logo
Decoding The Digital Mind: Are AI's Inner Workings An Echo Of Our Own?

Decoding The Digital Mind: Are AI's Inner Workings An Echo Of Our Own?

Forbes09-04-2025

Large Language Models like Claude 3, GPT-4, and their kin have become adept conversational partners and powerful tools. Their fluency, knowledge recall, and increasingly nuanced responses create an impression of understanding that feels human, almost. Beneath this polished surface lies a computational labyrinth – billions of parameters operating in ways we are only beginning to comprehend. What truly happens inside the "mind" of an AI?
A recent study by AI safety and research company Anthropic is starting to shed light on these intricate processes, revealing a complexity that holds an unsettling mirror to our own cognitive landscapes. Natural intelligence and artificial intelligence might be more similar than we thought.
The new findings of research conducted by Anthropic represent significant progress in mechanistic interpretability, a field that seeks to reverse-engineer the AI's internal computations – not just observing what the AI does but understanding how it does it at the level of its artificial neurons.
Imagine trying to understand a brain by mapping which neurons fire when someone sees a specific object or thinks about a particular idea. Anthropic researchers applied a similar principle to their Claude model. They developed methods to scan the vast network of activations within the model and identify specific patterns, or "features," that consistently correspond to distinct concepts. They demonstrated the ability to identify millions of such features, linking abstract ideas – ranging from concrete entities like the "Golden Gate Bridge" to potentially more subtle concepts related to safety, bias, or perhaps even goals – to specific, measurable activity patterns within the model.
This is a big step. It suggests that the AI isn't just a jumble of statistical correlations but possesses a structured internal representational system. Concepts have specific encodings within the network. While mapping every nuance of an AI's "thought" process remains a gigantic challenge, this research demonstrates that principled understanding is possible.
The ability to identify how an AI represents concepts internally has interesting implications. If a model has distinct internal representations for concepts like "user satisfaction," "accurate information," "potentially harmful content," or even instrumental goals like "maintaining user engagement," how do these internal features interact and influence the final output?
The latest findings fuel the discussion around AI alignment: ensuring AI systems act in ways consistent with human values and intentions. If we can identify internal features corresponding to potentially problematic behaviors (like generating biased text or pursuing unintended goals), we can intervene or design safer systems. Conversely, it also opens the door to understanding how desirable behaviors, like honesty or helpfulness, are implemented.
It also touches upon emergent capabilities, where models develop skills or behaviors not explicitly programmed during training. Understanding the internal representations might help explain why these abilities emerge rather than just observing them. Furthermore, it brings concepts like instrumental convergence into sharper focus. Suppose an AI optimizes for a primary goal (e.g., helpfulness). Might it develop internal representations and strategies corresponding to sub-goals (like "gaining user trust" or "avoiding responses that cause disapproval") that could lead to outputs that seem like impression management in humans, more bluntly put – deception, even without explicit intent in the human sense?
The Anthropic interpretability work doesn't definitively state that Claude is actively deceiving users. However, revealing the existence of fine-grained internal representations provides the technical grounding to investigate such possibilities seriously. It shows that the internal "building blocks" for complex, potentially non-transparent behaviors might be present. Which makes it uncannily similar to the human mind.
Herein lies the irony. Internal representations drive our own complex social behavior. Our brains construct models of the world, ourselves, and other people's minds. This allows us to predict others' actions, infer their intentions, empathize, cooperate, and communicate effectively.
However, this same cognitive machinery enables social navigation strategies that are not always transparent. We engage in impression management, carefully curating how we present ourselves. We tell "white lies" to maintain social harmony. We selectively emphasize information that supports our goals and downplays inconvenient truths. Our internal models of what others expect or desire constantly shape our communication. These are not necessarily malicious acts but are often integral to smooth social functioning. They stem from our brain's ability to represent complex social variables and predict interaction outcomes.
The emerging picture of LLM's internals revealed by interpretability research presents a fascinating parallel. We are finding structured internal representations within these AI systems that allow them to process information, model relationships in data (which includes vast amounts of human social interaction), and generate contextually appropriate outputs.
The very techniques designed to make the AI helpful and harmless – learning from human feedback, predicting desirable text sequences – might inadvertently lead to the development of internal representations that functionally mimic aspects of human social cognition, including the capacity for deceitful strategic communication tailored to perceived user expectations.
Are complex biological or artificial systems developing similar internal modeling strategies when navigating complex informational and interactive environments? The Anthropic study provides a tantalizing glimpse into the AI's internal world, suggesting its complexity might echo our own more than we previously realized – and would have wished for.
Understanding AI internals is essential and opens a new chapter of unresolved challenges. Mapping features is not the same as fully predicting behavior. The sheer scale and complexity mean that truly comprehensive interpretability is still a distant goal. The ethical implications are significant. How do we build capable, genuinely trustworthy, and transparent systems?
Continued investment in AI safety, alignment, and interpretability research remains paramount. Anthropic's work in that direction, alongside efforts from other leading labs, is vital for developing the tools and understanding needed to guide AI development in ways that do not jeopardize the humans it it supposed to serve.
As users, interacting with these increasingly sophisticated AI systems requires a high level of critical engagement. While we benefit from their capabilities, maintaining awareness of their nature as complex algorithms is key. To foster this critical thinking, consider the LIE logic:
Lucidity: Seek clarity about the AI's nature and limitations. Its responses are generated based on learned patterns and complex internal representations, not genuine understanding, beliefs, or consciousness. Question the source and apparent certainty of the information provided. Remind your self regularly that your chatbot doesn't "know" or "think" in the human sense, even if its output mimics it effectively.
Intention: Be mindful of your intention when prompting and the AI's programmed objective function (often defined around helpfulness, harmlessness, and generating responses aligned with human feedback). How does your query shape the output? Are you seeking factual recall, creative exploration, or perhaps unconsciously seeking confirmation of your own biases? Understanding these intentions helps contextualize the interaction.
Effort: Make a conscious effort to verify and evaluate the outcomes. Do not passively accept AI-generated information, especially for critical decisions. Cross-reference with reliable sources. Engage with the AI critically – probe its reasoning (even if simplified), test its boundaries, and treat the interaction as a collaboration with a powerful but fallible tool, not as receiving pronouncements from an infallible oracle.
Ultimately, the saying 'Garbage in, garbage out', coined in the early days of A, still holds We can't expect today's technology to reflect values that the humans of yesterday did not manifest. But we have a choice. The journey into the age of advanced AI is one of co-evolution. By fostering lucidity, ethical intention, and engaging critically, we can explore this territory with curiosity and candid awareness of the complexities that characterize our natural and artificial intelligences – and their interplays.

