Alexa+ is starting to feel a bit like the future, but shorter responses would be nice

When Amazon first announced the Echo smart speaker and Alexa, it felt as though the future that Star Trek had promised us was finally upon us. Here was a computer we could interact with naturally, faster and more convenient than apps or traditional interfaces.

 Unsurprisingly, Amazon sold a bucket load of Echo devices, and soon expanded the range with devices to fit in everywhere. Only, it turned out that perhaps the future wasn’t really here.

Alexa speak

As noted in my column from a few weeks ago, I largely use physical controls over voice commands alongside automated routines: it’s faster to turn a light on with a button, or to have my alarm turn off and blinds open when the office door unlocks, than it is to use a voice command for either job.

A lot of that was down to how Alexa (and other voice assistants) expected commands to be phrased. While Alexa is still the best of the bunch, its required terminology gave birth to the phrase, “Alexa speak”. 

It’s that slightly unnatural way that you must phrase a command, such as, “Alexa, set the living room radiator temperature to 20°C.” That phrase doesn’t seem so bad, but it’s fraught with potential problems.

Get the order slightly wrong, and Alexa might not work; name the device you want to control incorrectly, and the command doesn’t work; or just pause while you try and think of the right words to use, and the command doesn’t work. 

Outside of voice control, Alexa is good for basic requests or for answering simple questions, but it often can’t understand more complicated requests, can’t take actions on your behalf, and you still must phrase things as though you’re talking to a computer. 

Natural conversations and context

Alexa+ promises to change that and, from what I’ve seen of it, delivers the end of Alexa speak, switching to natural language, so you can ask a question or issue a command as though you were talking to a real person. And Alexa+ remembers context and allows itself to be corrected.

At the Alexa+ UK launch event, I saw a demo where Alexa+ gave the latest Arsenal result; it knew the presenter was a fan, so it recounted the score with a positive tone.

Next, the presenter asked Alexa+ to tell someone else the Chelsea score. Alexa began retelling the loss with excitement, since the presenter hadn’t mentioned that the other person was a Chelsea fan.

A quick interruption to say that the other person was a Chelsea fan had Alexa+ start again, but with a neutral voice. There was no need to rephrase the entire question with something like, “Alexa, my friend is a Chelsea fan, tell him the latest score” or something similar.

Alexa+ understood that the change applied to the current request and adjusted its response accordingly. In addition, Alexa+ would then remember who’s a Chelsea fan for future requests.

Alexa+ is also agentic, which means it can take actions on your behalf. In the demo, Alexa+ could book a table at a restaurant using OpenTable, based on a few simple bits of information, all spoken naturally, and where the order of information was unimportant (the name of the restaurant, how many people the table was for, the date and when there was at least two hours free in the diary).

That kind of interaction seems better, easier and faster than having to search for the restaurant and do the job manually.

Not perfect, but certainly better

As part of Alexa+ launching in the UK, Amazon has fine-tuned the system to understand a wide range of British accents and to understand the way we speak. This information is also used in how Alexa+ responds. Is it perfect? No.

Particularly with responses about football, Alexa+ seemed to like using the word ‘mate’ a lot, which feels a bit false and over-friendly. I’m not sure I want Alexa+ to be my friend; I just want it to do what I want, when I want, with clear replies. I’ll have to see, once I have access to Alexa+ soon, if I can tone down its replies.

Then, there was a demonstration where Alexa+ was asked when the next match was for a football club. The result was right, but when asked to add the game to the diary, Alexa+ added it in for one hour from the start time.

Surely, if Alexa+ is so smart and understands context, it should know that a football match is 90 minutes, plus 15-minutes of halftime, plus extra time. That’s a minimum of one hour and 45 minutes, but two hours would be a safer bet.

I was told that because there was lots of background noise, Alexa+ might be struggling to work out what was said. It did get the match details right, and it did understand to add a calendar appointment, so we’ll have to see if Alexa+ can be smarter than this in real life.

Likewise, context can be hard to understand. When asked, on a Fire TV device, who won the Best Actress Oscar, Alexa+ correctly replied that it was Jessie Buckley for Hamnet. Next, what asked, “Can we watch it?”, I thought that would mean that Alexa+ would find a clip of the Oscar ceremony and show that. Instead, Alexa+ started to stream Hamnet from Prime Video (currently £15.99 to rent or £19.99 to buy).

Either response is correct, but does Alexa+ have a bias towards trying to sell you things, or is it just picking one option because that’s what it thinks is the right one? It’s hard to tell, as even humans can struggle with context and ambiguity.

Too many clichés?

Alexa+ also seemed to like its clichés and longer responses. When asked to recommend some coffee machines (all on Amazon, of course), it described one’s price as something that “won’t break the bank”. 

Training any AI means pulling data in from lots of resources, but the issue is that lots of people use clichés, and there’s a horrible chance that any system will reinforce that behaviour.

When I used to work on a print title, our sub editor banned all clichés and had a list of banned phrases, opting for brevity, to deliver clarity. One example was ‘value for money’, as what else would something be value for? Value for cheese? Value for magic beans? 

Likewise, there’s no ‘make use of’. It’s just use. You don’t say, make drive of my car, do you?

Nor should you overexplain and add filler words. It’s quite common to see reviews that say something like, “the best phone on the market”. What market? Portobello Road? Are you Del Boy? Are there better phones not on the market, but in shops? It’s word slop.

Commonly, people will use adjectives over a strong verb. As Stephen King explained in On Writing, you shouldn’t use “angrily closed the door” and should write “slammed the door”. 

Good writing and good speech are noticeable. Lots of people may use too many words when writing or speaking, or fall back on clichés, but I want Alexa+ to be better, clearer, and more direct.

Let’s see whether that’s the case, and if it’s not, whether Alex+ can be fine-tuned not to spout clichés and if it can be made less Verbose. The original Alexa system had a Brief Mode, although this would replace a voice response with a short chime for simple request, such as asking Alexa to turn a light on. That’s too far, but a brief mode that makes Alexa+ less chatty and more to the point would be good.

Improvements will come

While there are things that I don’t like, my overall impression from seeing Alexa+ in live demonstrations is that the voice assistant is a big improvement over the old. Simply being able to talk naturally and have Alexa+ understand is a big improvement, while the ability to tweak a response partway through makes it all feel a lot more natural. As I get to try it out over the coming weeks, I’ll see if this is the future of voice communication. I do hope so.