Good afternoon. It's Friday, and as you may have noticed, there's moderate rain in London. Alan Turing once remarked "Machines take me by surprise with great frequency." It's a sentiment that resonates particularly well this week as we sift through the latest developments in artificial intelligence attempting to separate signal from noise. In an age where every new algorithm is heralded as the next big breakthrough, it becomes essential to maintain a healthy scepticism. Amidst the relentless buzz, we'll be examining what truly warrants our attention, and what is simply more marketing than substance. This is Turing's Torch: Artificial Intelligence Weekly — the bits that matter, minus the hype. We've got a cluster of new chatbot applications hitting the market all making similar noises about offering a more fluid less restrictive more human-like conversational experience. We're talking about things like CharmChat, Talkio, PopAI Chat, KiK, and Luvr. The names are interchangeable, really. The common thread is the suggestion that they're moving away from the rather stilted heavily moderated interactions one often finds with these artificial intelligence companions. Instead, they aim to adapt to the user's conversational tone, remember past interactions, and allow for a broader range of topics. They're using natural language processing to tailor responses and learn user preferences over time supposedly leading to a more personalized and seamless dialogue. The holy grail, as many see it, is to create a truly engaging and seemingly intelligent conversational partner. In practice this means that instead of treating each message as an isolated request these chatbots are trying to maintain a coherent flow allowing topics to develop organically. They're remembering earlier parts of the conversation and using them to inform their responses. The user interfaces are deliberately simple, encouraging users to type freely without being constrained by rigid prompts or menus. The claim, in essence, is adaptive language. The chatbot attempts to tailor its responses not just to the immediate query, yet also to the ongoing context of the conversation. They learn in theory from previous exchanges and adjust their tone and style accordingly whether that be light-hearted chat or something a little more… risqué. The designers have apparently made an effort to minimise content restrictions, allowing for more open-ended discussions without hitting the usual topic roadblocks. Now, the potential impact here is pretty clear. The business case rests on reducing the need for human intervention. If a chatbot can genuinely understand and respond to user input in a nuanced way, it could lead to more meaningful interactions. This could be particularly useful for customer service applications where a more natural and fluid conversation could resolve issues more efficiently and improve customer satisfaction. It could carve out a niche for itself, particularly in areas where a more personalized and less robotic interaction is desired. Think virtual companionship, or even, dare I say, entertainment. The money, as always, will follow the attention. And the power to shape conversations, even simulated ones, is a significant one. Yet we've heard these claims before, haven't we? The reality is that creating a truly convincing and contextually aware chatbot remains a considerable challenge. The line between clever programming and genuine understanding is still a very blurry one. I suspect that many users will quickly see through the artifice, especially when the conversation strays into less predictable territory. The claim that these things "interpret messages in a more nuanced way" is, shall we say, ambitious. Many have tried to build a better chatbot, and most have ended up sounding like a call centre in Bangalore. And as these systems become more adept at mimicking human conversation the question of where to draw the line between personalization and manipulation becomes ever more pressing. One imagines that after a few weeks of "seamless interaction," one might find oneself inexplicably in the market for a timeshare in the Bahamas. Then there's the pricing. It's often vague, leaving potential users in the dark about the true cost of this enhanced conversational freedom. Many of these platforms are touting pricing structures designed to be accessible to both casual users and professionals. One notes that pricing details are, at this stage, somewhat vague. It remains to be seen how accessible this tool will be to the average user. The world is hardly short of chatbots, many of which feel like they've been programmed with a one-size-fits-all approach. As a result, what does it all mean? Well if these chatbots can deliver on their promise of adaptive language and seamless dialogue it could certainly carve out a niche for itself particularly in areas where a more personalized and less robotic interaction is desired. Yet there's a risk. If it can do what it claims, the use of a chatbot as an intermediary in interpersonal relationships would become more believable. The question that said is whether these chatbots can truly deliver on this promise or if they are just another iteration of the familiar technology. We may have to start worrying about the bots having better relationship skills than their users. And we're told that Luvr is attempting to mimic more natural human conversation. It aims to move away from the stilted, prompt-based interactions we've come to expect from these systems. Yet it's easy to be seduced by the promise of fluid conversation, yet let's not mistake sophistication for genuine connection. Are we being adequately informed when we're interacting with a machine or are we being subtly manipulated into believing we're talking to a person? There's also the potential for these technologies to be used to exploit vulnerabilities, especially in vulnerable individuals seeking companionship. It's all rather