It's a rather grey and partly cloudy day in London yet let's not allow that to dampen our exploration of this week's developments in artificial intelligence. In the spirit of Alan Turing, who once remarked, "I'm not telling you it's going to be easy. I'm telling you it's going to be worth it," we'll dive into the. task of separating signal from noise in a landscape inundated with claims and counterclaims. In the world of artificial intelligence we often find ourselves sifting through a myriad of announcements many of which promise transformative change yet offer little in the way of substance. This week we'll examine some of the most talked-about advancements and discern what truly merits our attention versus what may simply be noise. This is Turing's Torch: Artificial Intelligence Weekly — the bits that matter, minus the hype. Insurance companies, bless them, appear to be having a bit of a mare with artificial intelligence. Word is, they're struggling to make effective use of it. Yet before we start imagining rogue algorithms driving up our premiums, the problem isn't actually the artificial intelligence itself. No, it's the rather more mundane issue of their own data. What this boils down to is that these firms are finding it difficult. to feed their artificial intelligence systems the right information in a usable format. Picture it like trying to bake a cake with all the ingredients jumbled together in one big bag. The algorithms might be sophisticated, yet they're only as good as the data they're fed. If that data is poorly organised incomplete or just plain incompatible the artificial intelligence will struggle to produce accurate predictions or make informed decisions. The insurance industry is, after all, sitting on a veritable goldmine of data, everything from customer demographics to claims histories. Yet if they can't effectively leverage this data through artificial intelligence, they risk falling behind more agile competitors. We're talking about the potential to automate underwriting personalize pricing detect fraud and improve customer service all of which translates to cost savings increased efficiency and ultimately greater profitability. Yet those benefits remain tantalisingly out of reach until they sort out the data plumbing. This is just another example of how the promise of artificial. intelligence often clashes with the realities of legacy systems and outdated processes. Everyone wants the shiny new artificial intelligence yet few are willing to do the unglamorous work of cleaning up the data behind it. And it's not just about technology; it's about organisational culture. Insurance companies like many established institutions can be slow to adapt to change and that inertia can be a major obstacle to artificial intelligence adoption. One wonders if perhaps they should have spent less time boasting. about artificial intelligence pilot projects and more time hiring competent database administrators. It appears, once again, that basic competence is the limiting factor. The insurance industry will need to address these data challenges head-on. if it hopes to remain competitive in the age of artificial intelligence. It's all very well talking about the dazzling future, yet someone has to actually lay the cables, doesn't it? And while we're on the subject of financial institutions and their digital wrangling Mastercard has announced a new system for spotting fraudulent transactions using what they're calling a large tabular model. Now, that's a bit of a mouthful. In essence, it's a specialised form of artificial intelligence designed to analyse vast quantities of transaction data. Instead of dealing with text or images as many of these systems do this one focuses purely on the numbers and details associated with each payment. Mastercard have apparently trained it on billions of past transactions, teaching it to recognise patterns that might indicate fraud. The idea is to improve security by more accurately identifying suspicious activity which is becoming increasingly important as online payments become more common and as a result do the opportunities for criminals to exploit vulnerabilities in the system. By focusing on the structured data of transactions rather than trying to apply more general-purpose artificial intelligence Mastercard hopes to reduce the number of false alarms that can disrupt legitimate payments. The broader context here, of course, is the ongoing battle against cybercrime in the financial sector. Companies are under constant pressure to improve their security measures and this sort of targeted artificial intelligence could become a vital tool in that fight. It's also worth noting that this approach reflects a growing trend towards specialisation in the field. Instead of relying on one-size-fits-all artificial intelligence, we're seeing more systems designed for specific tasks, like fraud detection. Of course, the proof will be in the pudding. It remains to be seen whether this new system will actually make a significant difference in reducing fraud. After all, criminals are constantly adapting their methods, and any security system is only as good as its ability to keep up. One might even suggest that announcing such a system publicly is simply providing a roadmap for those looking to circumvent it. In the meantime it's a reminder that the world of digital payments is a high-stakes game and the technologies designed to protect it are constantly evolving. It's a bit like announcing your new, state-of-the-art burglar alarm system to all the local burglars. Visa, not to be outdone, is also experimenting with artificial intelligence, this time to automate the payment process itself. Instead of a person actively initiating a purchase, artificial intelligence agents would identify needs, find the best prices, and execute transactions automatically. The idea is that software would handle tasks currently done by humans, from spotting deals to completing purchases. Envision an artificial intelligence constantly monitoring prices for an item you want and