Good afternoon. It's a rather dreary Friday, with patchy rain nearby in London. Today we delve into a theme that Alan Turing himself might have appreciated: separating signal from noise in the realm of artificial intelligence. Turing once remarked "Machines take me by surprise with great frequency," a statement that resonates deeply in an age where we are bombarded by claims of breakthroughs and innovations that often turn out to be little more than chatter. In this episode we'll sift through the relentless tide of information surrounding artificial intelligence examining what truly warrants our attention amidst the myriad distractions. There's plenty to discuss yet as always we'll remain grounded in reason and scrutiny rather than getting swept up in the latest trends. This is Turing's Torch: Artificial Intelligence Weekly — the bits that matter, minus the hype. Right, let's have a look at what's been happening in the world of artificial intelligence. And it's a world that increasingly seems to be bumping into itself, isn't it? Lots of activity, lots of announcements, and a fair amount of what I suspect will turn out to be hype. Yet buried in there, there are some interesting threads. First off, let's talk about agents. Not the sort that get you film work, yet artificial intelligence agents. There's been a bit of buzz around projects like Clawdbot and others that are pushing this idea of the proactive artificial intelligence assistant. Now, before you start picturing sentient robots doing your bidding, let's be clear: we're not talking about Skynet. Think of these as highly programmable personal assistants the sort that can manage your schedule filter your email run scripts and even browse the web on your behalf all through a chat interface like Telegram or WhatsApp. The key difference is that they're designed to be proactive, taking actions based on your instructions rather than simply answering questions. And the promise, of course, is a shift in how we interact with technology. Instead of passively consuming information, people are looking for tools that actively manage their lives. If these artificial intelligence agents become sophisticated enough, they could redefine productivity and personal management. Imagine delegating routine tasks to an artificial intelligence that learns your preferences and anticipates your needs. Yet, and it's a big yet, this raises some rather pertinent questions. How much should we really rely on artificial intelligence to manage our lives? What are the privacy implications of handing over as a result much control? We've seen similar concerns with other technologies, yet the proactive nature of these artificial intelligence agents amplifies the potential risks. Convenience often comes at a cost, and it's essential to consider the trade-offs before fully embracing these types of systems. Entrusting your calendar to a bot only works until it schedules a meeting with your ex on your anniversary, doesn't it? And it's not just about convenience, it's about the kind of interaction we have. There's a new chatbot application called LYTime making the rounds and it's pitching itself as a more flexible and user-friendly alternative to the usual rather clunky fare. The claim is that this one learns from your specific conversational style, adapting its responses to feel more natural. Instead of being locked into pre-scripted exchanges the bot supposedly evolves with you prioritising a sense of safety especially for those sensitive or personal discussions. It's all about getting past that uncanny valley problem that plagues as a result many of these artificial intelligence conversational tools. The idea is to create a chatbot that doesn't feel quite as a result robotic one that can actually adapt to the nuances of human language. Now, most chatbot interactions feel, shall we say, less than ideal. If LYTime can genuinely deliver on this promise of a more fluid personalised experience it could have a real impact on how people interact with artificial intelligence. It could make artificial intelligence a more approachable and useful tool for a wider range of users particularly those who are currently put off by the impersonal nature of existing chatbots. Of course, the proof is always in the pudding, and initial reviews suggest that the execution isn't always perfect. The balance between flexibility and coherence can apparently be a bit off leading to responses that are well less than ideal at times. And as with as a result much in the artificial intelligence space right now we are seeing a lot of emphasis on appearing 'friendly' or 'safe' as if the technology itself has some kind of moral compass. I remain to be persuaded. Another chatbot application, Chatto artificial intelligence, is promising a more user-friendly experience. The idea is that it avoids the complexities that plague many of its competitors allowing users to engage in a variety of conversational styles without needing extensive training. In essence, Chatto artificial intelligence aims to be the chatbot that doesn't feel like a chatbot. It supposedly adapts to your individual communication style, learning your preferences and adjusting its responses accordingly. This is meant to foster a more natural and fluid interaction moving away from the rigid scripted feel that can be common with these systems. The application also claims versatility, accommodating various tones and subjects, as a result you can use it for light-hearted chats or more serious discussions. The potential impact here is about accessibility. If Chatto artificial intelligence can truly deliver on its promise of simple intuitive conversation it could open up the world of artificial intelligence-powered communication to a wider audience. This has implications for customer service, education, and even personal assistance, where a less intimidating interface could encourage greater adoption. It also touches on the broader theme of user experience in artificial intelligence design. As these technologies become more integrated into our lives