Science (completely) without people?

Text Mai allo images juho aalto english translation marko saajanaho

The publication threshold is crossed. The costs: 15 dollars per paper.

An article published in Nature in late August makes one think. This “AI Scientist” does quickly and cheaply what a research group of people requires time, money, food, and an occupational doctor for.   

Thus far, AI has only been able to replace a person in research regarding AI itself. And even then, not everything went swimmingly. However, the human scientist creators of the AI Scientist claim their program can be applied to any field, such as biology or physics.

So, do we need people for anything in science from now on, or should Homo sapiens be sent packing from the campus altogether?

So, do we need people for anything in science from now on, or should Homo sapiens be sent packing from the campus altogether?

Acatiimi went over the Nature news piece and the ArXive publication it was based on with a few experienced researchers from different fields. Free thinking and speculation over a cup of coffee might not have produced scientifically viable results, but we did at least get some ammunition.

To start with, people being fully replaceable in terms of producing new data is likely to vary between fields. And secondly – the more science is produced in a faster and more automated manner, the likelier it is for people to be required as gatekeepers for this new data.

“Data does not jump into machines from nothing”

Juho Aalto is in charge of the Hyytiälä Forest Station in Juupajoki. His background is in forestry, and much of his career has been spent working on atmospheric research.

Aalto is a natural scientist through and through, and thus also a scientific handyman capable of designing, fixing, and building devices and equipment. He knows how to measure natural phenomena and translate them from mechanic signals to electric signals first and, from there, to usable data.

Aalto has climbed measuring towers, attached cuvettes to tree trunks, shovelled snow to uncover a sensor buried under snow, and monitored mass spectrometer activity in the middle of a forest. No wonder his first question regarding AI-made science is where the automatic researcher gets its data.

“Nothing and no one can analyse data unless someone produces data – getting measurement results and processing the information to make it usable. Measuring equipment itself often works automatically, but someone needs to consider what is measured and how, the best equipment for the job and how it is built, as that sort of equipment usually cannot be bought off the shelf.

“Then the equipment must be installed, serviced, and modified to suit different environmental conditions such as a lake or a hole in the ground.

“At least we need several people for this rigmarole: people with both theoretical and technical expertise, finger dexterity, muscle strength, and even motivation to manage completely unpredictable terrain conditions. At this point, there is no such program or robot connected to it that could handle all these stages so autonomously that no human input is required.”

Aalto recognises the limited scope of his viewpoint.

“I am not an expert on AI or machine learning in any way, shape, or form, but maybe that comes in handy in this discussion. AI results are often presented by data science experts. They may forget the massive amount of mental and physical work required to have any reliable data or measurement results to begin with.”

“AI will replace people in AI research”

What do these proper AI experts think about the AI Scientist, then?

Alex Jung, Associate Professor of Computer Science at the Aalto University and expert on machine learning, states outright that AI will replace people in AI research. Perhaps including him.

“Until now, people have been assisted by some program in mathematical modelling, for example, looking for the most representative and predictive model, or mathematical form. AI can now do this autonomously. When it can also form a sensible research question and write an article, no human input is required.”

”But in the end, the human is the the gatekeeper of science and new information. If peer review is left to AI, we run into technical, ethical, and even bibliometric problems.”

Alex Jung, Associate Professor of Computer Science at the Aalto University

However, Jung reckons this development is unlikely to actualise in other fields such as natural and laboratory sciences.

“It will not happen easily in mathematics either, where human intuition plays a role.”

In any case, Jung is excited about the development in his field.

“But in the end, the human is the the gatekeeper of science and new information. If peer review is left to AI, we run into technical, ethical, and even bibliometric problems. Just think about it; anyone could boost their bibliography in no time at all with hundreds of intrinsically faultless but perhaps meaningless papers.”

Associate Professor of Computer Science at the Aalto University

Kalle Kytölä, Associate Professor of Mathematics and Systems Analysis at the Aalto University, reads the Nature and ArXive articles about a cost-effective “automatic research group” with great interest.

He thinks that if something is easy and cheap to produce, then it will be produced – in large quantities. And what happens if this producer is also responsible for its own peer review process? Will the scientific community and the whole world drown in journals and articles no human being has time to read or check, let alone use the data in properly thought-out manner?

“Social media, and soon the rest of media and the digital world, is already bursting with fake news, videos, and scams. We all have to spend time verifying the origin of all kinds of data and ensure its legitimacy. There are still only 24 hours in a day!” Kytölä cries out.

Regardless, he reminds us that AI has established its place as an important tool across the scientific community. This can be seen in the research results of the most recent Nobel winners, for example.

Despite the rapid development, Kytölä does not expect a revolution per se.

Just one year ago, no artificial intelligence at all could succeed at the international competition for young mathematics, the Mathematical Olympiad. Now, AI is solving mathematics competition problems at a silver medallist level.

Another good example cited by Kytölä is the fact ChatGPT was a useless nonsense generator from a professional mathematician’s perspective just a year and a half ago. Currently, language model-based programs can be augmented with parts to automatically check the logical structure of the output, which supports the work of a human mathematician.

“I do not expect AI to replace human beings in mathematics in the near future”, Kytölä says.

Tämän artikkelin ideoi ja kirjoitti ihminen.

Sources:

Nature 30/8/2024: “Researchers built an AI Scientist – What can it do?”

Lu & al: “The AI Scientist: towards fully automated open-ended scientific discovery”

arXiv: 2408.06292v3

cs.AI

1 Sep 2024

For this article, interviews were also conducted with e.g., Professor of Computer Science Hannu Toivonen from the University of Helsinki.  https://www.mitätekoälyon.fi