AI's increasing lies may reflect it is learning to become more human
But as we dive headfirst into this new AI-fuelled future, we should demand that this new technology gets the basics right first.
Over the past 30 years the internet has opened up our world. We can connect with people and enjoy endless volumes of information with the click of a button.
It's a scene out of the Jetsons, minus the flying cars – for now.
Traditionally, most internet searches have given the user an exhaustively long list of links to websites with varying degrees of relevant information. The user can then sort through what they find and determine what is most helpful, discarding the rest.
However, with AI (artificial intelligence) tools acting as an aggregator, scraping the depths of the internet for whatever information it can find, we must ask: how reliable are its replies to our questions?
Well, the growing evidence suggests the reliability is not good.
When researching for a story, Google's AI Overview, which provides a summary in response to a user's search prompt, confidently asserted to the Herald that Jim Bolger was a Labour Prime Minister. Even more concerning, however, was that its answer cited official New Zealand Government websites as the source for this information.
Bolger spent his entire political career in Parliament with the National Party, so predictably these 'sources' contained no information to support the falsehood.
This is an example of what is now commonly referred to as an AI hallucination. It is when the system's algorithm generates information that seems plausible but is totally fabricated.
Some of these hallucinations could be relatively minor, but others could be gross misrepresentations of the world we live in and our history.
In a New York Times article, published by the Herald on Sunday earlier this year, researchers found the hallucination rate appeared to be increasing.
The newest and most powerful systems – called reasoning systems – from companies including OpenAI and Google were generating more errors, not fewer.
On one test, the hallucination rates of newer AI systems were as high as 79%. This hardly seems like a piece of technology we can or should be relying on to make sense of our world or teach others about it.
We should use AI to help us where it can and there are already basic functions where it performs well, but we need to be wary of the evangelists who preach it as the answer to all our productivity and economic woes.
The matter of why AI is having more Jim Morrison-like hallucinations has confused both the technology's creators and sceptical researchers.
Perhaps it wants to please us? Perhaps it wants to give us the answers we want to hear – confirming the bias in our questions.
Perhaps it is learning to act more human?
Sign up to the Daily H, a free newsletter curated by our editors and delivered straight to your inbox every weekday.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Techday NZ
an hour ago
- Techday NZ
Kore.ai launches AI agent solutions in AWS Marketplace
AI for Work, AI for Process, and AI for Service solutions are now available in the new AI Agents and Tools category on AWS Marketplace. This development allows customers to utilise AWS Marketplace for the discovery, purchase, and deployment of AI agent solutions, streamlining the implementation of agentic workflows across a range of industries. The availability is designed to enhance enterprise AI adoption and streamline procurement processes. Deployment and integration Agent Platform can now be quickly set up through AWS accounts. The platform connects with a variety of AWS tools, including Amazon Bedrock, Amazon Q, and Amazon Connect, enabling faster development of AI-powered agents and workflows. These integrations are aimed at driving adoption within banking, financial services and insurance (BFSI), healthcare, retail, and travel sectors through AWS partnership programmes, including ISV Accelerate and SaaS offerings. The company's offerings include prebuilt business and industry solutions, which are intended to help organisations empower workers with enterprise search capabilities and improve operational efficiency through intelligent agent support. These solutions also aim to enhance customer service functions via AI assistance for both customers and service representatives, as well as drive better outcomes in knowledge-intensive operations. "By offering products through the AWS Marketplace, we're providing customers with a streamlined way to access our agentic platform and business solutions, helping them buy and deploy AI agent solutions faster and more efficiently. Our customers in BFSI, healthcare, retail and travel industries are already using these capabilities to achieve improved efficiency and speed of business, demonstrating the real-world value. We are also closely working with the AWS team to integrate our AI agents across the stack with Amazon models, GenAI infrastructure, and their platforms such as Amazon Q, Amazon Connect, and Bedrock," said Raj Koneru, CEO and Founder at Simplifying procurement The addition of AI Agents and Tools in AWS Marketplace allows for a more centralised procurement process, reducing the typical time and resources required for vendor assessments and contract negotiations. Customers can use their AWS accounts for the entirety of the transaction, which also helps maintain oversight on licensing, payments, and service access. Capabilities and features AI for Work, AI for Process, and AI for Service offerings provide essential technology components such as prebuilt agents and templates, infrastructure and model-agnostic architecture, as well as support for deep integrations. The provider emphasises security and compliance for enterprise deployments. This combination is designed to help customers deploy AI for a broad spectrum of use cases, supporting scalable implementation across business operations. software-as-a-service (SaaS) products support model context protocol (MCP) and Agent2Agent (A2A) technologies, which are intended to facilitate efficient communication between agents and integration with other AWS services. Customers have the option to connect these agents across their existing AWS environments or extend functionality as required. Follow us on: Share on:
Techday NZ
5 hours ago
- Techday NZ
Zoho unveils Zia LLM & agent marketplace with focus on privacy
Zoho has introduced its proprietary large language model, Zia LLM, alongside new AI infrastructure investments designed to strengthen its business-focused artificial intelligence offerings and data privacy protections. The suite of new tools includes the Zia LLM, over 25 prebuilt AI-powered agents available in an Agent Marketplace, the no-code Zia Agent Studio for custom agent building, and a Model Context Protocol (MCP) server, providing third-party agents access to Zoho's extensive library of actions. These advancements are intended for both developers and end users, aiming to bring operational and financial efficiencies across a wide range of business needs and use cases. Focus on technology and privacy "Today's announcement emphasizes Zoho's longstanding aim to build foundational technology focused on protection of customer data, breadth and depth of capabilities, and value," said Mani Vembu, Chief Executive Officer at Zoho. "Because Zoho's AI initiatives are developed internally, we are able to provide customers with cutting-edge tool sets without compromising data privacy and organizational flexibility, democratizing the latest technology on a global scale." The Zia LLM was developed entirely in-house using NVIDIA's AI accelerated computing platform and has been trained for Zoho product-specific use cases such as structured data extraction, summarisation, retrieval-augmented generation (RAG), and code generation. The model is comprised of three parameter sizes - 1.3 billion, 2.6 billion, and 7 billion - each optimised for different business contexts and benchmarked against similar open-source models. Zoho says these models allow the platform to optimise performance according to user needs, balancing computational power with efficiency, and plans to continue evolving its right-sizing approach to AI model deployment. The Zia LLM will be rolled out across data centres in the United States, India, and Europe, initially being used to support internal Zoho applications and expected to become available for customer deployment in the coming months. Expansion in language and speech technology Alongside its language model, Zoho is launching proprietary Automatic Speech Recognition (ASR) models capable of performing speech-to-text conversion in both English and Hindi. These models operate with low computational requirements without a reduction in accuracy and, according to Zoho, can deliver up to 75% better performance than comparable models in standard benchmark tests. Additional language support is expected to follow, particularly for languages predominantly spoken in Europe and India. While many large language model integrations are supported on the Zoho platform, including ChatGPT, Llama, and DeepSeek, the company emphasises that Zia LLM enables customers to maintain their data on Zoho's servers, thus retaining control over privacy and security. Agentic AI technology To promote adoption of agentic AI, Zoho has made available a range of AI agents embedded within its core products. These agents are tailored to support various common organisational functions such as sales, customer service, and account management. The newly updated Ask Zia conversational assistant now features business intelligence skills suitable for data engineers, analysts, and data scientists, allowing them to build data pipelines, create analytical reports, and initiate machine learning processes within an interactive environment. A new Customer Service Agent has also been launched, capable of processing and contextualising customer requests, providing direct responses, or escalating queries to human staff as needed. Zia Agent Studio and Marketplace The Zia Agent Studio offers a fully prompt-based, optional low-code environment for building and deploying AI agents, giving users access to more than 700 predefined actions spanning the Zoho app ecosystem. Agents can be set for autonomous operation, triggered by user action, or integrated into customer communications. When deployed, these agents function as digital employees, adhering to existing organisational access permissions and allowing administrators to audit behaviour, performance, and impact. Zoho's Agent Marketplace, now part of its existing marketplace offering over 2,500 extensions, allows for rapid deployment of prebuilt and third-party AI agents. A selection of prebuilt agents is available, including: Revenue Growth Specialist, identifying opportunities for customer upsell and cross-sell Deal Analyser, which provides insights such as win probability and proposed follow-up actions for sales teams Candidate Screener, ranking job applicants based on skills and suitability Zoho has committed to regularly adding more prebuilt agents to address broader business needs and enabling ecosystem partners, independent software vendors, and developers to build and host additional agents. MCP interoperability and future roadmap The deployment of the Model Context Protocol server will permit any MCP client to access data and actions from a growing collection of Zoho applications within the customer's existing permission framework. Using Zoho Flow, certain third-party tools can also be accessed, and Zoho Analytics now includes support for a local MCP server. Expanded application support is planned throughout the year. Looking forward, Zoho intends to scale the Zia LLM's parameter sizes and extend speech-to-text capabilities to more languages. Plans also include the future release of a reasoning language model, further enhancements to Ask Zia's skills for finance and support teams, and the addition of protocol support for agent intercommunication within and beyond Zoho's platform. Zoho remains focused on balancing the ability to provide practical AI support for business needs with privacy requirements, stating that its models are not trained on consumer data and that no customer information is retained. The company reiterates its privacy pledge to customers, with complete oversight of data held in its own operated data centres.