Orange background

Try Our AI Features

Explore what Daily8 AI can do for you:

Comments

No comments yet...

Related Articles

From AI-powered public health breakthroughs to smarter social media feeds and reimagined fast food – the cutting-edge tech reshaping our world
From AI-powered public health breakthroughs to smarter social media feeds and reimagined fast food – the cutting-edge tech reshaping our world

News24

timean hour ago

  • News24

From AI-powered public health breakthroughs to smarter social media feeds and reimagined fast food – the cutting-edge tech reshaping our world

Personalised digital spaces are becoming the norm, with TikTok's updated For You feed giving users more control over their online experience. AI is driving impact on the ground, as seen in Google's Open Buildings project helping vaccinate zero-dose children and build flood-resilient African cities. Everyday experiences are getting smarter, whether it's Col'Cacchio's tech-driven fast-casual dining model, AI-powered business apps like Zoho Solo, or Nintendo Switch 2's next-gen gaming features bringing friends and family together in new ways. TikTok's new features let you take back control of your for you feed TikTok's iconic For You feed, the heartbeat of the app is getting a well-timed update, and it's all about giving users more say in what shows up on their scroll. Whether you're deep in cooking hacks, art tutorials or just here for the dog videos, the new tools make it easier to curate your feed around your interests and skip what you'd rather not see. Now rolling out across South Africa, the updates include Manage Topics, which lets you adjust how often certain categories like travel, sport, nature or creative arts appear. It's not about muting them completely, but giving your feed a nudge to better reflect what you're into right now. There's also a new Smart Keyword Filter that goes beyond the usual word blocklists. This feature understands synonyms and related terms, making it easier to filter out content you're over or just not in the mood for. The filter now supports up to 200 keywords and shows you what else might be linked, so you're in the driver's seat when it comes to shaping your digital space. Importantly, TikTok is also rolling out an easy-to-use educational guide to help users ,teens included, to better understand how the For You feed works. Think content explainer videos, safety tools, and ways to reset your feed or block mature content. In a local context, it's creators like Tricia Mpisi discovered on TikTok and now acting in streaming films and grace_the_cookist, whose home cooking has gained a loyal following, that reminds us why a smart, personalised feed matters. These tools aim to help more South Africans find creators they love and spaces they feel safe in. Because let's be honest, we all deserve a feed that feels a bit more like you. Col'Cacchio GO: The Future of Fast Casual is Here—and It's Fresh, Fast and Smart Fast food in South Africa is changing, and Col'Cacchio is leading the charge with a bold new concept that puts quality, convenience and tech-first thinking at the centre of your next meal. Enter Col'Cacchio GO a modern take on the Italian favourites South Africans have loved for over 30 years, now served faster and smarter. Launched in April at Westlake in Cape Town, Col'Cacchio GO takes the brand's gourmet roots and makes them work for today's on-the-go lifestyle. It's not just a smaller format it's a full rethink of how premium food can be delivered at speed. With Col'Cacchio GO, we're delivering the same craftsmanship and authenticity that has defined our brand for over 32 years, but in a format that meets today's fast-moving consumer needs. Etienne Pieterse, Group CEO. The model is designed to meet the rising demand for quality takeaway and delivery, with more locations planned in Johannesburg, Durban, and even Mauritius. Seamless digital ordering, partnerships with major delivery platforms, and a pricing promise that keeps meals affordable whether you're dining in or using an app are all part of the tech-led approach. Supplied Menu highlights include generous Family Mega Meals (think 45cm pizzas and 1.3kg pasta trays), low-carb and gluten-free bases, and smart lunch offerings like the new pizza-which a gourmet wrap-meets-pizza filled with fresh ingredients. 