(Acknowledgments for COST can be found at the bottom of the page.)

As you might have concluded from my about page I’m on a ‘crusade’ if you will, to put a new footprint in academia and after visiting the 1st Dutch Speech Tech Day, I got aquainted with people from Campus Fryslân, (Language, Technology & Culture), feeling quite the connection. Although at first my abstract wasn’t accepted, I feel obligated to thank the people who decided not to come at last notice, so I had the chance to present my insights on current developments in speech technology. If you’ve read my page on the BIAS presentation (2019), it’s something that I’m very interested in, but also worry about. Although at that time GPT- (Generative Pre-trained Transformer) systems for LLM’s (Large Language Models) based on NLP’s (Natural Language Processing) didn’t particularly fly this high yet.

I like to share my insight in results and embedding of these developments in a more existentialist and philosophical context of us humans or ‘Homo sapiens sapiens’ as a species of animals, for I am a biologist by trade. I look at our behaviour and activities for the kind of animals that we are and most of the time I will not be kind to interpretations putting us on a developmental pedestal. For me there are two kinds of actual behaviour… those serving us to arrange for a tenable place in our habitat and the kind of behaviour that I’d define as ‘play’ and not actually needed as such. Play of course has a function for the first kind of behaviour, because it will give room to express behaviour without consequences in general.

Perhaps needless to say that I consider most computer assisted activities as ‘play’. Interesting exceptions being the ones where these systems can help us with observation, categorisation and remembering our habitat (which apparently now includes electronically stored references). We can develop all kinds of algorithms to do so, but we have to keep in mind that the process it’s following is trackable. As such, I keep telling people that A.I. systems should be able to explain themselves about their output. This means that as it’s doing a rehash of information found in its databases to express the findings in a most ‘Human’ way, it must be able to reference for it. For the semantics rendered, it should be able to show the context it’s derived from. As long as this is not the case, any content automatically rendered is superfluous as information and should not be used for content to share.

Of course, what I am telling you on this page is content I deem worth sharing. This is in short, the basis for my presentation at LITHME 2023 and I’ll give you both abstract and main pages of the powerpoint presentation. Their site may lead you to to the actual presentation as it is recorded, but I consider those presentations contextual and fleeting. The introduction you’ve just been reading, was the reasoning behind using the title:

Stochastic Parrots, Storytelling and Behavioural Complexification:

Throw Away After Use!

For the definition ‘Stochastic Parrots” and references: (Emily Bender & Timnit Gebru, 2021, https://dl.acm.org/doi/10.1145/3442188.3445922)