Alien Insights: How AI Predicts Discoveries

0:00The wait is over. Dive into Audible's most anticipated collection, The Best of 2025. featuring top audiobooks, podcasts, and originals across all genres. Our editors have carefully curated this year's must-listens from brilliant hidden gems to the buzziest new releases. Every title in this collection has earned its spot. This is your go-to for the absolute best in 2025 audio entertainment. Whether you love thrillers, romance, or nonfiction, your next favorite listen awaits. Discover why there's more to imagine when you listen at audible.com slash best of the year. What can 160 years of experience teach you about the future?

0:44When it comes to protecting what matters, Pacific Life provides life insurance, retirement income, and employee benefits for people and businesses building a more confident tomorrow. Strategies rooted in strength and backed by experience. Ask a financial professional how Pacific Life can help you today. Pacific Life Insurance Company, Omaha, Nebraska, and in New York. Pacific Life and Annuity, Phoenix, Arizona. An incredible new scientific report was just published in the journal Nature, and it kind of goes over this really interesting new use case where they actually used AI to help them expand scientific discoveries by predicting and generating hypotheses that humans might not have considered.

1:27So the idea behind these is that they were essentially the researchers built these models that generated scientifically promising quote unquote alien hypotheses that would be very unlikely for an actual person to predict or to come up with. So I think this is really interesting today on the podcast we're gonna be diving into what the implications of this are and how this model actually works and some of the findings of the model. So without further ado, let's jump into it. I think the research team was led by James A. Evans, the Max Pavley professor in the Department of Sociology, and also the director of the Knowledge Lab.

2:04And this whole research report kind of takes AI a step further. They essentially develop models that skirt human inference to generate what are called these alien hypotheses, which essentially, I think, really open up a lot of scientific possibilities that humans might only consider in the distant future, if at all. So these essentially are hypotheses that are very unlikely for a human to come up with. And so by being able to come up with those and by being able to test those, this brings a whole new field of what we're actually able to predict and discover in science thanks to AI. This is what Evans said about it.

2:42He said, our research aims to go at going beyond the current scientific frontier by understanding what people are doing. We can both enhance prediction and outpace them to fast track science more so by identifying what people can't do or won't be doing for years to come. We can bolster them with this complementary intelligence. So AI models trained on published scientific findings have previously been used to innovate valuable materials and targeted therapies. But a lot of those same models often overlook the human component. So essentially, like the dynamic distribution of scientists engaging in the process.

3:21So this study that just came out, it kind of looks over how human competition and collaboration research could influence AI's potential if they were built aware of human expertise. So by simulating reasoning processes via random walks across research literature, the team ventured to predict future discoveries. So they started with a property like COVID vaccination, and then they leaped to related papers or cited materials. And after running millions of these quote unquote simulated walks, their model significantly outperformed with a 400 % improvement in predicting future discoveries, particularly when relevant literature was pretty scarce.

4:06So that's usually a pretty big roadblock. And that's why it's super impressive that it was able to do that. I'm in that case. So I think impressively, it also managed to pinpoint over 40 % precision in the actual individuals would likely make these discoveries. So as it recognized their unique experiences or conflict or connections to the topic. So this to me is quite amazing, because really what they're doing is they're able to look at, you know, who's published what who's talking about what what scientific discoveries we need to make it's, you know, coming up with this alien hypothesis of something.

4:38And then it's been able to say, based off of this alien hypothesis, who would make this discovery based off of all the literature that's been published. And with a 40 % accuracy, they were able to predict, of course, you know, they'll retroactively do this in the past, but now we can start doing this into the future and predict who will make a discovery. So if you have this kind of knowledge, you can see the power here, right? You're essentially able to say, well, you know, if we want to get X, Y, and Z outcome, we're gonna need x y z person to be working on it and then you're able to get those scientists to actually start working on you know whatever the specific thing you're trying to accomplish is and with a 40 accuracy that is incredibly higher than just having you know random scientists working on it you're getting people that have the actual expertise that have knowledge in all the specific areas i think this is going to be a really big breakthrough for scientists and for scientific discovery and this is what evan says about it he said we've essentially created a double of the scientific system, allowing us to simulate likely occurrences and experiment with alternative possibilities.

