Inefficiency of text-based interfaces
So, I know everyone is totally excited right now about bots, but let me be the skeptical voice which thinks it’s not going to work out at all. Bots are
claimed to be all about being a ‘natural’ interface, and indeed, if we would have an actual AI the way we imagined them decades ago they would indeed provide an incredible interface. The fact of the matter however is that they are not — as this article correctly points out — meaning that what we’re left with is an extremely unnatural and inefficient interface. Even looking at the current generation of digital assistants we see them — for the most part — moving away from voice control to proactive (the Google Now homescreen) and context based information (Google Now on Tap).
That’s not to say there is no place for voice based interfaces, certain actions happen at moments when precise finger control isn’t an option. There is a
wide variety of those situations, from turning off the lights whilst lying in bed to playing a certain song when running. But situations where it makes
sense to order a pizza through voice control are far rarer. Not impossible to think off — e.g. when sitting in the car — but in the general case using an
optimized interface for this process including visual feedback is far more efficient.
Just compare how long it would take to order a pizza through a bot compared to an equally optimized dedicated app/website:
1. I would like to order a pizza
2. We have 32 different types of pizza, which would you like to order?
3. Which ones do you have?
4. Vertical list of 32 different types of pizza’s
5. A pizza xxx
6. Would you like to change any of the toppings or the sauce?
7. Yes, I would like extra cheese
8. This was not understood (we’re talking about a first time customer, not knowing the quirks of the AI). Would you like to remove a topping?
9. No
10. Would you like to add a topping?
11. Yes, I would like to add cheese
12. Do you want to add any other toppings?
13. No.
14. Do you want to change the sauce?
15. No.
16. The current order total is ten dollars and fifty two cents consisting of one pizza xxx with extra cheese. Would you like to add another pizza?
17. No.
18. Would you like to add a drink?
19. Just finish the order already.
20. Your answer was not understood, would you like to add a drink?
21. No.
22. Would you like to add any snacks?
23. No
24. Where would you like this order delivered?
25. xxx
26. Did I correctly understand you want to have this order delivered at xyx?
27. No, xxx
28. Which of the following addresses is your address: 1. xyx, 2. xxy, 3. yxx
29. 2.
30. [etc.]
vs for example the current Dominos flow:
1. User opens app
2. Enters address
3. Sees a list of address interpretations (for my home address that’s currently a list of like 20 options, as the Dominos app has a somewhat weird spelling of my street name)
3. Taps on valid interpretation of the address
4. Sees list of pizza’s, using the screen real estate optimally
5. Adds pizza xxx
6. Taps the customize link in the cart under the pizza xxx line
7. Taps on the cheese
8. Clicks through the snack and drink screen (it’s the same button each time)
9. [etc.]
A dedicated app/website is like the best a bot can offer, but with the freedom to design smarter more intuitive and efficient interfaces. A bot can’t show the order menu in two neat columns with the description of the pizza in a different lighter color. As long as we don’t build true AI’s there is not a single flow which can’t be more efficient in a dedicated website/app than it is as a bot. Why? For the simple reason that a website/app has all the freedom a bot has *and a whole lot more*. Yes, writing ‘I want to order a pizza xxx with extra cheese’ is quite efficient, but allowing that input as a starting point in normal apps/websites is just as possible. The difference is that a normal app/website can however show a button to show the cart on demand rather than requiring the user to learn a special language for each and every bot.
Now, lets be very clear here: I am a huge fan of using applications that register their respective Google Now hooks and I think it’s great when a food delivery app hooks in to the platform digital assistant to allow for existing customers to repeat their last few orders or start an application (e.g. ‘Okay Google, I want to order a pizza xxx with added cheese and a Fanta’ which would ask you which app/website you want to use to finish this command and then prefill your cart with those items). A case definitely is to be made that voice control will play an increasingly important part in human computer interfacing, but bots will always — till we get true AI’s — take a secondary place to custom made interfaces. The only advantage bots could have is that they could’ve been cheaper to make (as they don’t require graphic and UI designers), but reality teaches us that actually NLP is super hard to do right.