%0 Journal Article %1 Bajohr_2023 %A Bajohr, Hannes %D 2023 %I Herbert von Halem Verlag %J IMAGE %K artificial_intelligence machine_learning semantics %N 1 %P 58–70 %R 10.1453/1614-0885-1-2023-15452 %T Dumb Meaning: Machine Learning and Artificial Semantics %U https://doi.org/10.1453/1614-0885-1-2023-15452 %V 37 %X The advent of advanced machine learning systems has often been debated in terms of the very ‘big’ concepts: intentionality, consciousness, intelligence. But the technological development of the last few years has shown two things: that a human-equivalent AI is still far away, if it is ever possible; and that the philosophically most interesting changes occur in nuanced rather than overarching concepts. The example this contribution will explore is the concept of a limited type of meaning – I call it dumb meaning. For the longest time, computers were understood as machines computing only syntax, while their semantic abilities were seen as limited by the ‘symbol grounding problem’: Since computers operate with mere symbols without any indexical relation to the world, their understanding would forever be limited to the handling of empty signifiers, while their meaning is ‘parasitically’ dependent on a human interpreter. This was true for classic or symbolic AI. With subsymbolic AI and neural nets, however, an artificial semantics seems possible, even though it still is far away from any comprehensive understanding of meaning. I explore this limited semantics, which has been brought about by the immense increase of correlated data, by looking at two examples: the implicit knowledge of large language models and the indexical meaning of multimodal AI such as DALL·E 2. The semantics of each process may not be meaning proper, but as dumb meaning it is far more than mere syntax.

@article{Bajohr_2023, abstract = {The advent of advanced machine learning systems has often been debated in terms of the very ‘big’ concepts: intentionality, consciousness, intelligence. But the technological development of the last few years has shown two things: that a human-equivalent AI is still far away, if it is ever possible; and that the philosophically most interesting changes occur in nuanced rather than overarching concepts. The example this contribution will explore is the concept of a limited type of meaning – I call it dumb meaning. For the longest time, computers were understood as machines computing only syntax, while their semantic abilities were seen as limited by the ‘symbol grounding problem’: Since computers operate with mere symbols without any indexical relation to the world, their understanding would forever be limited to the handling of empty signifiers, while their meaning is ‘parasitically’ dependent on a human interpreter. This was true for classic or symbolic AI. With subsymbolic AI and neural nets, however, an artificial semantics seems possible, even though it still is far away from any comprehensive understanding of meaning. I explore this limited semantics, which has been brought about by the immense increase of correlated data, by looking at two examples: the implicit knowledge of large language models and the indexical meaning of multimodal AI such as DALL·E 2. The semantics of each process may not be meaning proper, but as dumb meaning it is far more than mere syntax.}, abstractnote = {The advent of advanced machine learning systems has often been debated in terms of the very ‘big’ concepts: intentionality, consciousness, intelligence. But the technological development of the last few years has shown two things: that a human-equivalent AI is still far away, if it is ever possible; and that the philosophically most interesting changes occur in nuanced rather than overarching concepts. The example this contribution will explore is the concept of a limited type of meaning – I call it dumb meaning. For the longest time, computers were understood as machines computing only syntax, while their semantic abilities were seen as limited by the ‘symbol grounding problem’: Since computers operate with mere symbols without any indexical relation to the world, their understanding would forever be limited to the handling of empty signifiers, while their meaning is ‘parasitically’ dependent on a human interpreter. This was true for classic or symbolic AI. With subsymbolic AI and neural nets, however, an artificial semantics seems possible, even though it still is far away from any comprehensive understanding of meaning. I explore this limited semantics, which has been brought about by the immense increase of correlated data, by looking at two examples: the implicit knowledge of large language models and the indexical meaning of multimodal AI such as DALL·E 2. The semantics of each process may not be meaning proper, but as dumb meaning it is far more than mere syntax.}, added-at = {2023-08-10T16:37:20.000+0200}, author = {Bajohr, Hannes}, biburl = {https://www.bibsonomy.org/bibtex/27bb9373216583c76d581c0087ed9d4e4/meneteqel}, doi = {10.1453/1614-0885-1-2023-15452}, interhash = {05023cea8cdff1a404ba55ac589cbb77}, intrahash = {7bb9373216583c76d581c0087ed9d4e4}, issn = {1614-0885}, journal = {IMAGE}, keywords = {artificial_intelligence machine_learning semantics}, month = may, number = 1, pages = {58–70}, place = {DE}, publisher = {Herbert von Halem Verlag}, timestamp = {2024-04-24T09:27:52.000+0200}, title = {Dumb Meaning: Machine Learning and Artificial Semantics}, url = {https://doi.org/10.1453/1614-0885-1-2023-15452}, volume = 37, year = 2023 }