5:39The system has a capacity to shed light on the inherent limitations of our current scientific practices, especially those aspects like graduate education, which seems to prioritize job marketability over discovery, right? This is a common problem that we have with this kind of graduate education. And of course, at the end of the day, in the grand scheme of things, job, you know, job market marketability is not that important when, you know, compared to actually discovering things and discovering new drugs or new therapies or new innovations in science. So the team's second experiment after that first one asked the AI model to identify scientific predictions that are plausible but least likely to be discovered by humans.

6:25So, you know, these alien hypothesis or complementary inferences. So these are likely to be better than human inferences as humans often exhaust an existing theory or approach before starting to look at a new one. So in contrast, these models step into uncharted territory, steering clear of common human scientific patterns. So Evans, who kind of was leading the study, he argued that viewing AI an attempt to replicate human intelligence, which is echoing, you know, Alan Turing's concept of the imitation game, does very little to accelerate our problem-solving capacity. So he says we should actually aim for a radical augmentation of our collective intelligence.

7:07And this is specifically what he said, quote, individuals in fields like science, technology, and culture often try to stay close to the pack to maintain influence. Our models counterbalance this bias by following signals of scientific plausibility while consciously avoiding the pack. And, you know, he said that because I think this really allows AI to venture outside of current methods and collaborations. And I think this is actually going to enhance humans' capability for exploration because we're really getting beyond what we traditionally do. We're getting beyond the traditional patterns that science follows and being able to think outside the box in a new way.

7:46So Evans proposes this shift in the AI paradigm from, quote, artificial intelligence to radically augmented intelligence. So that's what he says that, you know, the paradigm shift needs to be. But I think really what this means is it kind of implies a deeper study of individual and collective cognitive capacity. So understanding human limitations and designing systems to compensate for these is, I think, at the end of the day, going to be the most effective thing that we can do. he said by learning more about human cognition we can deliberately design systems that account for its limitations leading us to a more collective understanding so i think really a key takeaway from all of this is the potential for ai to illuminate some of the blind spots in human scientific exploration i think the concept of alien hypothesis suggests that while human scientists are primarily concerned with immediate and familiar research avenues ai can actually look further and it can actually go and identify promising but less considered pathways.

8:44So these inferences, which are considered alien because they are, you know, rarely ventured into by humans, I think these are actually going to bring some massive advancements in science. I think that harnessing the underutilized areas of the scientific landscape is going to be what makes some of our biggest discoveries. And this is exactly what AI is going to excel at and what this AI they've developed is doing right now. So I think this concept of, you know, quote unquote radically augmented intelligence offers a very compelling new direction for AI. I think it's an approach that doesn't just aim to mirror human thought process, but actually aims to challenge, augment, and really complement them.

9:24So enhancing overall our collective understanding and capacity. I think that by implementing this vision in particular, I think we're going to unlock a lot of different avenues for scientific discovery. And of course, these are ones that may have otherwise remained undiscovered. And so I think because of that, this is going to lead us into an entirely new approach of scientific innovation and understanding. If we really follow what came out of this study, what they learned, what the key takeaways were, I think this is going to unlock a lot for us. I think at the heart of this study overall is kind of the idea that AI can and should be designed to operate beyond the confines of existing human thought processes and biases.

10:03but I think that in doing this we're actually going to be moving towards a future where AI rather than simply mirroring human intelligence actually broadens it and I think the consequences of this could be a total you know a total paradigm shift in how we actually utilize AI in scientific discovery and reshaping perhaps you know like the nature of research and innovation as we see it today so I think it's gonna be amazing I think this is gonna open the doors of a future filled with a lot of scientific discovery. This is something I'm going to be following closely and I'm excited to take you along and I'll continue reporting on it in the future.