#Application of AI in professional fields

1 messages · Page 1 of 1 (latest)

dense jungle
#

​ ​ AI is not a simple "plug-in," but requires systematic business restructuring.
​ AI ignorance: technicians do not understand complex industrial mechanisms or legal logic, resulting in the development of tools "difficult to use".
​ Knowledgeable people don't understand ai: experts in traditional fields have fear or blind rejection of ai, and lack the ability to translate business problems into ai tasks.How can AI solve this problem?

stuck elkBOT
#

Hi @dense jungle. I'm kapa.ai, your friendly support bot. I will try to answer your question by searching through the knowledge base. :wave:

#

AI alone can’t fix this gap; organizations need socio-technical changes around it. The knowledge sources point to a few patterns that help bridge “AI people” and “domain experts”:

  1. Use an eval‑driven, iterative project lifecycle
    OpenAI’s partner examples show that successful AI deployments start with:

    • Deeply understanding the problem with domain experts (lawyers, doctors, operations staff, etc.).
    • Assembling real examples of inputs and desired outputs from the existing workflow.
    • Quickly building a V0 end‑to‑end system, then using expert review to correct outputs and design evaluation metrics.
    • Mapping evals to business metrics (cost, risk, speed) so both AI engineers and business owners share concrete success criteria.
    • Iteratively improving the system and evals together, instead of treating AI as a one‑off “plug‑in.” [Eval lifecycle]

    This process forces continuous translation between business reality and AI tasks, reducing “difficult to use” tools.

  2. Use AI to help structure expert knowledge into routines and workflows
    Reasoning models like o1 can take complex internal documents (policies, procedures, decision trees) and convert them into step‑by‑step routines that an LLM can follow. Each step is a clear action (respond, call a function, fetch data), which:

    • Makes the expert’s logic explicit and testable.
    • Gives AI engineers a concrete structure to implement and refine.
    • Reduces hallucinations and deviations from the intended business process. [Routine generation]
#

This is a practical way for “knowledgeable people” to encode their expertise without needing to become ML engineers.

  1. Close the gap with domain‑specific evaluation and fine‑tuning
    In legal and compliance use cases, companies like Harvey, SafetyKit, and Thomson Reuters:

    • Define very specific tasks (e.g., “extract verbatim supporting passages,” “apply nuanced moderation policy,” “compare contracts”).
    • Build task‑specific graders/evals (e.g., F1 on citation overlap, policy‑label accuracy).
    • Use reinforcement fine‑tuning (RFT) so models learn to follow complex domain rules more reliably than general models. [Harvey legal extraction; TR legal skills; SafetyKit moderation]

    This addresses “technicians don’t understand the domain” by letting the data, evals, and feedback from experts teach the model the domain logic.

  2. Build self‑evolving, feedback‑driven agents with experts in the loop
    In regulated fields like pharma, OpenAI’s “self‑evolving agents” pattern:

    • Starts with a simple baseline agent (e.g., summarizing regulatory sections).
    • Collects human or LLM‑as‑judge feedback on quality and compliance.
    • Uses that feedback to iteratively adjust prompts/agents until eval scores cross a target threshold, then promotes the improved agent as the new baseline. [Self‑evolving agents]
#

This gradually shifts experts from doing the work to reviewing and steering the AI, which reduces fear and builds trust.

  1. Choose the right model type for ambiguous, high‑stakes reasoning
    For complex legal, financial, or policy‑heavy tasks, OpenAI recommends reasoning models (o‑series) because they:

    • Handle ambiguous instructions and ask clarifying questions.
    • Find “needles in haystacks” in long, dense documents.
    • Reason across multiple documents and nuanced rules. [Reasoning best practices]

    Using the right model reduces the perception that “AI doesn’t understand our field.”

Putting it together, AI “solves” the problem not by magic, but by:

  • Embedding domain experts into an eval‑driven, iterative lifecycle instead of one‑off tool building.
  • Using reasoning models to structure expert knowledge into routines.
  • Applying domain‑specific evals and RFT so models learn real industrial/legal logic.
  • Running continuous feedback loops (self‑evolving agents) so systems and experts co‑adapt over time.

The knowledge sources don’t provide organizational change recipes (training programs, incentives, etc.), so anything beyond these technical and process patterns would be speculative.

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon: