There is no single meaningful score for “AI versus human traders.” Execution, research, portfolio decisions and risk oversight are different tasks. A useful comparison names the task, system, data and outcome being measured.

This article does not present a proprietary head-to-head performance study. It explains how to assess evidence without treating automation's speed or an AI label as proof of profitability.

Algorithmic Trading Is Not Synonymous With AI

A fixed order-slicing rule is automated, but need not use machine learning. A model that summarises a report may use AI without sending any order. Estimates of algorithmic trading activity cannot be relabelled as the share of trading controlled by AI.

Model designs also differ. Some are fixed after training; others update under a defined process. Neither automatically handles new market conditions well. People and software can both make mistakes under unfamiliar conditions.

What a Fair Comparison Would Require

Specify the instruments, period, information available at each decision, position constraints and all costs. Include losing periods and unsuccessful model versions. Compare drawdown and exposure alongside returns, and keep simulated results separate from live fills.

The same historical data should not be used both to choose the winning model and to claim an independent evaluation. Record how many alternatives were tried and test later data without selecting a new winner after the fact.

A successful institution's overall return does not establish which component came from AI, nor what a retail subscriber can achieve with a different product and infrastructure.

Assign Authority According to the Task

A research assistant can extract candidate facts, with a person checking them against the original filing or release. An execution system needs additional controls for quantity, stale prices, duplicate orders and failed connections. A generated explanation is not proof that the cited fact or proposed trade is correct.

If software can send orders, define the maximum authority, monitoring and shutdown procedure. Human approval can add a check but is not an assurance of correct judgement.

What This Means for a Retail Product

Ask the vendor which component uses AI, which decisions remain fixed rules, and whether the advertised record belongs to the current version. Inspect permissions and custody before connecting an account.

The CFTC warns against AI trading-profit promises. Neither “humans always win during crises” nor “AI makes retail traders obsolete” follows from the available description of a tool. Assess a bounded task and its documented results rather than predicting a universal winner.

Risk Disclaimer: Trading involves substantial risk of loss. Past performance is not indicative of future results. See our full risk disclaimer.