Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
The Japanese emerging Sakana said that artificial intelligence had been born The first scientific publication reviewed by peers. But although the claim is not incorrect, there are great warnings that must be observed.
the The discussion is discussed on artificial intelligence and its role in the scientific process It grows more severely a day. Many researchers do not believe that artificial intelligence is fully prepared to work as a “co-player”, while others believe that there are potentials-but they admit that they are the first days.
Sakana is located in the last camp.
The company said it used the AI system called the world of artificial intelligence-V2 to create a paper presented by Sakana and then to an ICLR workshop, a long and good Amnesty International Conference. Sakana claims that the workshop organizers, as well as ICLR leadership, agreed to work with the company to conduct an experience to review the manuscripts created from artificial intelligence.
Sakana said that she collaborated with researchers at the University of British Columbia and the University of Oxford to present three papers created from artificial intelligence to the aforementioned workshop for reviews. AI-V2 created the papers “from finish to end”, including scientific hypotheses, experimental experiences, data code, data analyzes, perceptions, text and addresses.
“We have created research ideas by providing an abstract workshop and description of Amnesty International,” Robert Lang, a research scientist and founding member of Sakana, told Techcrunch via email. “This guarantees that the papers created were on an appropriate topic and presentations.”
One out of the three sheets is accepted in the ICLR workshop – a critical lens pillar sheet on training techniques for artificial intelligence models. Sakana said that she immediately pulled the paper before it was published in favor of transparency and respecting ICLR agreements.
“Each of the acceptable paper offers a new promising way to train nervous networks and shows that there are remaining experimental challenges,” Lang said. “It provides an interesting data point to raise more scientific investigation.”
But the achievement is not impressive as it might seem at first glance.
In a blog post, Sakana admits that artificial intelligence has sometimes committed “embarrassing” quotation errors, for example incorrectly attributed to the 2016 paper instead of the original 1997 work.
Sakana paper also did not subject to a lot of scrutiny, like some other publications that the pendants reviewed. Since the company pulled it after the preliminary peer review, the paper did not receive an additional “twisting reference”, as the workshop organizers had rejected it in theory.
Then there is the fact that the admission rates for conference workshops tend to be higher than the admission rates for the main “path path” – a fact that Sakana reminds frankly in her blog. The company said that none of its studies created from artificial intelligence have passed its internal tape to publish the ICLR course.
Matthew Gozdiel, a researcher of artificial intelligence and assistant professor at the University of Alberta, described Sakana’s results as “somewhat misleading.”
He said via e -mail: “Sakana people chose the papers from a number of number of forms that were created, which means that they were using human rule in terms of choosing the outputs that they believe might enter.” “I think this indicates that human beings in addition to artificial intelligence can be effective, and not that artificial intelligence alone can create scientific progress.”
Mike Cook, a research fellow in Kings College in London, specialized in artificial intelligence, questioned the accuracy of the auditors and the workshop.
“New workshops are often reviewed, such as one, by more novice researchers,” Techcrunch told Techcrunch. “It should also be noted that this workshop relates to negative results and difficulties – which is great, I have been running a similar workshop before – but it is easier to get Amnesty International to write about failure convincingly.”
Cook added that he was not surprised that artificial intelligence could pass a peer review, bearing in mind that artificial intelligence exceeds the writing of human prose. partially-Created papers Cook pointed out that the review of magazines is not new, as Cook pointed out, and that the moral dilemmas offered by science.
AI’s technical palaces – like her tendency to Hallucinogenic Make many scientists be careful not to support hard work. Moreover, experts fear artificial intelligence simply It ends with noise generation In scientific literature, not raising progress.
“We need to ask ourselves whether the result of (Sakana) revolves around the quality of artificial intelligence in the design and conduct of experiments, or whether it is about the quality of ideas for humans – which we know is really wonderful in that,” Cook said. “There is a difference between peer review and contributing to a field.”
Sakana, at its estimate, does not claim that artificial intelligence can result from scientific work – or even new – in particular -. Instead, the aim of the experiment was to “study the quality of research created from artificial intelligence,” the company said, and highlighting the urgent need for “the rules related to the science created from artificial intelligence.”
The company wrote: “(R) are difficult questions about whether the flag (which was created from artificial intelligence) should be judged by its special advantages first to avoid bias against it.” “To move forward, we will continue to exchange opinions with the research community on the state of this technology to ensure that it does not develop into a position in the future where its only goal is to pass the peer review, thus undermining the meaning of the process of reviewing the scientific peers significantly.”