How Artificial Intelligence Could Have Prevented New York’s Massive Cannabis Screwup
New York’s rollout of recreational cannabis was easily not only the worst cannabis rollout by any state ever, it was one of the worst examples of competent governance I’ve seen in my thirty years in and around politics. Using artificial intelligence could have prevented a lot of what went wrong.
Albany legalized recreational cannabis in March 2021. The bill set out a complicated structure for how legalization would work. The Office of Cannabis Management (OCM) was created as part of the bill. OCM then took a very long time putting everything together – promulgating rules 2.5 years after the bill was signed, covering eight different classes of cannabis businesses: plant nurseries, cultivators, processors, cooperatives, distributors, dispensaries, delivery services and microbusinesses.
When the rules finally were promulgated, they didn’t make sense. In fact, the Albany County Supreme Court explicitly struck down the marketing rules, saying: “There is nothing in the record to establish precisely how OCM developed the regulations, which staff members participated in the process or how they addressed the litany of issues that were raised...”
The multi-year delay created a weird dynamic where weed was legal but not legally available. New Yorkers knew they were allowed to possess cannabis and when they saw stores pop up everywhere selling cannabis, it’s hard to expect them to know which were licensed (a handful) and which were not (everything else). So the market does what it always does – fill a void, this time with thousands of illegal weed shops. Estimates are up to 8,000 in New York City alone.
The City and the State both refused to crack down on the illegal shops, each insisting the other was responsible for enforcement. As a result, the people who went through the licensing process were totally screwed because getting the license was extremely expensive and then all of the costs of running their businesses are even higher – every product has to be inspected, they pay taxes, comply with labor laws, etc… They can’t compete with the illegal shops who are doing exactly none of that. There are still only 140 licensed dispensaries (to NYC Mayor Eric Adams’ great credit, he has now cracked down aggressively on the illegal shops, closing well over 800 in the past twelve weeks).
In March 2024, after months of outrage from New Yorkers and applicants, Governor Hochul ordered a top to bottom review of the program to figure out how the program had fallen apart – and why so few qualified applicants had received licenses. The report found, among other problems, there was no centralization of the application process, an inability of staff to adapt to changing policy priorities, a lack of communication with applicants, poor customer service offerings, and disparate and uncoordinated IT systems.
Had AI instead been implemented as a tool to better inform regulatory policy from the start (aiding the drafting, putting the rules out for public comment, digesting the comments), operationalize the licensing process and award the licenses rather than working through a dysfunctional bureaucracy, here’s what likely would have happened instead:
The rules could have been put together in a fraction of the time it actually took — months, instead of two and a half years. And the licensing process could have started much sooner this way too. AI could have enabled the centralization of licensing operations, as well as given the regulatory rulemaking process a framework to start with, as well as provided better customer service, collected critical data to inform the licensing KPIs and better utilized IT systems – all problems cited in the report. As it says in the report, “the work of setting regulatory policy and designing programs is intertwined with operations.”
If you apply AI tools to organize and execute the basic operational tasks that OCM clearly failed to do from the start, then the work of setting regulatory policy and designing licensing programs should also run more efficiently and more quickly. For example:
AI could have reviewed the applications and issued licenses. The report showed that OCM failed to follow ITS-recommended solutions’ and instead spent their time essentially creating spreadsheets and CRMs in silos, rendering them useless. As we know, AI is about collecting as much information as we’ll allow.
AI could have weighed all of the various factors that mattered to the legislature to determine the best applicants based on equity (people from communities disproportionately impacted by the enforcement of cannabis prohibition, minority and women owned businesses, distressed farmers and service disabled veterans), ability to run a successful business, location and any other factors.
By implementing AI functions to drive and audit the application and rulemaking process with the necessary inputs for compliance, the process would have become far more immune to lobbying, political pressure, media pressure or anything that subverts the process.1
AI would have allowed OCM to set a clear timeline like 100 licenses per month and stuck to it. The report cited the massive bottleneck of applicants awaiting determinations. With AI, you wouldn’t have the bottleneck. OCM had an entire application review team. Just scheduling their meetings likely took as long as AI running all the criteria and data and reaching a determination. When the process includes nine different OCM staff members across four different business units prior to an enforcement review, background checks, management review, board approvals and compliance checks, of course it’s going to be a disaster.
The report said “by creating new structures rather than implementing approaches based on best practices, the leadership wasted valuable time and resources.” AI could have immediately identified and incorporated best practices.
OCM struggled to fill licensing jobs. If you were performing most of the functions with automation, hiring is no longer an issue.
The report cites the complexity and obscurity, the lack of transparency, regarding the entire licensing process. All of the AI’s work can be publicly available, shown and auditable.
All of that would have quickly created a legal market, preventing the rapid rise of the illegal market. It also would have not materially disadvantaged the license holders and applicants. In fact, bringing in the State’s Office of Information Technology Services sooner to set up a plan to use AI would make a lot of sense, both for OCM but also in general for rulemaking, issuing licenses, judging RFPs and more.
As a result, we would now not have thousands of illegal shops, a totally discredited cannabis program, license holders economically screwed over, a massive backlog of license applications, and perhaps most importantly, because the illegal shops sold indiscriminately to teenagers, a generation of addicted teens who are now behind in school and suffering mental health problems. A lot of real world harm occurred because of bureaucratic malfeasance.
I know we have low expectations for government and a high expectation (even tolerance) for governmental delays, mishaps and mistakes. We shouldn’t. Bad government is not a victimless crime. And it’s often entirely preventable. Technology is not the only way to make government work better but AI is going to offer opportunities to perform certain functions far better than we’re doing them today. Unions and others will oppose any changes but the government doesn’t exist for the good of the special interests that feed off of it. It exists for the good of the people. And in this case — and in many more — the people would have been a lot better off with a different approach.
There is a NY statute saying that AI cannot be used to make final decisions so while the definition of final is probably very debatable, the statute would either need to be changed or a step at the very end would have to be added in to make the AI decisions not final in a way that doesn’t risk all of the problems above.