Why AI Hasn't Taken Over Science (Yet)
Added 2024-11-27 00:37:18 +0000 UTCAI is evolving so fast that soon it might be able to take over scientific research...but what would that look like?
In our latest video, Kate wonders whether spending all that time in school and the field for her PhD was worth it :)
You can watch our it here on Patreon or on Youtube: https://youtu.be/FiCjb4ripBM
Thanks for your support! <3
-David
Comments
I'm a research programmer in an AI department. I have two observations not covered by the video. 1.So far, AI has no initiative. Ultimately, a human must ask (or even build) an AI system to do something. There is an emerging profession called "Prompt Engineer" that's all about getting AI to do complicated things that haven't been done before. 2. Today's leading models (the Large Language Models, LLMs) are trained on massive amounts of human input. We can expect breakthroughs in Undiscovered Public Knowledge, areas where all the pieces are there, but nobody has put them together yet. LLMs in particular are not very good at designing new experiments, much less realizing that new experiments are needed, or coming up with truly novel hypotheses. I'm not saying AI won't get there, but that LLMs are not sufficient for these tasks.
La Monte H. P. Yarroll
2024-11-27 14:38:27 +0000 UTC
