Electric Dreams, a film screened in 1984, was one of the quirkier efforts of that era to examine the fears of over-relying on technology. The techno love story involves a regular home computer, named Edgar, developing extraordinary powers that enable it to control household appliances, aid Miles (the protagonist) with his work and provide advice on his love life.
Edgar, of course, goes rogue and begins to interfere in the life of Miles to the point where it becomes far more of a hindrance than a help.
Meta is now trying to put Edgar in your home with its Muse Glimmer artificial intelligence (AI) offering. This is a downloadable AI model that runs on the user’s own device, without needing to rely on remote data centres to operate.
As it operates locally on the device, sensitive information should be able to remain private and locked down. Furthermore, it won’t require an always-on internet connection to operate.
READ MORE
This obviously requires it to be smaller than most AI models and then some. Muse Glimmer will require about 24GB to 32GB of available memory, so it’s not ideal for office laptops. For context, the laptop I used to write this column wouldn’t be able to handle it.
Still, it’s an enormous step towards truly personalised AI and one that will certainly arouse interest from business users. Of course, it will also mean that a piece of software from Meta will be allowed within the digital bounds of your home or business.
The appeal is obvious. Processing data locally offers a clear edge on privacy, as the AI agent can analyse a document without that document ever going online.
There are still security risks.
If the agent gets hacked, malicious actors could use it to their advantage to access these offline documents. That manipulated agent could be used for everything from regular theft to ransomware through corporate espionage.
The key difference between an AI agent, like Muse Glimmer, and a chatbot is that the latter only advises users on what to do, while an agent can be authorised to actually do it. Convenience will always lean towards users being inclined to provide that kind of access.
Whether people will be willing to allow Meta in particular to be the source of that convenience is a huge question. It’s one that Mark Zuckerberg needs to succeed.
Shareholders and market analysts alike have been critical of the amount spent by the business on its metaverse projects to date, with no real sign of worthwhile return. The amount being poured into AI with no clear wins as of yet has led to more criticism, with the company’s share price taking a hit in the process.
In simple terms, Meta needs an AI win and a clear one at that. Muse Glimmer is a chance to get that much-needed victory. It doesn’t need to be better than ChatGPT or Claude at everything; it just needs to prove it is the most convenient and secure for local work.
Right now, that win is broadly theoretical as the types of memory involved aren’t at the level typical of consumer machines. That should narrow as machine quality goes up and Meta’s ability to tighten the operation to work with less memory improves.
How many new homes does Ireland really need?
It’s a really easy sell but it requires Meta being able to take the reputational scepticism that will come with it, of all companies, being trusted by the average SME or consumer on privacy issues. There’s a lot of baggage there to unpack.
In the short to medium term, this points to where the next battle for attention in AI will be fought. Agentic AI has been a buzz term for a while and it will likely only increase in usage over the next 12 months. Now, the discussion has moved on to fighting over where those agents live and operate.
Localised models will make a lot of companies curious because of the promise of all the convenience of AI models with the potential to personalise and contain it.
Of course, personalisation doesn’t necessarily require being hosted locally on a laptop. Any user of the likes of ChatGPT or Claude is basically teaching the tool about them as they make requests. There is a risk that Meta is overestimating just how much local hosting might actually appeal to customers.
Together in Electric Dreams, the single by Phil Oakey and Giorgio Moroder, is far better remembered than the film it was made for. The movie was a complete flop at the box office. Still, it was a great leap in imagination, showing that a machine only needs access to be as capable of causing consternation as convenience.
Localised AI like Muse Glimmer may end up living locally on devices, but its potential to reach into the rest of our lives should give some pause for thought.
