This article asks the question of when – or more pertinently, if – what we presently refer to as artificial intelligence will make its way into integrated home automation systems. Stephen Dawson writes.

When the editor put that question to me and asked me to write about it, I did what any modern writer would do: I asked artificial intelligence (AI) to answer the question for me.

But first I had to formulate the question. After having played with a some of the free versions of AI, I’ve already learned some lessons. Such as, to get a relevant answer you must frame your question quite well. And even so, it may well lie to you.

Why, just last night Bing Chat told me in that conversational tone it adopts, while reeking of certainty, an utter falsehood. The AI community uses the euphemism “hallucination” for false answers, but to me a false answer stated with authority isn’t merely a dream, it is indistinguishable from a lie (which isn’t to say that AI is useless. I couldn’t remember that “hallucination” was the word I wanted, so I asked Bing Chat “what term is used when ChatGPT provides a false answer?” and it provided the word).

So, I asked:

Will AI be incorporated into integrated home automation systems? If so, when is that likely to happen?

Bing Chat came back within, seriously, just a few seconds, with the following:

Yes, AI is already being incorporated into integrated home automation systems. Smart home devices are becoming more intelligent and capable of learning from user behaviour to provide a more personalised experience. For instance, smart home systems can learn your daily routine and automatically adjust the temperature, lighting and other settings to your preferences. They can also detect when you leave the house and turn off the lights or lower the thermostat to save energy.

As for when this technology will become more widespread, it’s difficult to say. However, given the rapid pace of technological advancement in recent years, it’s likely that we’ll see more AI-powered smart home devices in the near future.

I hope this helps! Let me know if you have any other questions.

Bing Chat is always so cheerful!

Bard, which is Google’s version, gave a similar answer – at 300 words, too long to copy here – but added a bunch of examples. Bing Chat, by contrast, included four footnotes which linked to articles on the web.

Neither answer was particularly useful. I played with some variations, and the answers were also variations of those already given.

ChatGPT and similar “artificial intelligence” agents are often impressive, but they have enormous limitations.

Ad hoc consumer level automation vs professional integrated home automation

You see, all those answers essentially led back to either individual devices – your Roomba is a smart, network connected device – or one of the consumer-level smart home systems, such as Amazon Alexa or Google Assistant/Nest. They have their place, but they are not the basis of the type of home systems integration we are most interested in here.

Not once did I see a reference to, say Control4 or to Crestron, nor any of the other professionally installed systems.

The current impressive generations of AI use large language models, essentially predictive systems built on being fed enormous quantities of existing text. There are billions of words on the web about Alexa and Assistant. Nowhere near as many about Control4. So, the AI answers are always going to be strongly biased, in this area, towards those consumer-level “smart” systems. That’s one problem with current AI.

The main difference between those consumer systems and the installations built by professional installers is, in essence, reliability.

Anyone who has tried something as seemingly simple as controlling a networked home theatre receiver by voice using Alexa will know that the setup is far from consumer friendly. Whether it will work at all is iffy, and that’s for the simplest things, like changing inputs and changing volume level. If there’s a problem, where is it? The devices might be fine, but perhaps the consumer-level router installed by the customer’s ISP needs to be rebooted on a regular basis to ensure smooth operation. And the devices which can be controlled – the “smart” lights, “smart” plugs, “smart” curtain openers and closers, “smart” security systems and myriad of other “smart” devices – come from multiple vendors, frequently with differences in implementation.

An inveterate tinkerer can have a lot of fun building a home automation system with relatively inexpensive consumer-level devices and equipment, but that isn’t what the great bulk of smart home customers want.

What they want is a straightforward, easy to use system that is utterly reliable. Which is why those long-standing home system integration systems are still around.

Having failed in my quest to answer the question with artificial “intelligence”, I turned to an expert in systems integration and home automation.

Human thoughts on the present and future of AI in integrated home automation systems

One such expert is Chowmain Software chief technology officer, Richard Mullins. This company specialises in developing drivers to allow devices to be operated by a wide range of professional-level home automation systems. And he has thoughts, which he shared with me in a wide-ranging conversation. What follows is a distillation of that conversation, so if there are any errors, they’re most likely to be mine.

First, he educated me on the difference between consumer-level and commercially-installed smart home systems. I suspect that Richard would be quite happy as one of those inveterate tinkerers to whom I earlier referred, and in fact I gather that he is the software equivalent of one in his current role.

He is slightly sceptical of AI in general, correctly noting some of the problems with it that I pointed out above. For one thing, it isn’t really intelligence at all. For the moment, it’s merely a remarkably effective prediction engine, guessing through a model of (principally) the English language and an ungodly amount of data what words and sentences would appropriately answer a query. It has no internal conception nor understanding of the question. Consequently, despite appearances, the inherent literalism of computers remains in place.

It would be nice to see a virtual butler in your integrated system, with which you could converse to make your desires clear. You know, you could essentially program a series of tasks as easily and naturally as you could ask your spouse: “Could you look after things with the kids tonight while I go out?” Certainly, a system with a voice interface could receive the command, and if there’d been a great deal of preparatory programming so that the command would invoke a previously defined “look after things with the kids” routine – not any old one, but the one for evenings – then it might be able to close curtains, order a pizza, turn on a TV and so on. But all that seeming “intelligence” is simply an interface layer to a standard algorithmic system.

And given the willingness of current AIs to lie – sorry, to hallucinate – at worst, or simply follow false trails due to the preponderance of the data upon which they’ve been trained, reliability in performing even relatively simple tasks seems pretty iffy.

And then there’s language. Like it or not, English is the language of the internet, and the current versions of AI have been disproportionately trained on text in that language. It will be quite a while before they are as effective – to the extent that they are – in other languages. Or a further translation layer may have to be inserted in the interface, leading to even greater opportunities for things to go awry.

Richard also noted that the major demand from customers for home automation systems seems to be simplicity. He pointed out that both smart devices and control systems offer many more features than are often used. For example, some smart power adaptors can monitor current flow and provide a notification upon a change of state. A system can note this and, as a consequence, implement a series of actions: switch on associated equipment, for example. But there is little demand for these advanced features at this point. Indeed, the great majority of systems are essentially standard installations with only enough customisation to work properly in the customer’s environment.

Adding the unreliability of having AI attempting to interpret complicated commands and then applying them to the moving parts of a home automation system is likely to lead to frustration.

So, it all looks dark then

After all those downers, I should note a couple of points of optimism. First, a company like Josh.ai has been working with major systems integration systems for years, developing both hardware and software for effective voice control. And more recently, it has incorporated JoshGPT, an intelligent agent. I note that the intelligence used is similar to that in Bing Chat and Bard, and in its promotional video it seems to offer those same kinds of features provided by consumer systems. That is, you can ask it questions, have it make suggestions for what to watch or listen to and so on. The main advantage is that the Josh.ai voice interface works with it, and seemingly allows chained commands: Josh, close the curtains and switch on the TV.

With a specialist company like Josh.ai on the job, it seems to me that I may be unduly pessimistic about this subject.

The other point to note is that on several occasions over the past thirty years I have been asked in my capacity as an expert to opine on what the future holds. In each case, the passage of a few years (in one case, just a few weeks) has shown my opinion to be quite wrong.

All of which means: watch this space. AI, voice interfaces and this whole area is very fast moving and has a lot of extremely smart people working on it. Five or ten years from now, things could be very different.