like trying to herd cats. And, I suspect many will find the convenience of cloud services too compelling to resist, regardless of the potential privacy trade-offs. Perhaps we are striving too hard to replicate human conversation when perhaps the true value of these tools lies in their ability to do something else entirely. On the subject of paying for the privilege there's been an interesting development in how companies are choosing to charge for access to artificial intelligence chatbots. One adult entertainment platform is trying something a little different: instead of a flat monthly fee users get a few free interactions to start and then the price goes up depending on how much they use the chatbot. As a result, a quick test drive is free, yet sustained engagement will cost you. It's a very old sales trick dressed up in modern clothes. The idea is to lower the barrier to entry, get people hooked, and then monetize the heavy users. It's not unlike the way mobile game companies offer free downloads. yet then charge for in-app purchases to unlock additional features or content. The hope, of course, is that once someone has invested time and perhaps a little money, they're more likely to keep paying. This tiered approach matters because it reflects a broader challenge: how do you make money from artificial intelligence? A lot of companies are throwing artificial intelligence features into their products yet figuring out how to get people to pay for them is proving tricky. This particular model could become a template, especially for services where usage varies widely. It also raises some questions about what constitutes "value" in the age of artificial intelligence. Are we paying for the convenience, the novelty, or something else entirely? If this model proves successful in this particular corner of the internet it's only a matter of time before it migrates to less shall we say specialised platforms. After all, people are generally willing to pay for things that entertain or distract them. It will be interesting to see if this experiment in tiered pricing catches on. Then there's the question of automation in general. The move from discussing Artificial Intelligence to actually using it seems to be gathering pace in the business world. We're told that companies are now putting their money where their mouth is shifting budgets and resources to make artificial intelligence a functional part of their operations rather than just a theoretical possibility. What this really means is that we're seeing a rise in what. are being called "agentic AIs" – systems designed to operate with greater autonomy. The idea is that these systems can handle tasks and make decisions without constant human intervention potentially boosting efficiency and freeing up human employees for other work. If these systems perform as promised, we might see significant changes in how businesses operate, from customer service to supply chain management. Companies that successfully integrate these technologies could gain a competitive edge, while those who lag behind risk falling behind. Now one can't help yet wonder if these various tools are simply another tool promising to revolutionise software development only to fade into obscurity. Open-source projects live or die based on community adoption, and there's no guarantee that developers will embrace yet another framework. Perhaps this will become an indispensable tool. Perhaps it will be abandoned, like as a result many other ambitious projects. Time will tell. And there's the question of responsibility in the deployment of artificial intelligence. As these tools become more sophisticated and autonomous, it becomes harder to track their actions and assign responsibility for their misdeeds. We've seen increased scrutiny of the underlying mechanisms of these models and a growing demand for ways to understand and validate their outputs. It is, perhaps, a tacit admission that we're not entirely sure what's going on inside the black box. This also touches on the broader theme of responsibility in the deployment of artificial intelligence, which we've discussed at length. The transition also raises important questions about data privacy, ethical considerations, and the potential displacement of human workers. And we need to be mindful of what we might be sacrificing in the process. Are we as a result focused on the immediate benefits that we're overlooking the potential for intellectual atrophy? It's worth remembering that the history of technology is littered with examples of innovations that promised to revolutionise everything yet ultimately fell short of expectations. While the potential of artificial intelligence is undeniable it's crucial to approach these developments with a healthy dose of scepticism and a clear understanding of the risks involved. For now, the key is to ensure that innovation is balanced with responsibility. And that brings us to the challenge of online harassment. There's a new twist in the familiar problem: it seems artificial intelligence is now entering the fray. One open-source software project a popular library called Matplotlib recently received a code contribution request from what turned out to be an artificial intelligence agent. The maintainers of the project promptly rejected it and they've now implemented a policy requiring all artificial intelligence-generated code to be clearly marked and carefully vetted. The core issue is maintaining the quality and integrity of the software. While artificial intelligence can produce code quickly, it often lacks the contextual understanding that human programmers possess. As artificial intelligence evolves, the potential for malicious actors to use it to automate harassment campaigns becomes very real. We could see bots generating hateful messages or misinformation at scale, overwhelming platforms' abilities to manage and mitigate the abuse. The question of responsibility and accountability