In financial AI, transparency makes outputs and decision processes reviewable, supporting scrutiny, compliance, and public trust. It can provide an audit trail and help people investigate questionable recommendations. section 2.5.2 section 12.7.3
Practical transparency measures include:
- Show the sources behind generated summaries and use prompt guardrails to encourage objective outputs. section 2.5.2
- Audit outputs for demographic or categorical imbalance; transparency supports scrutiny but does not replace fairness audits. section 2.5.2
- Keep a decision log recording available data, recommendations and reasoning, flags, and outcomes. section 12.7.3
Listing 12.13 Decision log output for human review
for log in simulation_log:
print(f"\n{log['date']} (Conviction: {log['conviction']}/10)")
print(f" Reasoning: {log['reasoning']}")
w = log['weights']
weight_str = " ".join([f"{etf}={w[etf]:.0f}%" for etf in ETF_LIST]) #1
print(f" Weights: {weight_str}")
if log['flags']:
print(f" Flags: {log['flags']}") #2
Review a sample log format that exposes recommendations, reasoning, portfolio weights, and flags for human inspection. section 12.7.3