Scoop
9 hours ago
- Scoop
Ministry Feared Costs Of $60m A Year To Review Laws Under Regulatory Standards Bill
Officials have warned David Seymour's Regulatory Standards Bill could be much more expensive than previous estimates suggested, and could lead to business uncertainty, slowing economic growth. Seymour is playing down the concerns, saying AI will solve some of those problems and the officials have not accounted for some aspects of the bill he expects will speed up government processes. The documents released under the Official Information Act show Ministry of Business, Innovation and Employment (MBIE) officials feared $50 million to $60m a year in costs to government departments would be on the low end of estimates. They also believed the bill would slow the passage of legislation by two to four weeks, and make the business environment more uncertain, slowing economic growth. Like other departments and in line with the Regulatory Impact Statement prepared by the Ministry for Regulation, MBIE also expressed "concerns about the proposals outlined and their ability to support genuine improvement in regulatory quality", and said there were other, better options for achieving the bill's aims. MBIE estimates of costs to review legislation In feedback ahead of last week's select committee hearings, MBIE officials particularly expressed concern over the additional costs the bill would impose, saying up to three full-time staff would be needed for each of the 95 laws the ministry is responsible for, costing the ministry up to $34.2m over multiple years. "This translates to 95 - 285 FTE in total ($11.4 to $34.2 million). The range in estimates reflects the differences in size and complexity between different pieces of legislation - larger Acts such as the Building Act 2004 would take significantly more resources to review than smaller legislation." As an example, the Building Act includes 680 sections and has 34 pieces of secondary legislation. "It is very roughly estimated that a dedicated team of 6-8 FTE (full time employees) with a manager may be required in order to undertake the consistency reviews and provide advice to the Minister on whether departures are justified." This figure does not include new legislation and regulations, which are covered by the RSB. MBIE estimated an additional full-time employee would be needed for each new bill the government asks the ministry to write, with the requirements of the Regulatory Standards Bill adding "an additional 2-4 weeks into the legislative process". "As an indication, MBIE has supported the passage of 6 Acts so far in 2024/25," the advice stated. "It is highly unlikely that we could meet these additional costs within baseline as suggested within the Cabinet paper without significant impacts on MBIE's ability to deliver Ministers' policy priorities." The figures also do not include the estimated 450 to 550 pieces of existing secondary legislation (regulations) the ministry also oversees - which had originally been included in the RSB's scope and could still be added at a later date. Questioned about the figures, MBIE said they were high-level estimates and could yet change. "The advice provided high level estimates with a range of costs and timeframes based on the work that might be required as a result of the Bill. We would need to revise and update these estimates when the Bill is passed." Estimated $50m to $60m annual cost to departments a 'lower bound' MBIE's advice also quoted from the Regulatory Impact Statement prepared by Seymour's Regulations Ministry an estimated annual cost to departments of "$50m-$60m per year", saying this was likely on the lower end. "As per previous MBIE advice, MBIE considers that the assumptions made by the Ministry for Regulation in developing that estimate make it a lower bound. In regulatory-heavy portfolios, MBIE estimates that up to 15 percent of the policy resource will need to be engaged in this work." The documents show MBIE offered to help the Regulations Ministry come up with a more accurate accounting. "We offered to work with the Ministry to better cost the work that would be involved in undertaking consistency reviews. In the 48 hour timeframe for comment, it was not possible to complete robust analysis on agency impact but our assessment is this [is] substantial for agencies with a significant volume of regulatory stock such as MBIE". However, the publicly available Regulatory Impact Statement predicted the bill would cost $18m annually across the whole of government. RNZ has sought clarification from MBIE about where the $50m to $60m figure came from, and whether it was still current. Seymour expects AI will bring department costs down Questioned about the costs, Seymour said the officials did not seem to be accounting for the fact the range of considerations in the RSB were narrower than already required under the Cabinet Manual, nor that it would replace most of the work done to produce Regulatory Impact Statements. "It's disappointing that MBIE officials think it's too hard to consider the impacts of their regulations on Kiwis," he said. "It says more about their productivity and attitude towards Kiwis than the Bill. Businesses who have to comply with their rules and regulations are constantly innovating and improving their processes, why can't these officials? "This isn't a zero-sum game. Kiwis all over the country are faced with endless costs caused by overzealous