Some years ago, the kids bought a Meccano Meccanoid from a charity shop. Since then, it has largely been an ornament. Quite an ambitious shape for an ornament standing over 4 foot tall, but it had settled into the role.
This weekend we used AI to help revitalise it. It worked. An object we had become accustomed to having around became something to do together again.
With AI's help, we built Robot club, a little browser control room for a rather large robot. There are controls for its lights, wheels, arms and head, and ways to make routines. You can try it without owning a robot, which considerably reduces the charity-shop luck required to join in.
Screenshot of Robot club in practice mode; no physical robot is connected.
Naturally, all this led to Short Circuit.
I remembered it being brilliant. Johnny 5! A robot coming alive! The sheer possibility of it!
Watching it again, I found the narrative and storytelling awful. My affection survived, but it had considerably less support from the film than I remembered. There is a particular disappointment in returning to something you loved and discovering how much of the good version you have been carrying around yourself.
We had attempted two revivals in one weekend. The robot was considerably more cooperative than my memory of the film.
Still, Johnny 5 is an excellent companion for this sort of project. In the 1986 film, a lightning strike precipitates Number Five's apparent awakening. He wants to learn, insists he is alive and wants to survive. The audience is invited to recognise a someone in what everyone else sees as a something. Whatever I now think of the storytelling, that invitation has stayed with me.
It also makes “we brought the robot to life” a very tempting way to describe our weekend. It captures the pleasure perfectly, while making a complete muddle of the engineering.
Meccanoid was already programmable when it launched. Its original features included recording movements, app controls and motion capture. We were returning to possibilities built into an old toy, with a new way to get at them.
Our browser interface sends commands to the Meccanoid. Its spoken jokes come from voice actions already built into its firmware; there is no conversational AI inside it. The camera feature uses Google's MediaPipe pose tracking in the browser to estimate a person's pose and turn it into joint targets. Without Bluetooth, that movement drives the on-screen 3D preview alone. The preview illustrates commands rather than simulating the robot's mechanics; the complete camera-to-hardware path still needs physical verification.
Earlier work made our project easier: the repository credits existing Meccanoid protocol knowledge and preserved original apps, alongside MediaPipe. Physical measurements mattered too. The project notes record checking the joints and their comfortable ranges. Real objects retain a stubborn interest in where their arms can actually go.
That leaves plenty to enjoy without needing the Meccanoid to have woken up. It also gives us a useful place to notice the temptation behind some present-day suggestions of AI consciousness: something responds engagingly, and we reach for the language of an inner life. A toy with familiar routines and a modern AI system pose very different technical questions. The human shortcut is recognisable in both.
Butlin and colleagues' research on indicators of AI consciousness considers how systems work, drawing on competing scientific theories. Speech and other apparently humanlike behaviour can be imitated, so they are an uncertain guide to whether anything is being experienced. The indicators themselves remain uncertain. There is no settled test, and “it is software” does not resolve the argument either: serious theories disagree about what kinds of physical or computational organisation could support experience.
Johnny 5 has the enormous advantage of being written to convince us. A real system owes us rather more evidence. Affection is allowed to arrive sooner.
Finding the mechanism does not make the enjoyment fraudulent. Knowing which part is a stored routine, which part is a camera model and which part is our interpretation makes the project more interesting to me. Knowing how the trick works can give you another reason to enjoy it. You can start wondering what else you could make it do.
That curiosity needs somewhere to go. With an old connected toy, even finding the necessary software can become a project of its own.
The archived iOS listing records its last update as 9 November 2017. The last Android version I could trace is 4.02 from 2018. In his 2021 Meccanoid repair account, Neil Fraser describes missing firmware-updater downloads and software dependent on files at manufacturer URLs that no longer worked. He preserved and repaired parts of the route back in.
Spin Master later closed Meccano's Calais factory in late 2023, while the brand continued through a global licensing agreement with Addo Play announced in January 2025. Calling that “Meccano going bust” would muddle a factory closure with the fate of the brand and its owner. A continuing brand and a usable old app are separate things.
That history does not explain every year our robot spent as an ornament. It does show a preventable obstacle for whoever next takes an interest. Durable downloads, protocol documentation and a way to replace vanished endpoints would give them a better start. Open-sourcing the releasable software could let them go further.
Pebble shows what that further work can achieve. Fitbit acquired Pebble software and firmware assets in 2016, excluding hardware products; Google acquired Fitbit in 2021, then released most of PebbleOS as open source in January 2025. The Rebble community had kept watches useful with replacement online services since 2018. Google's release gave Pebble founder Eric Migicovsky's separate rePebble/Core Devices venture a substantial starting point for making new watches.
Google's snapshot could not be built as released: proprietary components were missing. Further engineering, services and manufacturing turned that opening into something tangible. In July 2026, Core Devices reported more than 19,000 Pebble Time 2 watches in the field. Community persistence, an owner's decision to release code and a new company's work all mattered.
I would like manufacturers to commit at launch to an open-source handover, triggered by a defined event such as ending updates or withdrawing an essential service. Release the code they can legally share, protocol and build documentation, and the files or interface details needed to replace disappearing dependencies. Preparing that material and arranging rights costs money; proprietary components may still need replacements. Planning while a product has a team and a budget gives the promise a chance of being useful.
A decent retirement plan for a robot should include leaving its paperwork where somebody can find it. Whoever picks it up will still have work to do. At least they can spend more of their effort making something happen.
Our own project is modest hobby engineering with AI assistance. I cannot tell you how long it would have taken without AI. I can tell you that we did it, enjoyed returning to the robot, and have left the code and project notes on GitHub for somebody else to explore.
Practice mode uses an in-memory robot and never asks for Bluetooth. Try the lights, explore the joints or build a routine. The camera preview needs camera permission; connecting a real Meccanoid needs a browser with suitable Web Bluetooth support, with Chrome on a Chromebook recommended by the project.
As I argued in Every Generation Gets Its Purity Test, curiosity is a good enough reason to begin. Understanding can grow as you go, with greater care as the consequences grow.
We taught an old robot some new tricks this weekend. The most welcome change was how interested we were in it again. I would like the next person who feels that spark to find the instructions waiting for them.
Short Circuit may take a little longer.
(Views in this article are my own.)