Government surveillance is hardly new. What artificial intelligence threatens to change is not government’s desire to watch, but its ability to understand everything it sees. That distinction could fundamentally alter the balance between the citizen and the state.
For most of history, surveillance was constrained by human limitations. Governments could collect records, intercept communications, monitor suspects and operate cameras, but somebody ultimately had to examine the information. There were only so many investigators, analysts and hours in a day. Mass surveillance therefore produced its own obstacle: masses of data.
AI potentially eliminates that obstacle.
Imagine government possessing access to millions of cameras, license-plate readers, public records, financial transactions, social-media activity, location histories and other digital information. Without AI, connecting those dots requires enormous manpower. With sufficiently capable AI, the state could potentially search, correlate and analyze those records continuously.
The ethical question is therefore larger than whether government should use AI. It is whether Americans are prepared for a government possessing something approaching automated omniscience.
There are legitimate uses. AI could help locate missing children, identify suspected terrorists, detect sophisticated financial crimes, uncover human-trafficking networks or find connections among criminal suspects that investigators might otherwise miss. Refusing to use powerful technology against genuine threats would make little sense.
But constitutional government was never constructed around the proposition that government should exercise every useful power available to it.
The Fourth Amendment exists precisely because public safety is not the only American value. The founders understood that government invariably possesses compelling explanations for wanting more information, more authority and fewer restrictions. The constitutional answer was to establish boundaries between legitimate law enforcement and the private lives of citizens.
AI makes those boundaries more important, not less.
Consider the difference between investigating a suspect and investigating everyone until the government discovers a suspect. Traditional American law enforcement generally begins with some reason for suspicion and then seeks appropriate legal authority to investigate further. AI creates the technological possibility of reversing that process: analyze the population continuously, identify supposedly suspicious patterns, and then decide whom to investigate.
That is an extraordinarily consequential change.
A free society should be uncomfortable with algorithms examining millions of innocent citizens searching for statistical indications that somebody might become interesting to the government.
The problem becomes even more serious when surveillance extends beyond criminal activity. AI could theoretically classify political associations, map social networks, identify participants at demonstrations, analyze religious gatherings, monitor journalists’ contacts or determine which citizens frequently associate with particular organizations.
A government need not arrest political opponents for surveillance to chill political freedom. Citizens behave differently when they believe authorities are permanently watching them.
Conservatives should understand this danger particularly well. Distrust of concentrated government power is not paranoia; it is embedded in the architecture of the American constitutional system. The question should never be whether today’s officeholders can be trusted with extraordinary surveillance capabilities. The question is whether we would willingly give those same capabilities to our worst political opponents.
If the answer changes depending upon who occupies the White House, the power is probably too broad.
This does not require rejecting AI-assisted law enforcement. It requires separating targeted technological assistance from indiscriminate algorithmic surveillance.
Government should be able to use AI to analyze evidence lawfully obtained during legitimate investigations. If investigators possess a warrant covering thousands of records, AI can help examine those records. If authorities are searching for a kidnapped child, facial-recognition technology may be enormously valuable.
But allowing government to continuously analyze the behavior of innocent Americans simply because technology makes it possible crosses a fundamentally different line.
Congress and state legislatures should establish those lines before bureaucratic practice establishes them by default.
Warrants should remain meaningful in the AI era. Agencies should be prohibited from circumventing constitutional protections by purchasing information from private data brokers that they would otherwise require legal process to obtain. Citizens should have meaningful remedies when surveillance powers are abused. AI systems used to generate investigative leads should also be subject to rigorous standards, particularly when algorithmic conclusions contribute to searches, arrests or prosecutions.
Transparency matters as well. Secret algorithms should not become invisible witnesses against defendants. If an AI system materially contributes to the government’s accusation that an American committed a crime, courts must be capable of examining how that conclusion was reached.
There is also a broader temptation that must be resisted: predictive policing becoming predictive governance.
Once government possesses systems capable of identifying patterns associated with crime, officials may inevitably ask those systems to identify patterns associated with extremism, misinformation, civil unrest or other increasingly subjective categories. At that point, technology intended to identify criminal conduct can begin identifying undesirable thought.
That road should alarm Americans regardless of political affiliation.
Artificial intelligence could make government extraordinarily effective at protecting citizens. It could also make government extraordinarily effective at monitoring them. Those two possibilities are not mutually exclusive.
The conservative position should therefore be neither technological panic nor technological surrender. Use AI where constitutional government permits it. Demand warrants where warrants are required. Preserve due process. Require accountability. Prohibit mass surveillance untethered from legitimate investigations.
America should have intelligent law enforcement.
It should never have an all-seeing state.