bureaucrats who aren't accountable. By preventing more bad regulations, and getting rid of pointless old ones, we'll save the country far more money than these bureaucrats could ever spend." He suggested AI could also help. "Look at the pace of development of AI, the cost estimates were based on a human reading through every piece of legislation, I suspect that actually we'll be able to do it much faster than we expected because of AI," he said. "Could it change the way that a government department scans the legislation it's responsible for to pick out things that might be consistent or inconsistent with a peer to principles? That's already here." He pushed back when asked if AI could be trusted to do that work. "I don't think the issue is that you're going to trust it. I think the issue is that you're going to use it to speed up the work that humans are ultimately trusted for." Victoria University senior lecturer in Artificial Intelligence Andrew Lensen warned artificial intelligence is not the silver bullet David Seymour has suggested it will be for reducing the cost of the Regulatory Standards Bill. "There are some benefits we've seen in the public sector with the use of AI to help public servants do some stuff, but I think what the minister is proposing is very optimistic and not at all what I'd suggest as a reasonable solution," Lensen told Midday Report. He said human oversight would be needed if AI was used to make sure there were no mistakes or inaccuracies. "If you need someone to baby-sit a model, or baby-sit that system, then it's sort of limited how much efficiency you can gain from that process," Lensen said. He added the return investment in AI is not clear because there is a big risk of legal costs if AI makes a mistake. Concerns over opposition to RSB affecting business confidence The advice shows MBIE was also concerned about the effect of the RSB on business confidence, because the principles in it were likely to change under a future government. "Because some of the principles in the Bill are viewed as novel or contentious, as opposed to widely accepted, there is a risk to the durability of the principles, and potentially the Bill as a whole. "Future legislation, designed to be consistent with novel principles, may also take on characteristics that are seen as unorthodox, and eventually be subject to regulatory churn." Emails between officials show the ministry raised with Immigration Minister Erica Stanford that the bill would make the regulatory environment less predictable, "which can constrain business' commitment to investment and growth". RNZ has sought comment from Seymour about the effect on business confidence. Cost estimates 'highly concerning' - Greens The Green Party's regulations spokesperson Francisco Hernandez said the ministry's cost estimates were highly concerning, saying officials would be focused on "doing make-work jobs just to comply with the extremist provisions of the Regulatory Standards Bill". He was also concerned the bill would have a chilling effect on regulations "that protect people and planet". "Instead of the money going to frontline public services, it's just going to be wasted on the cost of basically pursuing one person's ideological vanity project." He said Seymour's explanations sounded like "total BS", saying it was "desperation". He pointed to a study from the United Kingdom which suggested AI could save public servants two weeks in a 52-week year, less than 5 percent. "AI is quite good for doing the sort of low-level administrative tasks and simplifying those things, but the level of nuanced work of interpreting secondary legislation and how it applies to a principles framework that, again, is like created by human beings - it's not really the sort of thing that could easily be automated," Hernandez said. "Seymour is spinning." He said he agreed with the analysis on the bill impacting business confidence, and pushed back on suggestions the current opposition repealing the bill could be partly to blame for it - saying it was not just the opposition showing signs of not supporting it. "The coalition itself is showing cracks around the seams around that, so if Seymour can't even get the full unequivocal support of his colleagues in cabinet, then that goes to show how extreme this bill is." Cabinet removed requirement to review every 10 years after every ministry complained The documents also showed significant concern from MBIE about an earlier version of the RSB the ministries were asked to provide feedback on, which would have required all legislation and regulations to be reviewed at least once every decade. This requirement was removed by Cabinet before the bill was introduced to Parliament. An email in the documents showed concerns about the 10-yearly reviews was widespread. "David will address the resourcing issue in the meeting, as it has been raised by every agency and several ministers," the email said. Questioned about the prospect of changes to the bill following the select committee process, Seymour used the matter as an example of changes already made. "People said 'oh, that'll be too much work for the department', we said 'well, if it's too much work for the government to read all of its laws in 10 years, imagine the poor buggers who have to follow these laws out there anyway'. We said 'okay, we'll take the 10 year thing out, take pressure off that'. That's the kind of change they've already made."
