How Stack-Up proves that automation simply doesn't stack up

Stack-Up (1985)

Stack-Up (1985)

It’s funny how your job description gets thrown out from under you. Years ago I was employed as a Customer Success Manager, but really it was a nice and fancy way of saying I was a Project Manager who took plenty of guff from clients. But by the end of my stint there, I had become a Chatbot Programmer, Developer, Troubleshooter, Optimiser, Humaniser, AI Tester and indeed AI Detester in all but name.

Yes, if you’ve ever gone on to a company’s website with the aim of lodging a complaint, only to find yourself confronted by a “hip”, “cool” chatbot who says “Great! I’m happy to hear that” upon being told that your grandmother has died, then I need to confess. Yes, the person behind that bot was me, and I’m truly sorry for all that.

Now that I’ve left that job, it’s probably safe for me to tell you that chatbots simply don’t work. Automation in general doesn’t work. You knew that yourself already, of course - how many positive interactions have you ever had with bots? Zero, right?

---

It’s not just user-error causing bots to fail, though. I can also tell you that, front-end, back-end, ass-end, whatever way you look at bots, they just don’t function correctly. They are always, always getting stuck, or going down altogether until somebody intervenes and brings it back.

That’s no good for business owners, is it? Imagine your human agent, as we have to call them now, imagine they just zonked out for an hour and left everyone on read? You wouldn’t be too happy if your staff elected to have a nice sleep for a couple of hours, would you?

Some trendy sales executive may attempt to counter this by pointing out that bots have a much lower labour cost than FTEs. This is your typical flowery, salesy way of saying that your company won’t need to pay minimum wage to a robot. But unless you don’t want dumb little PilchardBot representing your brand, you’ll have to pay fools like me good money to set it up right and give it a bit of personality, maybe even a cute little avatar.

Give me a few days and I’ll build you a chatbot about as complex as the controls in an elevator; click the menu option and, assuming the bot stays upright long enough, it could even give you an extremely signposted response. Try and implement a bot that can interpret free text questions, though, and you’ve got to use NLP & NLU, and that’s where things get exceptionally fruity.

Yes, I know it might all sound cool, AI and automation and neural networks and machine learning. But in reality, in the back-end, chatbot developers like myself and half of Asia are just hammering key phrases and keywords into a massive list, in desperate hope that your ignorant bot might understand some words in the future.

This is actually quite appropriate, because the end user will be doing some hammering of their own too, angrily raining blows down on their keyboards, typing out the words “agent” or “human” in a futile attempt to gain some human recognition. Nowadays I’ve even started to see businesses advertising the fact that their customer service teams are entirely human. Businesses will never get it right, will they?

The things you have to do to become a subject matter expert, eh? AI is a lie, or at least its usefulness has been blown way out of proportion, and it’s all been propagated and perpetuated in order to keep CEOs and shareholders spending money to keep ahead of their competitors until the next trend comes along.

Give me a real-life, physical robot any day of the week. Old R.O.B. for instance, or the Robotic Operating Buddy to give him his full name - now that’s what you call a classic design. There’ll be no chatting with this bot, oh no. In fact, he doesn’t have many functions at all.

I respect this little robot for his role in gaming history but I have to say, what his scope document promised and what he actually delivered were miles apart. R.O.B. bowled a strike through the US gaming market, and promised us a brand new gaming experience, a cool little robo-buddy who could play the game with you.

In reality, this little hunk of plastic is smaller, lighter, and more fragile than you think, and he’s not going to be playing NES games like Contra and Zelda alongside you. No, try to plug him into those games and you’ll find him a most untrained bot, and you’re crazy if you think he’s going to start responding – he’ll be as gleefully ignorant to your requests as the rubbish old chatbots I used to throw up onto “the cloud”.

The best R.O.B. will give you is rock-bottom performance in a whopping two NES games: Gyromite and Stack-Up. Gyromite at least gives R.O.B. a chance to get in on the action, quite literally opening doors for you. The game is mostly garbage anyway, but at least it’s slightly interesting.

R.O.B.’s other compatible game, Stack-Up, requires a teeny bit more bot development before you can Go Live with it. The game comes in a specially large NES box, which contains all manner of bulwarks and buttresses that you attach onto R.O.B. As you probably know, the more moving parts you add to a package, the greater the chance of something going wildly wrong.

Actually, keep that one in your back pocket because it’s a great one to fob customers off with whenever shit doesn’t work and they’re demanding to know why - just remind them that there’s loads of moving parts. Eager not to look stupid and untechnical, the client will suddenly accept it and you’ll buy yourself a few more days.

Anyway, back to the Stack, to see how it all stacks up. Included in your $1,000 box is a bunch of coloured “blocks” which you need to place into R.O.B.’s holders. Switch on the Stack-Up game on your NES, try not to roll your eyes at the fact that they never even changed the title screen from its Japanese moniker of Robot Block, and go into the main game.

Ah, sorry, I said game, but it’s really just a list of chores. The game wants you to select inputs on the screen that will make R.O.B. dance and sing a tune in the real world. Well, actually, he’ll execute the orders onscreen, and move one stack of blocks from one of his holders to another… and that’s it.

I’m sorry, what?! Has there been some sort of mistake? This is the sort of thing you do in testing, or don’t do in testing as is more often the case. Did they accidentally ship the training program instead of a fun, playable game?

Stack-Up seems to operate on the honour system – the Gyromite game at least gave R.O.B. a finicky, Doc Brown-style way to access your NES controller, allowing him to press buttons and interact with the game. In Stack-Up, there is no game at all, just a laundry list of processes. You decide when to continue, even if R.O.B. looks completely wrong. It’s a bit like correcting your own homework.

Worse than that, Stack-Up succinctly brings home everything that’s wrong with bots, this supposed automation revolution. Yes, I know it’s from 1985, but nothing has changed: R.O.B. was an overpriced, overwrought, overthought solution that, at worst, still manages to literally fall over, and at best, does almost nothing constructive at all – and it takes all day to do it. Take my advice on this one please, and terminate all robots and automatons you see with extreme prejudice.

21 March 2025