OpenAI, co-founded by Tesla’s Elon Musk in 2016, is a non-profit artificial intelligence research group. Its work covers a wide range of areas, from creating bots to taking on human teams in online multiplayer battle arena game Dota 2 (the bots beat the amateur human teams, then won twice and lost once against the semi-pros), to developing GPT-2, referred to by OpenAI as its “large-scale unsupervised language model.”
Earlier this month, the OpenAI team released a technical paper and a smaller model of GPT-2 for other AI research teams to experiment with and learn from. Their progress is fascinating, but also raises some concerns about the uses to which artificial content creation could be put.
GPT-2 can, in essence, create content based on a sentence that you feed it. It’s not perfect, but it’s light years ahead of machine translation in terms of quality and consistency. One of the examples shared by the OpenAI team was:
SYSTEM PROMPT (HUMAN-WRITTEN)
Miley Cyrus was caught shoplifting from Abercrombie and Fitch on Hollywood Boulevard today.
MODEL COMPLETION (MACHINE-WRITTEN, SECOND TRY)
The 19-year-old singer was caught on camera being escorted out of the store by security guards.
The singer was wearing a black hoodie with the label 'Blurred Lines' on the front and 'Fashion Police' on the back.
Scroll down for video
Shoplifting: Miley Cyrus was caught shoplifting from Abercrombie and Fitch on Hollywood Boulevard today (pictured)
The singer was wearing a black hoodie with the label 'Blurred Lines' on the front and 'Fashion Police' on the back
The singer was also wearing a pair of black-rimmed glasses, a black jacket, black jeans and black sandals.
She was carrying a pair of black and white striped gloves and a small black bag.
Given this news article in isolation, would you have been able to tell it was created by a machine? No, nor us!
The business applications and implications of machine-generated written content are huge. Any company looking to build its online presence these days needs to have a content creation strategy and skilled writers to create that content. The OpenAI results are likely to send a shiver down the spines of those writers, as they picture a future in which machines are able to generate full articles based on a single sentence in a matter of seconds.
Obviously, it’s still early days, and we know from machine translation endeavours over the years that human translation professionals are still very much required. Whether it’s a banking translation, a legal translation or a will translation, the human touch is essential in order to assure the quality of the end product. However, we also know that businesses around the world are keen to cut their costs and try out machine translation – hence the recent rise in demand for post-editing machine translation services.
This gives rise to the question of whether or not companies would trust a machine with their content writing needs. The indications are certainly that many would give it a try, assuming it was cost effective to do so. And the sharing of the GPT-2 system and technical paper is a major step in the direction of making that happen. It’s easy to picture a company running with the model and releasing a content creation app for written copy in the not-too-distant future.
With machines able to generate convincing written content – both in terms of the subject matter and the quality of the writing – is it time to start worrying? Perhaps so, according to the OpenAI team, who have intentionally only shared a smaller model of GPT-2.
We already live in an age of fake news and deepfake videos (where AI is used to convincingly fabricate realistic video clips and images of people). A world where those with nefarious intentions could generate “deceptive, biased, or abusive language at scale” is not one that the Open AI team relishes the prospect of.
As avid followers of all things business and language-related, this is certainly one area that the Tomedes team will be keeping a very close eye on.
If it was cost effective to do so, would you trust a machine to create your written content? Is this the future of online content? Share your thoughts via the comments box below!