Position: There are futures that benchmark-driven AI cannot see

Sobhan Lotfi, Ava Iranmanesh, Lachin Naghashyar, Ali Shirali, Fateme Nateghi Haredasht, Sanmi Koyejo, Philip Torr, Yong Suk Lee, Fazl Barez, Joel Lehman, Peter Norvig, Arvind Narayanan
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:171375-171387, 2026.

Abstract

In biology, traits evolved for one function sometimes become decisive for another; this is exaptation, and scientific progress works the same way. AI’s benchmark-centered selection environment taxes exaptation: when one selection rule dominates, ideas that do not fit it have nowhere to persist. The cost grows acute as the field shifts from asking can machines exhibit intelligent behavior? to asking can they do so while remaining aligned, interpretable, and safe? These are philosophically distinct questions that may require discoveries we cannot specify. We propose mechanisms to restore exaptive capacity without abandoning benchmarking: plural evaluation regimes, protected venues for non-comparable work, long-horizon funding, and training norms that encourage researchers to question selection rules, not only optimize within them.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-lotfi26a, title = {Position: There are futures that benchmark-driven {AI} cannot see}, author = {Lotfi, Sobhan and Iranmanesh, Ava and Naghashyar, Lachin and Shirali, Ali and Haredasht, Fateme Nateghi and Koyejo, Sanmi and Torr, Philip and Lee, Yong Suk and Barez, Fazl and Lehman, Joel and Norvig, Peter and Narayanan, Arvind}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {171375--171387}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/lotfi26a/lotfi26a.pdf}, url = {https://proceedings.mlr.press/v306/lotfi26a.html}, abstract = {In biology, traits evolved for one function sometimes become decisive for another; this is exaptation, and scientific progress works the same way. AI’s benchmark-centered selection environment taxes exaptation: when one selection rule dominates, ideas that do not fit it have nowhere to persist. The cost grows acute as the field shifts from asking can machines exhibit intelligent behavior? to asking can they do so while remaining aligned, interpretable, and safe? These are philosophically distinct questions that may require discoveries we cannot specify. We propose mechanisms to restore exaptive capacity without abandoning benchmarking: plural evaluation regimes, protected venues for non-comparable work, long-horizon funding, and training norms that encourage researchers to question selection rules, not only optimize within them.} }
Endnote
%0 Conference Paper %T Position: There are futures that benchmark-driven AI cannot see %A Sobhan Lotfi %A Ava Iranmanesh %A Lachin Naghashyar %A Ali Shirali %A Fateme Nateghi Haredasht %A Sanmi Koyejo %A Philip Torr %A Yong Suk Lee %A Fazl Barez %A Joel Lehman %A Peter Norvig %A Arvind Narayanan %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-lotfi26a %I PMLR %P 171375--171387 %U https://proceedings.mlr.press/v306/lotfi26a.html %V 306 %X In biology, traits evolved for one function sometimes become decisive for another; this is exaptation, and scientific progress works the same way. AI’s benchmark-centered selection environment taxes exaptation: when one selection rule dominates, ideas that do not fit it have nowhere to persist. The cost grows acute as the field shifts from asking can machines exhibit intelligent behavior? to asking can they do so while remaining aligned, interpretable, and safe? These are philosophically distinct questions that may require discoveries we cannot specify. We propose mechanisms to restore exaptive capacity without abandoning benchmarking: plural evaluation regimes, protected venues for non-comparable work, long-horizon funding, and training norms that encourage researchers to question selection rules, not only optimize within them.
APA
Lotfi, S., Iranmanesh, A., Naghashyar, L., Shirali, A., Haredasht, F.N., Koyejo, S., Torr, P., Lee, Y.S., Barez, F., Lehman, J., Norvig, P. & Narayanan, A.. (2026). Position: There are futures that benchmark-driven AI cannot see. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:171375-171387 Available from https://proceedings.mlr.press/v306/lotfi26a.html.

Related Material