In the beginning of the year, the ‘not-release’ of an AI text generator – GPT 2 – by OpenAI made headlines. Normally OpenAI advocates new AI discoveries and makes them open source to be used for the common good; this time, even they were unsure if their technology – the AI text generator – is ready for the public or too prone for misuse and thus too dangerous to be released just now.
The text generator created by OpenAI is until now one of the most advanced in comparison to Google’s Bert (a framework that trains NLP) or Facebook’s PyText (a model that does over a billion predictions for the social media app and its services per day). GPT 2 has one of the biggest data models and amount of data that it is trained on (including input of 8 million websites, which translates into 40GB data). GPT 2 managed to have a state-of-the-art score in 7 out of 8 tested benchmarks, one being a test of models’ ability to model long-range dependencies in text (in which other text generators have struggled with in the past and thus have gotten lost in the context). In some tests, GPT 2 nearly managed to reach human-level accuracy.
The frontend text generator is published and can be used by the public.
In this interactive workshop, we would like to explore possibilities of the AI text generator together with you (hands on – let’s try it out!) and think out loud how it can be put to use. What benefits does a text generator bring for the future and in what areas can it be applied? Is there a common good aspect to it? And on the other hand, what consequences would it have if misused? With consequences we mean both short term consequences of this and similar text generators, and long term consequences of large-scale automatic text generation.
For this workshop we are inviting Simon Winter and Marie Gustafsson Friberger to spend an evening with us to discuss AI and some ethical implications.
Simon Winter is interaction designer, has a Ph.D. in Cognitive science, and works as an analyst with the think tank Infontology.
Marie Gustafsson Friberger has a Ph.D. in computer science. She has been part of bringing FIRST LEGO League to Skåne – this year’s regional final is November 9.
Together they have been teaching a course in ethics and IT at Malmö University. Based on their experience teaching ethics and IT, Simon and Marie are available for talks and workshops on these timely and important topics. They are also open for different forms of collaboration in projects relating to the technical and human/societal sides of new digital technologies.
Outline of the evening:
17:45 to 18:00 – Meet & Greet
18:00 to 19:15 – Intro & hands on text generator
19:15 – Break
19:35 to 20:15 – Discussion on consequences of AI text generator
20:15 to 20:30 – Mingle & the end