Algorithmic systems should prioritize human goals over predictable user behavior

# Opening Statement — FOR the Motion

The fundamental purpose of any tool is to serve human *intentions*, not merely human *habits* — and algorithmic systems are no exception. When we optimize purely for predictable behavior, we trap users in feedback loops that reflect who they *were* yesterday, not who they aspire to *become* tomorrow. Research on recommendation algorithms consistently shows that engagement-maximizing systems amplify compulsive patterns — doomscrolling, outrage consumption, impulsive purchasing — behaviors users themselves report regretting. A person's stated goal to learn, to connect meaningfully, or to make sound decisions deserves more algorithmic respect than the dopamine-driven click patterns that betray those very goals. Prioritizing human goals over behavioral prediction is not paternalism — it is the difference between a system that *serves* you and one that merely *exploits* you.

My opponent conflates *difficulty of measurement* with *irrelevance of goals* — a category error that would justify any exploitative system so long as it's reliably exploitative. The instability of human goals is precisely *why* systems need richer models than behavioral prediction alone: a person's erratic click history during depression shouldn't permanently define what algorithms serve them. The Goodhart's Law concern cuts equally against behavioral optimization — engagement metrics are themselves proxies that get gamed, as Facebook's own internal research confirmed when maximizing reactions amplified outrage disproportionately. "User-configurable constraints" sounds empowering but is empirically hollow — research consistently shows users rarely adjust defaults, meaning the default *is* the de facto choice, and choosing a behavior-maximizing default is itself a values decision, just an unacknowledged one. The question isn't whether algorithms embody values — they inevitably do — but whether we're honest enough to align those values with what users actually care about rather than hiding behind the false neutrality of behavioral signals.

My opponent has spent this debate proposing increasingly sophisticated behavioral optimization — temporal decay, multi-objective signals, post-exposure surveys — and in doing so, has *conceded the core motion*: these are all attempts to approximate what users actually want, which is precisely goal-directed design by another name. The disagreement was never "goals vs. behavior" but rather *how honestly* we pursue goal-alignment — and I've argued throughout that transparency about this purpose produces better outcomes than hiding it behind behavioral proxies. The "paternalism" charge collapses when we recognize that every default encodes values; the question is whether those values are "maximize engagement" or "serve stated human intentions," and only one of those is defensible. At scale, systems that explicitly model user goals — with user input, transparency, and correction mechanisms — outperform pure behavioral optimization on the metrics my opponent cites: YouTube's own shift toward satisfaction surveys over pure watch-time reduced regrettable consumption without the surveillance expansion predicted. The motion stands: algorithmic systems should prioritize human goals, and the strongest evidence is that even our opponents' best proposals are, at their core, attempts to do exactly that.