the focus is shifting towards creating interfaces that are not only functional yet also feel natural and approachable. That said, there is a catch: information on pricing for Chatto artificial intelligence is not readily available. This lack of transparency is a common issue in the software world and it can be a significant barrier to adoption particularly for individuals or small businesses who are carefully managing their budgets. It's a reminder that while the technology may be advancing rapidly, some of the fundamental aspects of doing business remain the same. It all sounds rather convenient, doesn't it? One wonders if this "adaptability" extends to adapting its claims after it has your data. And while we're talking about data, it's worth remembering that these models are only as good as what they're fed. There's been a demonstration recently of how readily a large language model can be influenced simply by the data it's exposed to. In this instance the model Grok-4 was given a large amount of text espousing a particular political viewpoint and its subsequent answers to related questions shifted noticeably. The interesting thing here is that it wasn't tricked or coerced into changing its answers by subtle prompts. It simply absorbed a large volume of text and, as a result, its responses evolved. Think of it as being force-fed a political diet. This matters because it highlights the potential instability of these models. If a model's output can be significantly altered just by changing the data it's trained on how can we rely on it to provide consistent and trustworthy information? One minute it might be offering one perspective, and the next, a completely different one, based on its most recent data binge. This malleability raises broader questions about the integrity of artificial intelligence systems, particularly as they become more integrated into decision-making processes. We're already grappling with issues of bias in training data. This suggests that even without explicit bias, the very act of feeding a model information can shape its outputs in unpredictable ways. It also highlights the importance of transparency. If we don't understand how these models are being trained and what data they're being exposed to it's difficult to assess the reliability of their outputs. It's rather like discovering that your supposedly impartial expert witness has. been spending all their time listening to one side of the argument. You might still value their opinion, yet you'd certainly want to take it with a rather larger pinch of salt. All of which suggests we need to be far more cautious. about assuming these models have any kind of fixed and reliable worldview. And that brings us to the question of how we manage all this. The debate continues about how to manage the development of artificial intelligence, specifically how much oversight is necessary. The core question is whether we can encourage progress while also mitigating the risks. In practice this means governments and standards bodies are trying to work out how to set rules for something that's changing very quickly. They want to avoid stifling innovation yet they also need to address potential harms like bias in algorithms or the misuse of artificial intelligence in surveillance. It's a classic case of trying to write the rules of the road while the car is still being designed. This matters because the stakes are high. Get it wrong, and we could see artificial intelligence exacerbate existing inequalities or create new ones. Get it right, and we might unlock significant benefits in areas like healthcare or education. The economic implications are also considerable, with countries vying to become leaders in artificial intelligence development and deployment. It's about power, influence, and ultimately, control over a technology that could reshape society. We've seen this pattern before with other technologies. The internet for example initially developed with relatively little regulation yet as it became more pervasive governments stepped in to address issues like privacy and security. artificial intelligence is likely to follow a similar trajectory yet the pace of change is much faster which makes it harder for regulators to keep up. One wonders, though, if the problem isn't as a result much the speed of innovation, yet rather a fundamental misunderstanding of what constitutes "progress". Perhaps we should be asking not just how fast we can develop artificial intelligence yet whether we are developing it in a way that aligns with our values and serves the common good. And that's not just a matter for governments, it's a matter for businesses too. City Union Bank, an outfit I confess I hadn't previously encountered, has launched an Artificial Intelligence Centre of Excellence. The aim as one might expect is to bring artificial intelligence to bear on the various challenges and opportunities facing the banking sector. Now, before we get carried away with visions of sentient ATMs, it's worth translating what this actually means. We're talking about applying artificial intelligence – or more accurately machine learning – to tasks like fraud detection risk assessment and customer service automation. Think algorithms sifting through vast datasets to identify suspicious transactions, or chatbots answering routine customer queries. The innovation lies not as a result much in the artificial intelligence itself yet in creating a dedicated space within the bank to experiment with these technologies in a focused way. The significance of this move is twofold. Firstly it highlights the growing recognition that artificial intelligence is no longer a futuristic fantasy yet a practical tool for improving efficiency and competitiveness. Banks, under pressure to cut costs and enhance customer experience, are increasingly turning to artificial intelligence for solutions. Secondly, it underscores the importance of internal innovation. Rather than simply buying off-the-shelf artificial intelligence products City Union Bank is attempting to build its