then buying it for you the moment it hits a pre-defined price. This shift could have significant implications for how we interact with money. It promises faster transactions and potentially lower costs, yet it also raises questions about security, trust, and control. Who is liable if the artificial intelligence makes a mistake, buys the wrong thing, or is compromised by hackers? The banking sector is already showing signs of movement in this direction, as a result it's not just theoretical. It also reflects a broader trend of automating tasks previously performed by people, and the financial sector is no exception. As with all artificial intelligence applications, there are significant regulatory and security considerations. Robust safeguards and clear lines of responsibility will be essential if such a system is to gain public trust. One wonders if handing over complete control of purchasing decisions to an algorithm, even with the promise of efficiency, is truly progress. After all, the occasional impulse buy, the odd unplanned extravagance, is often what makes life interesting. A perfectly optimised existence sounds rather…dull. At any rate, it's another example of how artificial intelligence continues to reshape the landscape of everyday finance. Soon, we'll be nostalgic for the days of actually deciding what to buy, instead of simply being told what we need. And the US Treasury clearly mindful of all this has published guidelines for banks and other financial firms on how to manage the risks of using artificial intelligence. It's a detailed framework intended to help these institutions understand and mitigate the potential downsides of incorporating artificial intelligence into their operations. Now, when we talk about artificial intelligence risk in finance, we're not just talking about robots making bad investment decisions. It's about the entire range of potential problems that can arise when algorithms are used to make critical decisions. This includes things like inadvertently discriminating against certain groups when issuing loans exposing customer data through security vulnerabilities or even just relying on artificial intelligence systems that are simply too complex to understand and properly oversee. The Treasury's framework essentially provides a step-by-step guide for identifying, assessing, and managing these risks. Why is this important? Well, the financial system is the backbone of the economy. If artificial intelligence systems within that system start causing problems, the consequences could be widespread. Think about the potential for biased algorithms to deny mortgages to qualified applicants, exacerbating existing inequalities. Or imagine a large-scale data breach exposing the financial details of millions of people. The Treasury is clearly trying to get ahead of these potential. disasters by pushing firms to adopt better artificial intelligence governance practices now. This move also fits into a broader trend we're seeing which is a growing recognition that artificial intelligence needs to be managed responsibly not just in terms of its technical capabilities yet also in terms of its ethical and societal impact. We've seen similar initiatives in other sectors with governments and regulatory bodies around the world starting to grapple with the challenges of artificial intelligence oversight. One can't help yet wonder, though, if this framework will actually make a difference. Will financial institutions genuinely embrace these guidelines or will they simply pay lip service to them while continuing to deploy artificial intelligence systems with inadequate safeguards? After all, regulatory guidance is only as effective as the willingness of the regulated to comply. That said, it's at least a start. And a sign that the authorities are watching closely. I suppose it's a case of better late than never though one does wonder why it took as a result long for them to notice. Now, shifting gears slightly, let's talk about tools. Nvidia has released a new toolkit aimed at making artificial intelligence systems safer for businesses to use. The idea is to provide a framework that allows companies to deploy artificial intelligence. agents while retaining more control over their data and how the artificial intelligence operates. Now, when we talk about artificial intelligence agents in this context, we're not necessarily talking about physical robots. More often, it's software designed to automate tasks, make decisions, or interact with customers. Think of a chatbot handling customer service inquiries, or a system that automatically adjusts pricing based on market conditions. The problem as Nvidia sees it is that businesses are worried about the risks involved: data breaches biased decisions or unintended consequences that could land them in legal trouble. This matters because the hesitation around adopting artificial intelligence is real. Many companies are sitting on the sidelines, even though they see the potential benefits, because they're concerned about the downsides. If Nvidia's toolkit can genuinely reduce those risks, we might see a wider adoption of artificial intelligence across different industries. It's about building trust or at least the perception of trust in a technology that is still viewed with considerable scepticism by many. Of course, the toolkit is open source, which raises the inevitable questions about security. Open source can be a double-edged sword: while it allows for greater transparency and community involvement it also means that potential vulnerabilities are more easily identified – and potentially exploited – by those with malicious intent. The hope, presumably, is that the benefits of open collaboration outweigh the risks. One imagines there will be a few sleepless nights for security officers regardless. And that's perhaps the core issue here. Tools can only go as a result far. Ultimately, the safety and ethical use of artificial intelligence depends on the people deploying it, and their commitment to responsible innovation. You can give