Amazon ditches 'just walk out' checkouts at its grocery stores
Amazon is phasing out its “Just Walk Out” cashierless checkout in U.S. grocery stores after it emerged that the supposedly AI-driven system depended heavily on more than 1,000 human reviewers in India, making it costly and hard to scale. Commenters see the retreat as evidence of current computer vision and “AI” limits in messy real‑world retail, and compare it with more pragmatic models like scan‑as‑you‑shop, RFID‑based self‑checkout, and traditional staffed or self‑service lanes. The change also reignites debates about privacy, shoplifting, customer experience, and how often automation really just means outsourcing low‑paid human labor.
Human labor behind “Just Walk Out” and the AI hype gap
- Many commenters latch onto reports that >1,000 workers in India were watching/labeling video; some interpret this as effectively off‑site cashiers, others as training/validation for ML models.
- A cited figure of “700 of 1,000 sales needing human review” is seen as evidence it never technically or economically scaled; Amazon is said to dispute that number but offers no alternative.
- This becomes a broader example of “Mechanical Turk” / “wizard‑of‑oz” AI: impressive demos fronting large amounts of low‑paid remote labor (parallels drawn to self‑driving oversight, delivery robots, LLM data labeling, receipt scanning, “AI” chatbots).
- Several see this as evidence current “AI” is overhyped, especially where the company—not the user—must bear the cost of errors.
Technical and economic constraints
- Computer vision for dense, cluttered, occluded grocery environments is described as extremely hard and expensive: cameras, compute, annotation, constant retraining as packaging changes, and tricky categories like produce/by‑weight items.
- Some with CV/retail experience claim there’s “no financially feasible way” to run full JWO at grocery margins today.
- RFID is widely praised (Uniqlo, Decathlon, libraries), but commenters stress tag cost vs grocery margins, form‑factor diversity, reliability issues, and supply‑chain coordination.
Alternatives: scan‑as‑you‑shop and self‑checkout
- Scan‑as‑you‑shop (handheld scanners or phone apps) is common in parts of Europe and some US chains; fans like bag‑as‑you‑go and skipping lines, critics dislike cognitive load, forgotten scans, and random recheck audits.
- Self‑checkout experiences vary dramatically: some report fast, frictionless lanes; others face constant weight‑sensor errors, age checks, and long waits for attendants.
- Debate over whether humans are still faster and whether stores intentionally understaff manned lanes to push self‑checkout.
- Loyalty programs and “digital coupons” are seen as price discrimination and data‑collection tools, sometimes required to use advanced checkout features.
Privacy, surveillance, and pricing
- High‑res in‑store cameras are viewed by some as creepier than existing POS data (e.g., potential to read phone screens, tailor prices in real time).
- Others argue stores already have “perfect” purchase histories via barcodes and loyalty cards, citing past retail analytics cases.
Customer experience, labor, and norms
- Some loved Amazon Go’s speed and lack of queues and are disappointed; others are “glad it’s gone” due to hidden labor and surveillance.
- Strong theme that self‑checkout and schemes like JWO mainly cut labor costs and shift work and liability onto customers, without lowering prices.
- Broader lament that grocery and many other industries have traded service quality for cost‑cutting and profit maximization.