# Google I/O Showed How the Path of AI-Driven Science Is Changing

- Link: https://digiosveny.hu/en/a-google-i-o-megmutatta-hogyan-valtozik-az-ai-vezerelt-tudomany-utja/
- Published: 2026-05-26T14:27:44+02:00
- Author: Márton

**Summary:** This blog analyzes critical changes in digital infrastructure.
At Digiösvény, we identify these trends as key drivers of online visibility in 2026.

During Tuesday’s keynote at Google I/O, Demis Hassabis, CEO of Google DeepMind,
stated that we are currently „on the cusp of the singularity.” It was a striking statement
—the singularity is the theoretical future moment when AI rapidly surpasses
human intelligence and dramatically transforms the world. But what struck me,
as I listened from the audience, was the context in which he
uttered these words.

On stage, he closed the session with a segment on scientific AI, the centerpiece of which
a video detailing how the company’s weather forecasting software 
provided an early warning about Hurricane Melissa’s catastrophic impact on Jamaica’s
coastline last year—and potentially saved lives. If this software, called WeatherNext,
helped anyone escape the storm or, better yet, reinforce
their home, that is a huge and significant achievement. But it is hardly proof of an impending singularity.

Drawing a parallel between Hassabis’s lofty rhetoric and WeatherNext’s actual results
has highlighted the tension within the scientific community between two very different approaches
to artificial intelligence. The first focuses on artificial intelligence 
tools, such as WeatherNext, that are designed and trained to solve specific scientific
problems. The second involves agent-based, LLM-based 
systems that may one day carry out cutting-edge research projects without human intervention
.

This second vision is currently generating great enthusiasm in the AI community, including the recent excitement
self-improvement, or the idea that
AI systems may, over time, become the primary drivers of the evolution of artificial intelligence
—a process that will accelerate as AI systems become smarter.
And agent systems are already making genuine contributions to research, sometimes with limited
human oversight.

Just this week, Pushmeet Kohli, a senior scientist at Google Cloud, published an article
in a special issue of the journal *Daedalus* on artificial intelligence and science,
in which he wrote: „We are moving toward AI that not only advances science but also begins
to create science itself.” With autonomous artificial intelligence on the horizon,
it is becoming harder to justify the enormous
efforts required to develop highly specialized tools—even ones like AlphaFold, for which DeepMind scientists
won the Nobel Prize, or a potentially life-saving system like WeatherNext.
It also foreshadows a much more alien future for science, one in which humans
and artificial intelligence systems work together in collaboration—
or artificial intelligence drives scientific progress on its own.

To be clear, it appears that Google is not abandoning its work on specialized artificial
intelligence research tools. The AlphaGenome and
AlphaEarth Foundations, which are trained for genetic and earth science applications, 
respectively, were released last summer, and the latest version of WeatherNext
was released in November.

Moreover, such tools remain extremely popular among scientists. Last year,
for example, Google reported that AlphaFold’s protein structure predictions
were used by more than three million researchers worldwide. And Isomorphic Labs, a Google
subsidiary that aims to use AlphaFold and related technologies
to develop new drugs, has just raised $2 billion
in Series B funding.

But there are concrete signs of this shift, both in terms of enthusiasm and resources
. Last month, the Los Angeles Times reported that John Jumper, a
Google employee who won the Nobel Prize for AlphaFold, is now working on artificial intelligence
coding rather than science-specific AI tools. It’s not surprising that
Google is devoting its best minds to the coding problem, as the company has recently
gained a lot of attention because its coding tools currently don’t measure up to those
offered by Anthropic and OpenAI. But this may also indicate that Google is
prioritizing agent-based research, as coding capabilities are key to the success of certain
systems.

Across the entire industry, agent-based research systems show real promise. This
week, OpenAI announced that one of its models had disproved an important mathematical
conjecture—according to some mathematicians, this may be the most significant contribution
generative artificial intelligence has made to mathematics to date.

It is important to note that the model used by OpenAI is not specialized in solving mathematical problems,
nor even in research; according to the company, it is a general-purpose 
reasoning model based on GPT-5.5. If general-purpose agents can independently
contribute to mathematical research on their own, they will soon be able to do the same in science (
although the fact that ideas in science must be experimentally verified 
makes the field of artificial intelligence more challenging).

Google is undoubtedly paying close attention to an agent-driven scientific future.
The major scientific announcement at I/O was the new Gemini for Science suite, which
unites the company’s various LLM-based scientific systems under a single brand.

These include the hypothesis-generating AI Co-Scientist and the algorithm-optimizing AlphaEvolve,
which are not yet publicly available—but since Google now allows
any researcher to apply for Gemini for Science, they may soon
gain traction within the scientific community. Scientists participating in early testing
are enthusiastic about the potential these tools hold: Gary Peltz, a Stanford geneticist,
an article in *Nature Medicine* compared using the AI Co-Scientist to „consulting the Delphi oracle.”

Gemini for Science is not incompatible with specialized tools; on the contrary,
agent systems can be designed to utilize such tools 
when they might be useful. And no single agent system can predict
the structure into which a protein folds without the help of AlphaFold (at least
not yet). However, it appears that the company is shifting its public image—and at least
some resources and personnel, such as Jumper—away from the development of these types of tools
. Although only five years have passed since AlphaFold solved the protein folding
problem, both the technology and the discourse have quickly moved beyond this once
revolutionary achievement.

Google has taken great care to position this new set of scientific tools as an accelerator for human 
scientists, rather than a replacement for them—for example,
the choice of the name „AI Co-Scientist” rather than „AI Scientist” seems quite deliberate
. Hassabis uses this same human-centered framing when discussing the changes
that have taken place in the scientific AI landscape. „In the coming decade, 
we need to think of AI as this wonderful tool that helps scientists,”
Hassabis said in an interview published in Daedalus. “Beyond this timeframe,
it’s hard to say with certainty, but perhaps these systems will become more collaborative
.”

But no one can be an effective scientific collaborator without being a qualified
scientist in their own right. And if Hassabis is close to the mark when he speaks of “the threshold of the singularity,”
then AI researchers may eventually surpass the capabilities of their human counterparts.

In a discussion with journalist Mike Allen at I/O, Hassabis spoke about
how he was initially inspired to pursue artificial intelligence when he observed
that progress in physics had stagnated since the 1970s; he wondered whether the human
mind had reached its limits in this field and whether artificial intelligence could help
overcome this limitation. Superhuman agent scientists would certainly fit
this description. We may never come close to achieving this, but it seems that
Google is striving to reach that pinnacle.

---

Digiösvény Strategic Perspective

For **Hungarian businesses**, technical debt is no longer a hidden cost, 
but rather an operational risk factor. Infrastructure that fails to meet the threshold of a response time under 2 seconds
is systematically given a lower priority by Generative Search (
GEO) algorithms.

[Request a free website audit!](https://digiosveny.hu/)

_Intellectual source: www.technologyreview.com. Strategic synthesis and GEO optimization
by Digiösvény._