someone the best hammer in the world, yet that doesn't guarantee they'll build a house, does it? NTT Data and NVIDIA have also announced a collaboration aimed at simplifying the adoption of artificial intelligence for businesses. In essence, they are offering a pre-packaged platform designed to make it easier for companies to implement and scale artificial intelligence solutions. What this entails is the integration of NVIDIA's powerful hardware specifically their graphics processing units and networking technology with their artificial intelligence Enterprise software suite. This software includes tools designed to streamline the development and deployment of artificial intelligence models. The promise is a system that can operate seamlessly across both cloud and edge computing environments. The significance here lies in the potential to lower the barrier to entry for businesses looking to leverage artificial intelligence. Many organisations struggle with the complexities of integrating artificial intelligence into their existing workflows. This offering aims to provide a more straightforward and scalable solution potentially unlocking the benefits of artificial intelligence for a wider range of companies. This also reflects a broader trend toward the commoditisation of artificial intelligence tools and infrastructure. As the technology matures, there is increasing pressure to move beyond theoretical possibilities and deliver practical, real-world applications. This initiative aligns with that imperative. One might observe of course that the history of technology is littered with "plug and play" solutions that turned out to require a considerable amount of fiddling. Time will tell if this particular offering lives up to its billing yet the underlying principle of simplifying artificial intelligence deployment is certainly a welcome development. After all, the easier something is to use, the more likely people are to use it. OpenAI the outfit behind various generative artificial intelligence systems is now offering a service called Frontier which aims to connect different business software systems. Now, the software world is full of acronyms, as a result let's unpack this a bit. Many businesses rely on what's called Software as a Service, or SaaS. This usually means paying a subscription fee to use someone else's software over the internet. Frontier proposes to act as a kind of universal translator, allowing different systems to talk to each other more easily. Think of it as a digital interpreter sitting between your customer relationship management system your data warehouse and all the other pieces of software that a modern business uses. The potential impact here is significant. If Frontier works as advertised, businesses might need fewer separate software subscriptions. Instead of paying for a dozen different tools that don't quite integrate they could use Frontier to make their existing systems work together more efficiently. This could lead to considerable cost savings, which is always of interest to those watching the bottom line. It also puts OpenAI in direct competition with the established SaaS vendors who naturally have a vested interest in maintaining the status quo. Of course, the big question is whether Frontier can actually deliver on its ambitious promises. Connecting disparate systems is notoriously difficult and there's always the risk that it will be another over-hyped solution that fails to live up to expectations. Yet the fact that OpenAI is even attempting this suggests a broader shift in how businesses think about software. Are we heading towards an era of greater integration and interoperability or will the silos of the software world remain firmly in place? It's a bit like the old joke about standards: everyone wants them, yet nobody wants to use someone else's. And when it comes to software, everyone seems to think their way of doing things is the best. LangChain a company known for its tools for building artificial intelligence applications has released something called "Deep Agents." Essentially it's a software toolkit designed to help artificial intelligence systems manage more complicated tasks. To unpack that a bit most artificial intelligence agents – think of them as digital assistants – are pretty good at handling simple one-off requests. Yet when faced with something that requires multiple steps remembering previous interactions or keeping track of various pieces of information they often struggle. Deep Agents is meant to address this by providing a structure for these artificial intelligence systems to better manage context and memory. It's like giving them a more organised workspace, as a result they don't get lost in the details. The significance here is that it could make artificial intelligence more useful in real-world scenarios. Imagine a customer service bot that can not only answer basic questions yet also guide a customer through a complex troubleshooting process remembering previous steps and adapting to their specific needs. Or an artificial intelligence assistant that can manage a multi-stage project, keeping track of deadlines, resources, and dependencies. The potential impact is on businesses that want to automate more complex workflows and provide more sophisticated artificial intelligence-powered services. This also touches on a broader theme we see quite often: the move from simple task-oriented artificial intelligence to systems that can handle more nuanced and dynamic interactions. As artificial intelligence becomes more integrated into our daily lives the ability to manage context and adapt to changing circumstances will become increasingly important. Of course, whether Deep Agents actually delivers on its promise remains to be seen. We've heard similar claims before, and the devil is always in the details. Yet if it works as advertised, it could be a significant step forward in making artificial intelligence more practical and useful. And it does seem we're moving toward a world of digital assistants that are not just helpful yet