I recently came across an interesting paper, Scaling Laws in Scientific Discovery with AI and Robot Scientists, which proposes a systematic framework for measuring AI’s evolution in scientific discovery from “tool” to “independent researcher.” In the past few years, AI has indeed made remarkable strides in fundamental research:
-
2020: An autonomous robot chemist for synthetic chemistry experiments graced the cover of Nature;
-
2023: CMU released the Coscientist agent, capable of co-designing and executing experiments;
-
2024: Lawrence Berkeley National Laboratory developed the A-Lab platform, achieving a semi-automated experimental workflow.
Today we’ll focus on one small part of the paper—the classification of AI-automated workflows—and connect it to the job-search process.
Similar to the levels used in autonomous driving, the paper divides AI in basic science into six tiers:
-
Level 0 – No AI: Experiments are carried out entirely by dedicated instruments and conventional software, with no intelligent component.
-
Level 1 – Tool-Assisted: Provides basic data retrieval and processing functions but lacks any decision-making autonomy.
-
Level 2 – Intelligent Assistant: Can gather multi-domain knowledge across the web and offer preliminary research suggestions.
-
Level 3 – Collaborative Partner: Deeply integrated with virtual or physical lab environments, supporting semi-automated experiment execution and precise manipulation.
-
Level 4 – Autonomous Researcher: Capable of independently designing and completing complex studies, tackling interdisciplinary problems on its own.
-
Level 5 – Pioneer: Achieves breakthrough discoveries without human intervention, leading entirely new scientific paradigms.
With this classification, it’s clear that most labs today operate around Level 2–3, while only a few top institutions are exploring Level 3–4 scenarios. You can see that Level 4–5 AI almost entirely replaces humans in experimental design and execution. Currently, we’re mainly at Level 2–3—AI remains a “knowledge advisor” or “intelligent assistant.” Once technology reaches the “singularity,” many frontline research roles could be at risk of automation, potentially within the next 10–20 years.
So, whether you’ve just left academia or are considering a career move, it’s worth asking yourself:
-
Which parts of my daily work are already automatable by AI?
-
What level of intelligence have current AI tools achieved in those areas?
Before negotiating salary, considering a company’s growth prospects, or evaluating team culture, it’s crucial to determine if your role is on the “AI-replaceable” list.
Should we therefore pivot into emerging AI roles? The answer is: sometimes yes, sometimes no. Take the “Prompt Engineer” role, for example:
-
In 2023, the market shortage led to a surge in hiring for this role, and salaries skyrocketed.
-
Two years later, as major firms have fine-tuned models to near-natural-language proficiency, standalone Prompt Engineer positions have been scaled back or cut.
But this doesn’t mean prompt-engineering skills have lost value—they’ve simply diffused into many specialized fields: legal research, historical document analysis, chemical experiment design, etc., all now require similar capabilities. In other words, while the Prompt Engineer title may fade, those skills are becoming part of many job descriptions.
Finally, I wish you all equanimity amid the AI trend, success in your job hunt, and a bright future!