You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
With the addition highly automated research labs coming to be, will AI/ML that works 24/7/365 be able to dominate cutting edge research by being able to speed up dramatically basic research not only by speed, but being able to operate 24/7/365? Some of this may be guided by humans but greatly increase the speed and reach of what is currently possible. Will AI/ML be able to learn by bruit force methods that no human can match in speed and reach? Not having humans involved will allow labs and other data gathering methods in new and innovative ways achieving efficiencies yet to be even considered or dreamed of. Any additions or comments sought and appreciated.
It's paywalled - but AI/ML has done some impressive things already in this area like being able to determine that some chemical compounds are cancer causing that were not previously realized/known:
An algorithm trained on the properties of more than 35,000 compounds can identify novel cancer-causing substances.
www.nature.com
The thing is though what we call "Artificial Intelligence" is really (imo) just advanced algorythms and pattern recognition. It's not reasoning, self-thinking, has no motivational capability, etc. It's a tool -- albeit a powerful one. This could help greatly accelerate scientific research.
However, for a lot of (scientific cultural/instutional reasons) - scientific productivity has steeply decreased in recent decades according to Sabine Hossenfelder:
(it's worth a watch). One take-away is that the wrong reward systems are in place for pushing science forward, and another is that practices often force people to work on things they have no interest in -- reducing the chance of real innovation in thinking.
I would like to see some examples of recursive AI exploring a problem...
With AI running 24/7, there's no downtime, which means it can be processing data and testing hypotheses around the clock—way faster than any human could. Imagine being able to analyze mountains of data in real-time, spotting patterns or connections almost instantly.
That’s where the real power lies in speeding up basic research, especially in areas like drug discovery or climate change modeling, where time and data are key.
But even though AI can learn at a crazy pace, it’s not all about brute force. Humans still bring the creativity and insight that AI lacks. While AI can crunch numbers and recognize patterns, it still needs humans to guide the research, interpret findings, and come up with the right questions.
The real advantage of AI/ML is to the ability to run literally hundreds to thousands of options in a time frame no human will ever match. Those that don't learn to harvest and use AI/ML to their maximum potential will lose.
The real advantage of AI/ML is to the ability to run literally hundreds to thousands of options in a time frame no human will ever match. Those that don't learn to harvest and use AI/ML to their maximum potential will lose.
I agree with you here, it's like natural selection, and only the most adaptable ones will remain with a good job and will help innovate humanity even more. But I bet there will still be jobs that are not that hard, and every average person should be able to do them
