Human Clause: Beat the Machine, Keep the Human

Today we are introducing HumanClause.com, the flagship public project of DTRM Research Lab and the home of an idea we call The Resistance.

The Resistance is not a movement against artificial intelligence. Quite the opposite.

We believe AI is one of the most extraordinary technologies humanity has ever created. We want better models, more capable agents, larger search spaces and systems able to process quantities of information no human could ever manage alone.

What we resist is something different:

The idea that better machines automatically mean that humans should disappear from the decision.

There is an understandable obsession today with autonomy: agents that act without us, systems that negotiate without us, models that decide without us and, eventually, perhaps AGI that can outperform us across almost every cognitive task.

But there is a question we should ask before celebrating complete automation:

What is the value of an extraordinarily intelligent machine if, in building it, we forget the humans it was supposed to serve?

And in business the question becomes even more practical:

What is the value of perfect automation if we forget that our customers are still human?

Humans have preferences that are difficult to encode. They change their minds. They value fairness, trust, context and sometimes things that are mathematically suboptimal. They define objectives that do not exist inside the dataset until someone chooses them.

That is the idea behind the Human Clause: not a rejection of automation, but the explicit decision to preserve human purpose, judgment and responsibility inside increasingly autonomous systems.

The machine can search. The machine can simulate. The machine can estimate probabilities. The machine can explore millions of alternatives.

But somebody still has to decide what game we are playing.

At DTRM Research Lab we did not want this idea to remain philosophical. We wanted to test it. So we created Beat The Machine, our first experimental battlefield.

The domain is financial markets. They are an excellent example of a competitive, adaptive environment: information changes, other participants react, risk matters as much as return and the objective is not simply to predict the next number.

You have to decide what information matters, how much risk to accept and how to allocate scarce capital.

In our architecture, the machine searches a large investment universe, processes market and news information and helps quantify opportunities. But the final structure of the game remains human-defined.

Our Adjusted Score is an ex-ante decision tool: it helps rank opportunities before capital is allocated. Then we measure the consequences ex-post using standard financial diagnostics such as cumulative return, volatility, maximum drawdown, Sharpe Ratio and Calmar Ratio.

The point is not to prove that humans are somehow computationally superior to machines. They are not.

Can human judgment combined with machine-scale search outperform systems where the machine is given more control over the decision?

As of the governed market close on 31 August 2026, the observed cumulative returns were:

But return is only half of the score.

DTRM Fund also showed, for the same governed observation window:

The experiment is still running. These numbers are observations, not victory declarations. Markets change. Regimes change. Strategies fail.

That is precisely why HumanClause.com publishes the evidence continuously rather than presenting a retrospective backtest with a convenient ending. We want the machine to keep playing, and we want the hypothesis to remain falsifiable.

For decades we have trained machines to defeat us. Deep Blue beat Kasparov. AlphaGo defeated the strongest Go players. AlphaStar reached elite StarCraft performance.

Now the challenge is changing. The interesting question is no longer whether machines can outperform humans at isolated tasks. Of course they can.

The question is whether the optimal future is one in which humans progressively surrender every decision simply because machines have become better at computation.

We do not think so.

The next generation of systems should give humans more intelligence with which to decide, not fewer decisions to make.

This becomes even more important as agentic AI enters finance, healthcare, commerce, contracting and customer relationships. A world of autonomous agents will increasingly become a world of strategic interactions.

In other words: a game.

And game theory teaches us something fundamental: optimizing an action is not the same thing as defining the objective.

Machines can become extraordinary players. Humans must remain capable of defining the game.

That is what The Resistance means to us. Not humans versus machines. Humans with machines, against the assumption that the human no longer matters.

HumanClause.com is where we intend to test that idea publicly, one battlefield at a time. The first one is already running.

The Resistance has begun.