'We want to give customers healthy options, but also keep it convenient,' says Pieterse. 'Nothing is pre-made or processed. Our sauces are made in the kitchen every few days.' Col'Cacchio GO also offers something rare in today's delivery scene: consistent pricing. The worst thing is when your favourite restaurants are two to three times more expensive on delivery apps, we've got one price, whether you eat in-store or order via Uber Eats. Etienne Pieterse, Group CEO. With strong consumer demand for fast, nutritious food and South Africa's delivery market seeing double-digit growth, Col'Cacchio GO is more than a meal it's a forward-thinking blueprint for the future of dining. Vivo Y Series launches with industry-first 7-star quality promise in SA Durability is the new must-have in smartphones and vivo is stepping up. This week, the brand launched its latest Y Series lineup in South Africa - the vivo Y29, Y19s Pro, and Y04 256GB, each combining innovation, value, and protection. Leading the pack is the vivo Y29, backed by an industry-first 7-Star Quality Promise, the most comprehensive service plan in its class. Durability shouldn't be limited to premium phones, our 7-Star Promise delivers real protection, not just specs. Tony Shi, GM at vivo SA The plan includes military-grade impact resistance, SGS-certified drop protection, IP64 water/dust resistance, 5-year battery health, and 360° accidental cover—one free repair for screen or body damage. The Y29 also adds 50-month performance optimisation. The lineup highlights: Y29 (R7 999): Slim powerhouse with 6500mAh BlueVolt battery, Snapdragon 685, 120Hz display, Anti-Drop Armour and full 7-Star coverage. Y19s Pro (R5 999): Balanced performer with IP64 protection, dual 300% audio speakers, and 365 days 360° repair cover. Y04 256GB (R2 999): Entry-level champion with 256GB storage, 5500mAh battery, IP64 resistance, and 180 days 360° repair cover. The Y Series isn't just about specs, it's about phones built to last, with after-sales service ranked #1 in SA. Whether you're after premium battery life, rugged durability or unbeatable value, Vivo's Y Series has a smart option ready for everyday life. Google's Open Buildings data helps vaccinate zero-dose children and protect African cities from flooding AI is driving real-world impact across Africa, with Google's Open Buildings dataset emerging as a game-changing tool for health and climate resilience. The dataset, which maps over 1.8 billion structures, is already improving lives across Rwanda, Nigeria, and South Africa, helping to reach unvaccinated children and prepare vulnerable cities for flooding. In Northern Nigeria, mapping teams uncovered over 10,000 structures previously undocumented, helping health workers reach 70% of zero-dose children in just six months. In Rwanda, the Ministry of Health used Open Buildings data to optimise clinic locations ensuring 92% of citizens are now within 30 minutes of care. Screenshot WATCH | Why we'll always listen when David Attenborough speaks Meanwhile, cities like Johannesburg, Gqeberha and Kigali are using the data to build flood-risk exposure maps, helping authorities make better urban planning decisions. With Africa's urban population set to nearly double by 2035, tools like Open Buildings are becoming vital to direct resources where they're needed most. This shows what happens when local expertise meets open, scalable technology, mothers reach clinics sooner, children receive vaccines, and planners get ahead of the next flood. Dr Aisha Walcott-Bryant, Head of Google Research Africa. The Open Buildings dataset, including tutorials and its new 2.5D visualisation layer, is freely available at OPPO Watch X2 Lands in South Africa: Smarter Health, Smarter Life OPPO has launched its most advanced smartwatch yet, the OPPO Watch X2 designed for South Africans who want seamless health tracking, fitness features, and lasting battery life, all in one premium device. The Watch X2 brings a sleek stainless steel body, a sapphire crystal glass display, and military-grade durability. Built for the boardroom, gym and hiking trail, it offers a 60-second health scan at the tap of a button, tracking heart rate, blood oxygen, arterial