Turing Test: Museum edition. Could a computer program write an exhibit label or an exhibit catalog that made visitors believe a human museum worker penned the text?
There’s a lot of controversy right now over how many jobs, and which kinds, will be displaced by emerging technologies. The inroads made by robotics on manufacturing in the 20th century may be reprised in coming decades by artificial intelligence on white collar professions. As I discuss in the Labor 3.0 Chapter of TrendsWatch 2016, programs such as IBM Watson, combining the power of cognitive computing,machine learning & the ability to respond to natural language queries, have already disrupted key functions in law (researching & writing briefs) and medicine (reading medical images, diagnostics). The Associated Press uses software to generate stories on corporate earnings, and an increasing number of small local papers use news bots to report on local and youth sports competitions. Heck, you can even use a free writing bot to auto-generate your own content, if you can feed it data to work with.
But law, medicine, sports results, corporate earnings–these are all, at their heart, factual. The resulting text needs to be accurate, even grammatical, but what about creative writing?
Which brings us to our Futurist Friday video short for the week: in the 9 minute film Sunspring, human actors perform a screenplay written by a “recurrent neural network” named Benjamin. (Originally “Jetson,” but the program re-named itself.) Watch and see what you think:
Incoherent? Well, yes. Though as this review points out, that may in part be a matter of the input Benjamin was working from. And before you snark too much, consider that Sunspring finished in the top 10 of the Sci-Fi London 48 Hour Film Challenge.
Perhaps, in the future, any museum will be able to send a version of Benjamin or Watson to search for open data on any subject–history, art, science–analyze, synthesize and summarize its findings and generate label or catalog copy*. There are two components to this assignment: researching the background information and turning it into coherent, compelling text. The technology needed to conduct such automated research already exists. The second–good writing, with all that this entails, may be further on the horizon, but Benjamin and his digital kin are evolving pretty quickly. So two questions for your FF assignment:
- How long do you think it will be before a research/writing bot available for anyone (students, writers, exhibit developers) to use in their work?
- Is there an essentially “human” element of writing that an artificial intelligence, no matter how advanced, will be unable to replicate?
*You many be horrified by that suggestion, but remember, doctors were just as appalled by the prospect of Watson interpreting symptoms–until IBM demonstrated that the program is more accurate than most human MDs.)