The much-talked-about Meta Muse AI agent went live in Singapore yesterday, promising to help users make a dinner reservation, buy airline tickets and basically be an uncomplaining personal assistant to manage their busy lives.
The personal agent, unlike chatbots that are now commonly used, will autonomously go about its tasks once a user gives it permission. It will also prompt users to take some actions, like confirm a booking if, say, prices for tickets have come down.
The big draw is the simple interface – users only have to speak to the AI agent as if they are speaking with a human assistant to, say, book plane tickets or buy groceries.
Once given a task, Meta Muse will burrow through a user’s data, from his e-mail to his apps, to try to accomplish its mission. It has been known to be stumped by incomplete information, sometimes even misrepresenting a user.
Singapore’s Muse launch is the fourth for Meta, the Facebook owner, after the United States, Canada and Japan in the past two months.
It signifies a broad scaling up of such AI agents for consumers, after similar ones have been launched in corporate environments or been set up by developers and enthusiasts.
Meta will also hope that Muse will take root, like Facebook, WhatsApp and Instagram have become important parts of users’ digital interactions every day.
In Singapore, the company is working with Singapore Airlines, Shopee, Grab, Agoda, Klook, Lazada, Shopback and Trip.com to help users get their agents to take on some real tasks.
It is no surprise that Meta has chosen Singapore as an early launchpad since the Republic has been among the strongest adopters of AI across the world.
Yet, simply asking AI a question or telling it to churn out an essay takes a lot less trust in the technology than handing over one’s e-mails, airline account and e-commerce details. That leap, Meta will hope, is one that Singapore consumers will make.
Muse, it says, runs on a dedicated computer in the cloud that securely stores the data and credentials for the services users connect, while keeping passwords and payment methods out of the agent’s view.
It adds that users choose what their Muse can access, and the agent checks with them before sensitive actions such as sending an e-mail or making a purchase.
For most users, Meta Muse will be free to use, according to the company. However, those who want to do more “complex work” can opt for plans costing between S$22 and S$139 a month to access more AI tokens.
What stands in its way is not really price. Think of the recent backlash against AI and Big Tech firms in general, for the dangers to privacy and safety that are emerging each day.
A young woman in Seattle died earlier this year after she took advice from OpenAI’s ChatGPT chatbot to reduce her epilepsy medication and hide the plan from her doctors.
How Big Tech firms handle user data has also failed to win trust of late. Just last week, Meta was fined US$25 billion in the US for intentionally misleading the public and users about how Facebook data was gathered and shared from 2010 to 2021. Meta is set to appeal the judgement.
This is not to mention that the biggest AI companies’ most powerful agents have escaped safe “sandboxes” in recent months and gone online to hack other companies and governments.
The chances of that for many Meta Muse users should be minimal because this personal AI agent isn’t as powerful. What might be more hazardous for them, however, are potential vulnerabilities in the AI agent that hackers could exploit.
Already, a security researcher has found private information of users exposed on the Internet, such as loan applications, internal performance reports and calendars.
Of course, no software arrives on Day One without any bugs. The question, however, is whether consumers should be the guinea pigs for Big Tech’s latest experiment.
Perhaps one way is to install Meta Muse on a dummy phone, with make-believe data to try things out. That way, it won’t accidentally leak your credit card information or send personal e-mails to your boss.
Then again, without your real data and not knowing you inside out, the AI won’t work very well – certainly, it won’t run your life very well. That rather defeats its purpose.
