With the rise of 'AI agents' that perform tasks on behalf of humans, we may be heading towards a future where even choosing Christmas presents will be left to AI.



Giving Christmas presents to loved ones may be a fun event, but some people may find the series of tasks such as careful research, purchasing, and arranging for delivery difficult. John Whittle, director of

Data61 , a digital technology research and development organization under the Commonwealth Scientific and Industrial Research Organization (CSIRO) of Australia , claims that 'in 2025, AI agents that can act in the real world may emerge and will be able to choose and purchase Christmas presents on your behalf.'

AI can't do your Christmas shopping just yet – but next year might be different
https://theconversation.com/ai-cant-do-your-christmas-shopping-just-yet-but-next-year-might-be-different-246132



Since OpenAI released ChatGPT in November 2022, various AI-powered services have emerged, but at the time of writing, there is no AI service that can perform the task of 'selecting a Christmas gift for someone, purchasing it on a shopping site, and shipping it to the recipient's home.' However, Whittle said, 'The situation may change by next Christmas. One of the things that many experts expect to happen in 2025 is the rise of AI agents that can act on your behalf in the real world.'

An AI agent that performs the task of 'selecting a Christmas gift for someone, purchasing it on a shopping site, and shipping it to the recipient's home' must have the ability to coordinate tasks across multiple websites. AI agents that run on a limited number of websites have already appeared on the market, and in a report released by AI development company Langchain, 51% of respondents said they use AI agents in production environments.



AI agent development projects are said to have raised

an estimated $1.8 billion in funding in 2024, and accounting firm Deloitte predicts that 25% of companies using generative AI will start AI agent projects in 2025. AI agents are attracting a lot of attention.

We are also starting to see the emergence of AI agents that consumers can customize themselves to automate browser-based tasks, rather than those provided by companies. In October, generative AI developer Anthropic released a feature called ' computer use ' that can control the user's mouse and keyboard to perform actions on websites.

In addition, in December, Google announced 'Project Mariner,' an AI that can understand information on Chrome and operate it automatically.

Google announces 'Project Mariner' that can automatically operate Chrome with AI - GIGAZINE



At the time of writing, these tools can't be used to automate Christmas gift shopping, but in the future, more helpful AI agents may emerge and humans may no longer be responsible for choosing Christmas presents.

Whittle points out that the technology challenges are not so great when it comes to building an AI agent to carry out Christmas shopping: 'I want to send photos to my family in the UK. Please select some happy family photos from my smartphone, search for a website that offers photo gift services, order the right gifts for each family member, and use your address book to send the gifts.'

In this case, multiple AI agents would be needed to execute the task: one to find the right photos on a smartphone, one to find shopping sites, one to find the right gifts for each person, one to buy gifts using a credit card, one to find family addresses, etc. However, there is no technical reason why these tasks cannot be performed with modern technology.

According to Whittle, there are two major barriers to the realization of multi-AI agents. One of them is that 'AI agents lack trust,' and there are still concerns about entrusting tasks such as using credit cards to AI. AI is also prone to hallucinations, which are false responses, and there is a risk of problems such as 'giving a gift to someone you didn't mean to give it to' or 'accidentally leaking credit card or bank account information.'

The second barrier is that 'for an AI agent to be useful, it needs to understand detailed context.' For example, to choose a gift that will please someone, knowledge such as 'what the person likes,' 'what they received last year,' and 'did they say they wanted something recently' is useful, but this information is generally implicit and is not something that an AI agent can find out by looking it up. In that respect, it is possible that AI agents are less accurate than humans at choosing gifts.



Despite some issues, AI agents are steadily evolving, and it's quite possible that we'll soon see AI agents that can do our Christmas shopping for us, Whittle concluded.

in Web Service,   Science, Posted by log1h_ik