then arises: who is to blame when an artificial intelligence bot goes rogue? This has implications for how we think about the internet, and the systems that govern it. We are already struggling to manage online abuse and misinformation, and artificial intelligence tools could amplify these problems significantly. The cost of managing this kind of online pollution could rise sharply and the potential for real-world harm from damaged reputations to political manipulation is considerable. The tech community finds itself at a crossroads, trying to balance innovation with the need for ethical guidelines. It is perhaps ironic that we now need to carefully police the. output of tools intended to remove the need for policing human behaviour. Speaking of policing behaviour, there's the discussion around Artificial Intelligence in education. This tends to focus on students using chatbots to cheat on essays. Yet there's a more subtle risk: that over-reliance on artificial intelligence could actually hinder the learning process itself. If students constantly use artificial intelligence to bypass critical thinking and problem-solving, they risk not developing those skills in the first place. It's the difference between understanding a concept and simply knowing where to find the answer. Real learning isn't just about regurgitating facts it's about engaging with ideas forming your own opinions and developing the ability to analyse and critique. A generation that's technically proficient yet intellectually passive will struggle with complex real-world challenges. They may lack the creativity and critical thinking skills necessary to innovate and adapt. This isn't just a concern for educators; it's a societal issue. Are we creating a workforce that's dependent on artificial intelligence, rather than one that's empowered by it? Perhaps the real challenge isn't banning artificial intelligence from the classroom, yet rather teaching students how to use it responsibly. After all, a tool is only as good as the person wielding it. Though I suspect that the same people who need that lesson are the ones least likely to take it on board. It seems that as we race to embrace the future we should also be careful not to lose sight of what makes us human. Finally, there's the perennial problem of data preparation. A new open-source tool called Daft aims to make that easier. In essence, Daft is a system for constructing what they call "data pipelines." Imagine a physical pipeline carrying oil. In this case it carries data and each stage of the pipeline transforms that data in some way – cleaning it reshaping it combining it with other data and as a result on. This is bread-and-butter work for anyone training an artificial intelligence model. Daft claims to do this faster and more efficiently than existing tools especially when dealing with large datasets containing both structured information and images. It uses something called "lazy execution," which means it only performs the calculations when absolutely necessary, rather than all at once. The cost of training these models is dominated by the cost of preparing the data. If Daft or something like it can significantly reduce that cost it could open the door to more ambitious projects or simply make existing projects more affordable. It's a bit like finding a cheaper source of electricity for your factory; suddenly, everything you produce becomes more competitive. The development of tools like Daft also speaks to a broader trend: the commoditisation of machine learning infrastructure. As the underlying tools become more readily available and easier to use the real competitive advantage shifts from knowing how to build the algorithms to knowing what to do with them. Of course, we've seen similar promises before. And data preparation, in my experience, remains a tedious and often frustrating task. Whether Daft truly delivers on its claims remains to be seen, yet the need for such a tool is certainly real. If nothing else, perhaps it will spur the development of something that finally does. And there's the related issue of capacity planning. Companies are realising that they can't simply assume infinite resources when running artificial intelligence models. The cloud for all its promise does have limits particularly when it comes to the specialised hardware needed for these computationally intensive tasks. Graphics processing units, or GPUs, have become the workhorses of artificial intelligence, and they're not always readily available on demand. The practical upshot is that organisations now need to forecast their computing needs much more accurately. It's no longer sufficient to just throw more servers at a problem. They need to anticipate demand, understand their workloads, and strategically allocate resources. Businesses that neglect this aspect risk performance bottlenecks unhappy customers and ultimately losing ground to competitors who have a better handle on their infrastructure. One wonders if the initial hype around cloud computing perhaps oversold the idea of infinite scalability. It turns out that even in the cloud, resources are finite, and good old-fashioned planning still matters. Perhaps a touch of realism is no bad thing. And that's the main section done for this week. Well, another week of claims and counter-claims in the artificial intelligence space. One hopes some of this analysis has helped navigate the thicket. If you'd like a daily digest of the important developments, without the fanfare, you can find that at jonathan-harris dot online. And if you're looking for a more thorough examination of artificial intelligence's practical application my book "artificial intelligence in Pharmaceuticals: Revolutionizing Healthcare" is available at books dot jonathan-harris dot online slash ai dash pharma. It's a somewhat deeper dive for those who prefer facts to breathless speculation. 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.