The first issue of Nature in 2026 has an article by science writer David Adam.
The Science of 2050 Nature explores the future breakthroughs that could shape our world.
The online version has a different title and subtitle but the text is the same. It begins with a quote from "futurologist" Nick Bostrum.
“There’s a good likelihood that by 2050, all scientific research will be done by superintelligent AI rather than human researchers. Some humans might do science as a hobby, but they wouldn’t be making any useful contributions.”
There's no attempt in the article to apply critical thinking to such a ridiculous prediction and the author doesn't consider the implications. If Bostrum (whoever that is) is right then that's the end of graduate studies and after 2050 nobody will be getting a Ph.D. in physics, biology, geology, or chemistry.
I hope I live long enough to see AI collecting and analyzing fossils in Greenland or studying volcanoes in Hawaii. Maybe I'll still be around when AI figures out how memories are stored or which transcription factor binding sites are functional in the human genome. And if I'm very, very lucky I'll see live to see all of my colleagues in the Department of Biochemistry abandon their labs and take up some scientific hobby like alchemy or intelligent design.
David Adam and the editors of Nature should be ashamed of themselves for publishing such nonsense.
6 comments :
Sorry to insist on the opposite position, but the rise of AI is clearly an imminent challenge to humanity's position as the main intelligence on Earth. There is nothing implausible about the idea that AI is doing all the science of 2050. What *is* implausible is that it would be business as usual for the rest of humanity (and Bostrom himself would understand this well enough). If scientists get replaced by AI, it will be because humanity in general has been replaced by AI.
I strongly disagree. Just consider Mitchell's chemiosmotic hypothesis. DO you believe that AI would have come up with the same based on the 1961 data. What about evolution theory? Could AI develop the same thoughts and ideas Darwin had back then?
Nick Bostrom is also a longtermist -- a movement of which I hold a low opinion.
If by promoting an "opposite position" you are referring to supporting the part about "all scientific research will be done by superintelligent AI rather than human researchers" and there "is nothing implausible about the idea that AI is doing all the science of 2050", then it's hard to believe you have a clue about what scientific research consists of, its full complement of critical thinking, background research, learning the relevant literature, and acquiring the skills and knowledge and being able to carry out the requisite activities to conduct "all" scientific research.
Science is an empirical endeavor that requires the acquisition of data. Maybe AI can aid humans in designing experiments and suggesting data to collect, but it's unlikely that AI robots will be able in the future to go out into nature everywhere and do what is necessary in even a fraction of scientific fields. Data can perhaps be analyzed by AI, but in the real world actual humans generally have to go out and collect and then analyze the data. Are robots and AI going to be able to go out and do what Jane Goodall did to analyze and understand primate behavior? By 2050? Or even ever? (Is a robot going to be able to be seen by various nonhuman animals as something worth interacting with?)
Are robots and AI going to explore and map caves and do the research that speleologists and biospeleologists and paleontologists do, and maintain the high goals and ideals of cave conservation in caves such as Lechugilla Cave or Fort Stanton's Snowy River Passage?
Ah, Nick Bostrom who somehow gets misquoted so people can make different absurd claims than the ones Nick actually makes. He's the source of the famous "simulation argument", where the original seems to be more giving suggestion that the simulation hypothesis requires certain things to be true, some of which are quite ridiculous if you have a bit of understanding of computation and physics.
Dear Anonymous who asks if AI could have come up with the important scientific theories of the past, or if AI could do field work, data collection, and literature reviews,
Have you ever looked into cognitive and computational neuroscience? This is the naturalistic study of human thought.
A visual neuroscientist like David Marr was able to explain much of how the visual experience of the world is constructed from the input to the optic nerve, by specialized neurons which calculate the existence of progressively more intricate features of the visual scene, as you follow the flow of information deeper into the visual cortex.
Scientific theories, and all the other higher products of human culture, are similarly the product of sophisticated algorithms embodied in our neural circuits. Of course it's more complicated than vision, it involves the work of many brains, practices learned from culture, and trial and error in the physical world. But fundamentally, how science happens in the nervous systems of human scientists, is something that can be studied as a kind of computational process.
Or at least that is what the naturalistic picture implies. The phenomenon of consciousness suggests to me personally that there are still some big gaps in our understanding, but more on an ontological level than an information processing level. The nature of some matter is such that it is conscious, and it is not yet evident what that implies from a physical-science standpoint. But from an information-processing standpoint, cognitive scientists have quite a grip on how to understand world models, discovery, hypothesis generation, in terms of concepts like representations, heuristics, various forms of optimization, etc. These are computational operations that can be carried out by neurons.
Just as biochemical mechanism has obviated the need to posit a vital force, the Turing theory of computation and the comp-sci study of algorithms provides a schema for the mechanistic explanation of all cognition. What occurred in the heads of Darwin and Einstein and in the research activities of all those who were involved in the genesis of evolution and relativity as scientific theories, was a distributed cognitive process occurring in thinking systems with a particular information architecture, interacting with material sources of external information.
Some of it boils down to things as simple as "identify all the possibilities, and test them all in turn". That is something that a scientist might do when facing an unanswered question, and it is an algorithm, it's a brute-force search through a possibility space. More sophisticated methods are simply more sophisticated algorithms. Even intuition is an exaptation of evolved capacities for pattern recognition.
This is why there is no form of human cultural activity that is eternally safe from AIs and robots. Human mental activity is computational in nature. That is one of the great modern insights. These AIs that we have now work quite differently to the human brain in many ways, but they have a family resemblance, they are our siblings in the emergent kingdom of artificial life. And the speed with which their capabilities are advancing, thanks to massive human investment and experimentation, is why I expect them to surpass us in the near future.
Post a Comment