I'm a pc scientist and a terrible Christmas shopper. Over the weekend I asked myself whether AI systems might have the ability to assist me.
Could I just ask ChatGPT to pick out a customized present for my cousin Johnny and have it shipped in time to succeed in him? Unfortunately, the cheerful chatbot couldn't help and told me it “can't make purchases or handle shipping directly.”
In the 2 years since ChatGPT launched, we've seen wave after wave of AI products and features that promise to avoid wasting us from on a regular basis tasks. But to this point they’ve not been capable of shop for gifts – other than the occasional built-in chatbot on a shopping website or app.
However, things may very well be different by next Christmas. Many experts expect the rise of AI agents in 2025: bots that may take actions in your behalf in the actual world.
Agents are already here
An AI agent can do greater than just suggest where you may get a Santa suit. They should purchase it for you and have it delivered to your door.
And the vision for “agency AI” is for teams of AI agents to work together. You would give your team of agents a prompt:
I'm cooking Christmas dinner this yr. Find my closest Facebook friends, send them invites, be sure that one in all them is a chef, and tell them to bring the turkey.
The agents would handle every part without you ever having to lift a finger. Crucially, AI agents should have the ability to coordinate across multiple web sites.
In fact, limited AI agents exist already. A report According to AI developer Langchain, 51% of respondents to its survey are already using AI agents in production.
Venture funds will spend money on 2024 an estimated $1.8 billion in AI agent projects. The latest from Deloitte Global forecast report argues that 25% of corporations using generative AI will launch agentic AI projects in 2025.
Research firm Gartner predicted that by 2028, 15% of each day work decisions will likely be made by AI agents.
Agents for everybody?
We are also seeing consumer agents emerge. These are systems that would potentially automate many browser-based tasks (including shopping).
In October, Anthropic – the corporate behind the favored generative AI bot Claude – released a “Computer use“Feature that permits AI to take over a user’s mouse and keyboard to browse and perform actions on any website.
Education expert Leon Furze created one Demo with computer use To routinely navigate to a learning management system, open the page for an project, create text for the project, and click on the Submit button. Everything happens routinely through a single text prompt.
Recently, Google Deepmind released its own version, Project Marinerwhich also enables an AI to autonomously navigate and perform actions within the Chrome browser.
Both systems are still early versions, with Project Mariner only available to a trusted group of testers. But they hint at what’s to come back.
None of those tools mean you can automate your holiday shopping today – at the least not easily. So what wouldn’t it take to make a really useful AI agent for holiday shopping?
The technology is there
The technological side of being a shopping agent is comparatively easy. As a user, I’d want to present a prompt, e.g
Sending photo gifts to my family in England. Select a number of fun family photos from my phone, find a web site that provides photo gifts, order matching gifts for every member of the family and send them using my address book.
To execute this might require multiple AI agents: one to seek out the photos, one to seek out the shopping sites, one to personalize the gifts, a bank card agent to buy, and an address locator agent.
Whether through computer use, Project Mariner, or another AI agent platform, there is no such thing as a technological reason why this can’t be done today.
The trust problem
However, there are two significant barriers to the usage of AI agents.
The first and most blatant is trust. Would you trust an AI agent together with your bank card details?
Despite two years of advances in AI since ChatGPT, hallucinations – where the AI ​​doesn't know the reply and just makes something up – are still an issue.
A current study showed that even in AI programming – some of the popular and useful applications of AI – 52% of AI-generated answers to coding questions contained errors.
All it takes is one mistake from the AI ​​to send Aunt Molly's gift to Uncle Joe. And let's just hope it's a harmless error, e.g. B. about poor allocation of gifts and never disclosing your checking account details.
What agents must know
The second and fewer obvious obstacle is that AI agents need to know context in an effort to be useful. Even with something relatively easy like buying gifts, context is every part.
I've known what my mother likes for years. I won't at all times get it right, but I'll do it significantly better than a generic AI answer. This knowledge is generally tacit and there’s simply no way that ChatGPT has access to the wealthy history of human interactions that result in that perfect gift.
However, AI bots already record details about their users. To prove this, simply ask ChatGPT, “What do about me?” Depending in your settings, you might be surprised by the reply.
Maybe someday the AI ​​systems we use recurrently will know enough about us and our families to have the ability to completely automate holiday shopping.
But this yr I still should deal with it myself. Bah humbug!