perhaps a little too deeply involved in our affairs. I'm not entirely sure I want my digital assistant knowing quite as a result much about me. Volcengine a company I confess I hadn't heard of until this morning has released an open-source tool called OpenViking designed to help artificial intelligence agents manage information more effectively. Now, the phrase "context database" might sound a touch opaque, yet the core idea is fairly straightforward. Instead of treating the information an artificial intelligence uses as a jumble of text OpenViking organises it more like a computer's file system with folders and files. The hope is that this structure will make it easier for artificial intelligence to find and use the right information at the right time leading to better decisions and more natural interactions. The potential impact is significant. If artificial intelligence agents can manage context more efficiently they could become more reliable and useful in a variety of applications from customer service to complex problem-solving. It also represents a shift in how artificial intelligence systems are designed moving away from simply throwing vast amounts of data at the problem and towards a more structured approach to information management. This echoes the increasing emphasis on explainability and transparency ensuring artificial intelligence systems aren't simply black boxes churning out answers yet can actually show their working. Of course, open-source doesn't automatically equal good. The project's success will depend on whether developers actually adopt and contribute to it. There's a certain irony in using a file system metaphor for something intended to operate autonomously one imagines a near future where artificial intelligence agents complain bitterly about disk quotas and the perils of accidentally deleting important files. Still, it's an interesting development, and one worth keeping an eye on as artificial intelligence continues to evolve. For now, it offers a potentially useful tool in a rapidly changing landscape. It's all very well building these things, yet someone has to teach them how to tidy up after themselves, doesn't it? And finally, Garry Tan, a figure well-known in Silicon Valley, has released an open-source coding toolkit called gstack. The proposition is that it makes using artificial intelligence assistance in coding a more structured and, dare we say, a reliable process. The toolkit essentially divides coding into distinct phases: planning, review, release, quality assurance, and as a result on. By isolating each of these, the thinking goes, it becomes easier to manage and less prone to error. At its heart gstack uses the Claude Code model packaging it into eight specific skills each aimed at a different part of the coding workflow. A persistent browser runtime is designed to keep everything running smoothly. The idea here is not entirely new. For years software engineers have sought ways to modularise their work to avoid the chaos that can arise when juggling multiple tasks simultaneously. This toolkit is simply applying that principle, with the added twist of integrating artificial intelligence more directly into each stage. Of course, the value of this approach will depend on its practical application. The tech landscape is littered with tools that promised efficiency yet ended up adding layers of complexity. If it does deliver, it could make developers more productive and reduce the risk of errors. It's also open source, which means it can be modified and improved by the community. It's worth noting that the separation of tasks is becoming increasingly common in various fields often driven by the need for greater control and accountability. Whether this toolkit truly delivers on its promise of streamlining coding remains to be seen. It may well be that developers find it adds yet another layer of abstraction to an already complex process. One can imagine a scenario where the tool itself becomes the problem requiring as much maintenance as the code it's supposed to simplify. Still it's an interesting experiment and one that will no doubt be closely watched by those keen to find ways to make coding both more efficient and more reliable. It's a bit like those recipe kits that promise to make cooking easier, yet end up requiring more washing up. Yet the real question, of course, is not just about building tools, yet about how we control these systems. There's been some discussion this week about how businesses are handling the increasing use of artificial intelligence agents within their operations. These agents, which automate tasks such as data queries and workflow management, are becoming more deeply embedded in production processes. The problem, it seems, lies in how these agents are authenticated. Many rely on shared credentials or static API keys, which essentially means they lack distinct identities within corporate systems. In practice this is akin to giving multiple employees access to the same master key without knowing who used it when or why. This creates significant governance risks. If an artificial intelligence agent makes an error, or worse, acts maliciously, tracing the action back to its source becomes incredibly difficult. The lines of responsibility become blurred, which can lead to compliance issues and security breaches. The proposed solution is an "

Another week, another deluge. Sorting the signal from the noise requires, I think, a certain… tenacity. If you'd like a curated dose of artificial intelligence news delivered to your inbox each day, visit jonathan-harris dot online. One email, that's all. And if you're looking for a bit more depth on a specific topic my book "The artificial intelligence Music Revolution: Creativity and Controversy" is available at books dot jonathan-harris dot online slash ai dash music. For those who prefer analysis to advertising. That's your lot for this week's Turing's Torch. If you want the daily brief, head to jonathan-harris dot online. Same time next week — try not to believe the press releases.