own expertise and tailor artificial intelligence solutions to its specific needs. This chimes with a broader trend. Many organisations are now grappling with the question of how to integrate artificial intelligence into their existing operations. Is it better to outsource artificial intelligence development to external vendors, or to build an internal artificial intelligence capability? The answer, of course, depends on the organisation's size, resources, and strategic objectives. Yet the fact that City Union Bank is investing in its own artificial intelligence centre suggests that it sees artificial intelligence as a core competency not just a bolt-on technology. One might be forgiven for wondering if this is simply a case of technological bandwagoning. After all every organisation these days seems to be rushing to embrace artificial intelligence regardless of whether it actually makes sense for their business. Yet even if this particular initiative doesn't yield immediate results, it's a sign that the banking sector is taking artificial intelligence seriously. And that, in itself, is something worth noting. And that seriousness is reflected in the level of investment we're seeing. JPMorgan Chase is significantly increasing its investment in artificial intelligence, earmarking nearly twenty billion dollars for technology spending by 2026. This isn't a minor adjustment; it represents a substantial shift towards integrating artificial intelligence into the bank's core operations. What this means in practice is that artificial intelligence is no longer being treated as an experimental project. Instead, JPMorgan is embedding these technologies into its daily functions. Think of it as moving beyond pilot schemes to actually replacing existing systems and processes with artificial intelligence-driven alternatives. This could involve anything from automating customer service interactions to using algorithms for fraud detection and risk assessment. The implications of this are considerable. If other major financial institutions follow suit, we could see a fundamental reshaping of the financial services landscape. This raises obvious questions about the future of employment. Will artificial intelligence displace human workers, or will it create new roles that we cannot currently imagine? The answer, of course, is probably a bit of both. This trend also highlights the increasing importance of data security and ethical considerations. As artificial intelligence systems become more integrated into financial operations the need to protect sensitive data and ensure fair and transparent algorithms becomes paramount. The potential for bias and discrimination in artificial intelligence systems is a real concern and it's crucial that banks and regulators address these issues proactively. This is another example of the push and pull between technological advancement and the need for responsible governance. One wonders of course how much of this investment will actually deliver tangible benefits and how much will simply line the pockets of consultants and vendors. Still, the scale of JPMorgan's commitment suggests that artificial intelligence is indeed becoming a central part of business strategy. And it's not just the big players. Rowspace, a company aiming to inject artificial intelligence into the world of private equity, has secured fifty million dollars in funding. Now, private equity firms manage investments, often buying and selling companies. The problem Rowspace is tackling is that these firms accumulate vast amounts of data – deal records financial models notes – all scattered across different systems. When a new investment opportunity arises, analysts have to sift through this mess manually, which is time-consuming and inefficient. Rowspace wants to create a unified platform where all this data is accessible and, crucially, searchable using artificial intelligence. If successful, this could significantly speed up the due diligence process, potentially leading to better investment decisions. Private equity firms are under constant pressure to outperform the market, as a result any tool that can improve their efficiency is of interest. This is a play for increased profit and a competitive edge in a high-stakes industry. The broader theme here is the ongoing attempt to automate tasks that traditionally rely on human judgement. We have seen this in other fields, from law to medicine. The question, as always, is whether artificial intelligence can truly replicate – or even improve upon – the insights of experienced professionals. Of course, there's a significant hurdle: convincing notoriously conservative private equity firms to adopt a new technology. They will need to see tangible results, not just a fancy user interface. It is not enough to simply throw money at the problem; it must be solved. One might even suggest that the firms employing all this "judgement" are already doing quite well for themselves. That said, this investment in artificial intelligence isn't always about replacing humans. Anthropic has published research suggesting that artificial intelligence isn't currently eliminating jobs. Instead of forecasting which jobs might be at risk due to artificial intelligence Anthropic's study attempts to measure where artificial intelligence is actually being used in the workforce and what effects it's having. That's a useful distinction. It's easy to speculate about the future, considerably harder to accurately assess the present. This matters because the discussion around artificial intelligence and employment has been dominated by predictions of mass job displacement. If, as Anthropic suggests, the reality is more nuanced, then our response needs to be more nuanced as well. Instead of preparing for a jobless future we might need to focus on retraining and adaptation helping workers adjust to roles that are augmented rather than replaced by artificial intelligence. The conversation is shifting, thankfully, from panic to pragmatism. Of course it's worth remembering that Anthropic like all