A 1998 toy's own code shows scripted, not open-ended, learning

The design

A furred robotic toy that appeared to learn English became one of the best-selling objects of 1998, twenty years before any companion app. Its own source code survives as a scanned patent-application exhibit, preserved on the Internet Archive, titled 'INTERACTIVE TOY (FURBY.ASM),' naming 'INVENTOR: Dave Hampton,' filed through the law firm Fitch, Even, Tabin & Flannery, and copyrighted '1996, 1997, 1998 by Sounds Amazing!,' Hampton's own company. Its modification log records 'final testing for shipment of code on 8/2/98.' Tiger Electronics' own launch website, live by December 1998, calls the toy 'the interactive pet' and offers a 'Furbish Dictionary,' the company's own name for its invented language.

What the evidence says

The source code itself, not marketing copy, is the more precise record of the 'language learning' claim. It implements sensor-triggered counters, tied to being tilted, fed, petted, or spoken to, each tracking a 'learned word,' and a comment in the code states that regardless of the toy's own simulated age, '25% of time' it draws speech from an earlier vocabulary tier so as to keep producing invented 'Furbish' even at older ages, a scripted probability shift, not open-ended learning. Tiger's own website confirms the company branded this invented vocabulary 'Furbish' and marketed the toy as interactive; it does not describe the underlying mechanism, which only the code itself shows.

What it asks of people

The design asks a child or owner to physically work through the toy's own named sensors, tilting it, triggering its feed sensor, touching its light and tilt inputs, talking to it, so as to 'teach' behavior that the surviving code shows was already scheduled by probability tables tied to elapsed interaction, not shaped by any specific lesson. It manufactures a sense of a developing relationship through scripted randomness rather than genuine responsiveness to what a child actually said.

Privacy and safeguards

Nothing in the surviving code or the 1998 website describes any connection to a company server or the internet; the code does reference an infrared 'sleep command,' evidence of short-range signaling between nearby units, but the 'learned word' counters it tracks are stored on the device itself and, this record shows, never transmitted to Tiger Electronics or anywhere beyond that. That is a meaningful contrast with networked companion toys and apps built later, which typically send comparable interaction data to a server, something this 1998 design demonstrably did not do.

• What does the surviving source code establish about Furby's 'learning' that marketing language alone would not show?
• How does a scripted, on-device probability table differ from what a modern companion app calls memory or personalization?
• Does the absence of any network connection in 1998 make this design safer, or simply differently limited, than later connected toys?

Furby's own code, filed with a patent attorney and preserved decades later, is a more reliable guide to what the toy did than any story told about it since.