019s right \u2014 by proving you aren\u2019t a robot, you\u2019re actually helping to train a robot to be better at pretending to be human. This is the world we live in.)<\/p>\n\n\n\n

Essentially, it\u2019s the Infinite Monkey Theorem on NZT-48<\/a>. And in this case, the typewriter-wielding monkeys are after my job.<\/p>\n\n\n\n

Some more entertaining examples of machine learning include the fun antics AI researcher Janelle Shane<\/a> has been having with her neural network, teaching it to do things like create recipes and name colors<\/a>. (Highlights include Sane Green, Snowbonk, and a lovely Kelly green named Bylfgoam Glosd.) There\u2019s also this example of Google\u2019s DeepMind AI learning to walk, which I could watch on loop for hours.<\/p>\n\n\n\n

\n