A London-based artificial intelligence company founded by former Google DeepMind researchers is claiming a major breakthrough in automated scientific research.
Inherent says its AI agent, Faraday, has demonstrated an ability to independently reproduce findings from published scientific papers.
The company says the system performed better on this particular task than larger models developed by some of the biggest names in artificial intelligence.
If independently validated, the development could represent an important step toward AI systems capable of conducting scientific research with less human intervention.
From chatbots to research agents
Most public AI systems are designed primarily to answer questions.
Newer systems are increasingly being designed to complete tasks.
These agentic systems can search for information, write code, analyse data and use software tools.
Scientific research is a particularly challenging application.
A research agent needs to understand scientific literature, develop a methodology, conduct experiments and analyse results.
It must also avoid making unsupported conclusions.
What Faraday is designed to do
According to Inherent, Faraday focuses on reproducing scientific research.
The system is given a published paper and must attempt to replicate its findings.
This is significantly different from simply summarising the research.
Replication requires the AI to understand what the original researchers did and then reproduce the process.
That could involve writing code, finding datasets and interpreting results.
Why replication matters
Scientific progress depends heavily on reproducibility.
Researchers need to know whether findings can be independently confirmed.
However, replication is often time-consuming.
Scientists must spend hours reading papers, obtaining data and recreating experiments.
An AI system capable of automating some of this work could significantly accelerate research.
It could also help identify errors or weaknesses in published studies.
The limits of AI research
Despite the excitement, AI research agents still face major challenges.
Scientific experiments can be highly complex.
Some require specialised equipment.
Others depend on information that is not publicly available.
An AI may also misinterpret experimental procedures.
Therefore, human researchers will remain important.
The more realistic near-term scenario is likely to involve AI assisting scientists rather than completely replacing them.
Researchers could use AI agents to perform repetitive tasks while focusing their attention on experimental design and interpretation.
Competition is intensifying
Inherent’s claims are also significant because the company is relatively young.
It emerged from stealth with a $50 million seed investment.
That illustrates the amount of investor interest in AI systems designed for specialised tasks.
The industry is gradually moving beyond general-purpose chatbots.
Companies are now building AI systems for coding, research, medicine, finance and other professional fields.
Specialisation could become a major trend.
What this means for Africa
AI research tools could also have significant implications for African universities and research institutions.
Many researchers face limited access to expensive laboratory infrastructure and large research teams.
AI systems could potentially reduce some of the time and resources required for literature reviews, coding and data analysis.
However, access to computing infrastructure remains a challenge.
African institutions may need stronger cloud and data-centre capacity to fully benefit from advanced AI.
The next frontier
The development of research-focused AI agents represents an important shift.
Artificial intelligence is gradually moving from systems that generate information to systems that can perform multi-step intellectual tasks.
Whether Faraday’s performance claims hold up under independent testing will be important.
If they do, AI could become a much more active participant in scientific discovery.
That would mark another major stage in the evolution of artificial intelligence.
