The silicon-based fundraiser

Deduplicating databases.  Proofing letters.  Selecting the donors who will receive a communication.  Data entry.  Everyone has a fundraising task (or tasks) they wish had been automated by now.  We were promised Rosie the robot maid and hoverboards and things that do the things we don’t want to do.  Or, depending on your preferred flavor of sci-fi, we were promised that robots would take our jobs or our lives.

And while we did get things that fit in your pocket that contain most of human knowledge, they also come with Facebook, so that’s pretty much a wash.

Fear not!  Or fear!  That future where you can focus on the more human fundraising tasks is closer than you might think…

It’s game time.  You get to have written correspondence with two alleged people.  Your job: determine which actually is a human and which is a computer.  If the computer can fool human interrogators like you, it has passed what its inventor Alan Turing called the imitation game (but we now know as the Turing test). 

Turing himself thought computers would be able to pass this test by the year 2000 (perhaps we would have gotten closer if he himself hadn’t been persecuted and prosecuted for his homosexuality, driving him to suicide).  We aren’t there yet, as anyone who has interacted with a phone tree that asks you to pay attention because their options recently changed can tell you.  When programs have been able to pass the Turing tests, it has been in very specific domains rather than as a generalized intelligence.

The innovation is called GPT-3.  While it lacks the punchiness of a HAL or Skynet, it could be similarly game changing. 

Researchers in the AI space have been going through what they term “GPT-3 shock.”  The shock isn’t that GPT-3 isn’t that much different from what’s come before — it used training materials to learn how to produce human-like texts.  The difference is in scope—10 times the parameters of its nearest predecessor—and its results.

The results are such that GPT-3’s researchers put out warnings with their model—that it could be used to create fraudulent academic essays, spam, phishing, social engineering, and more.  So far, people have used it

– to turn language into code (so if you ever wanted the HTML for “a button that looks like a watermelon,” look no further)

– to chat with famous people (what would Ben Franklin have thought of social media?)

– to generate memes (because Internet)

– to create news stories that people can’t distinguish from real news stories (possibly because when a gender reveal party burns down California, there is no news that is too weird)

In fact, The Guardian has a GPT-3-written, human-edited piece here that looks, well, like a Guardian piece.

My personal favorite is summarizing content for various grade levels.  So, for example, its summary of the Wikipedia entry for quantum computing for a second grader is:

“Some people think that if you had one hundred people working on one problem for one hundred years you could not figure out how to solve it with a regular computer but if you had one hundred people working on one problem for one hundred years using a special kind of computer called a Quantum Computer you could figure out how to solve it.”

Whereas for an eighth grader, it is:

“When we say quantum computing, we mean computing based on quantum physics. With classical computing we perform calculations one bit at a time. With quantum computing we try to take advantage of a superposition between 1 and 0. This is what we mean by qubits. This allows us to make connections between pairs of bits (qubits) using quantum logic gates. Quantum computing allows us to perform computations faster than classical computing.”

And the big part is it wasn’t trained to do any of these things.  It’s largely generalizable.

In other words, while it may not pass a Turing test perfectly, it’s getting there.  It also means that if trained on your fundraising letters and results, it may be able to make a compelling case for your mission.

Before that, though, the things that computers are supposed to be good at will come into to play — selecting donors, modeling responses, learning from our past to predict our future.  If you’d like to learn more about how we plan to do this, please join us September 29th for our free Death to Buckets! Toward more advanced and profitable segmentation webinar.

Sign up for Moore updates