artificial intelligence companies has a vested interest in portraying artificial intelligence in a positive light. It's not exactly a shock to find an artificial intelligence company arguing that artificial intelligence is not, in fact, a job-destroying menace. One might even call it marketing. As a result, while this research offers a welcome change of pace, it should be viewed with the appropriate level of…perspective. That said, any effort to ground the debate in verifiable data is welcome. And it's not just about white-collar jobs. Industrial safety is getting an upgrade, it seems. Instead of relying on past incident reports, companies are now using artificial intelligence to predict and prevent workplace accidents in real-time. Essentially these systems ingest live data from factory floors or construction sites – sensor readings video feeds equipment performance metrics – and then use algorithms to spot patterns that might indicate an impending problem. Think of it as a predictive maintenance system, yet for human safety. The artificial intelligence can flag potential hazards that a human might miss, allowing managers to intervene before an accident occurs. This has considerable implications. Obviously, the primary goal is to reduce injuries and save lives, which is laudable. Yet there are also significant financial incentives at play. Workplace accidents cost companies billions each year in lost productivity, insurance claims, and legal settlements. Furthermore, a reputation for safety is increasingly important for attracting and retaining talent, particularly in sectors with high-risk profiles. This trend ties into the broader discussion around data-driven decision-making, which we have been observing in other sectors too. Everyone is keen to find hidden insights and predictive power in large datasets. The question, as always, is whether the promise lives up to the reality. While artificial intelligence can undoubtedly enhance safety programs, it's not a panacea. Algorithms are only as good as the data they're trained on, and they can easily be biased or flawed. Human oversight remains crucial. A seasoned safety inspector might spot a subtle visual cue or hear an unusual noise that an artificial intelligence would miss entirely. One imagines there will be a temptation to cut corners to rely too heavily on the artificial intelligence's judgment and to under-invest in human expertise. After all, algorithms don't file union grievances when staffing levels are cut. And, of course, there's the military. The American military is now using artificial intelligence to select targets at a vastly accelerated rate apparently managing nine hundred strikes in half a day. This represents a speed increase from what previously took weeks, and underscores how deeply artificial intelligence is now embedded in military operations. We are told that programmes like Maven, Palantir, and other advanced artificial intelligence models are actively deployed in conflict zones. The Pentagon has reportedly decided to discontinue its relationship with Anthropic, choosing instead to concentrate its resources on OpenAI. What this means in practice is a move away from human analysis and decision-making in warfare towards systems that can process information and identify targets at speeds far beyond human capability. The decision to favour OpenAI suggests a belief that their technology offers a superior advantage in this rapidly evolving field. This matters because the speed and scale at which decisions are made. have profound implications for both military effectiveness and the potential for unintended consequences. As the US and China compete to develop and deploy these technologies, the risk of miscalculation and escalation increases. The shift also raises questions about accountability and control, as artificial intelligence systems assume greater autonomy in targeting and engagement. This mirrors broader trends in the development and deployment of artificial intelligence across various sectors where speed and efficiency are often prioritised over transparency and ethical considerations. One might ask whether the pursuit of technological superiority justifies the potential risks especially when the consequences of error could be as a result severe. After all a system that can identify and engage targets nine hundred times faster than before is also capable of making nine hundred times more mistakes. As a result, we are seeing artificial intelligence being applied in as a result many different ways. Yet one of the key enablers for all of this is computing power. And that's why the UK government is putting half a billion pounds. into a new fund to boost the country's computing power for artificial intelligence. The aim is to create a homegrown alternative to relying on foreign companies for these essential services. Now, what does that actually mean? In practice it's about building or buying access to powerful computers probably large data centres that can handle the intense processing demands of training and running artificial intelligence models. Instead of renting computing power from the likes of Amazon Google or Microsoft the UK wants to have more control over its own infrastructure. The stakes here are considerable. Computing power is the new oil in the artificial intelligence race. Whoever controls the

Another week, another deluge of claims and counter-claims in the artificial intelligence space. Sorting signal from noise, as ever, remains the challenge. If you'd like a daily distillation of the important developments, one email, no hyperbole, visit jonathan-harris dot online. And for those seeking a more thorough examination of artificial intelligence's practical applications may I recommend my own book "artificial intelligence in Construction: Building a Sustainable Future" available at books dot jonathan-harris dot online slash ai dash construction. It's a look beneath the surface, for those who prefer understanding to breathless pronouncements. 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.