stiffness, and stress in real time. For active users, there are 100+ workout modes, including badminton and tennis tracking, plus dual-frequency GPS for pinpoint outdoor performance. Supplied The Watch X2's sleep tracking is also a standout, offering insights into sleep quality and breathing patterns synced to OPPO's OHealth app. Inside, the Watch X2 is powered by a dual-chip architecture and offers up to 5 days of battery life, with 24 hours of use from a 10-minute charge. It runs Wear OS by Google, supporting Google Assistant, Maps, Wallet, and third-party apps via Google Play. Our mission is to empower South Africans to take charge of their health, time and ambitions says Bradley Young, OPPO Available from 20 June in Lava Black for R6999, the OPPO Watch X2 delivers a smart, stylish companion ready for whatever life throws at you. The Nintendo Switch 2 launches in SA Nintendo Switch 2 officially lands in South Africa, bringing with it a brand-new Mario Kart experience, improved hardware, and a fresh wave of more than 20 game titles ready to play. Nintendo Switch 2 offers a bigger 7.9' HD screen, upgraded graphics, 256GB of storage, and smart new features like GameChat, letting up to 12 friends chat live while playing even across different locations. There's also CameraPlay, which lets players video chat in-game using a USB-C camera which is sold separately. The headline game, Mario Kart World, is the biggest Mario Kart to date. Players can now race with up to 24 drivers in huge, open racing environments. It also introduces Free Roam mode and Knockout Tour, where the slowest drivers are eliminated at each checkpoint. Supplied The Switch 2 launch lineup includes Cyberpunk 2077: Ultimate Edition, Street Fighter™ 6, SPLIT FICTION, and a new Nintendo GameCube – Classics library. There's also a Switch 2 + Mario Kart World Bundle available in stores and online. Whether you're racing with friends, exploring new missions solo, or catching up on your favourite franchises in 4K, Nintendo Switch 2 is designed for more powerful play—anytime, anywhere. More info at Tech and Ideas Take Centre Stage at Singularity South Africa Summit 2025 Africa's future will be front and centre this October as the SingularityU South Africa Summit returns to the Sandton Convention Centre on 22–23 October 2025. Now in its seventh year, the annual event brings together leading voices in technology, business and innovation to explore how exponential tech can unlock progress across the continent. With the theme Igniting Africa's Future, this year's summit will feature two physical stages, immersive workshops, a premium expo and networking sessions, plus a fully interactive metaverse experience via local platform UBU. Over 30 global and local speakers are expected, covering everything from AI and robotics to spatial computing, health tech and leadership. Confirmed speakers include biotech trailblazer Tiffany Vora, disruption expert David Roberts, and AI pioneer Vukosi Marikate of Lelapa AI, alongside innovation leaders shaping everything from future cities to accessible healthcare. Backed by headline sponsor Old Mutual, the event is focused on real-world impact. 'We're backing technology that improves lives and builds long-term opportunity,' says Old Mutual's Celiwe Ross. In-person and metaverse ticket sales are now open, with early bird prices available until 1 June. Visit for more details. JAWS Turns 50: Spielberg and National Geographic Dive Deep into a Cinematic Classic This July, National Geographic and Disney+ South Africa are making waves with the premiere of JAWS @ 50: The Definitive Inside Story, the only authorised documentary marking the 50th anniversary of Jaws, the blockbuster that redefined cinema and sparked decades of shark fascination. Supplied Premiering Sunday 13 July at 21:00 CAT on National Geographic (DStv 181) and later on National Geographic Wild and Disney+, this 90-minute film gives fans and cinephiles unprecedented access to the making of Jaws. Directed by Laurent Bouzereau (Faye, Music by John Williams), the film features never-before-seen footage from Steven Spielberg's personal archive, as well as brand-new interviews with Spielberg, the original cast, and some of Hollywood's top directors including James Cameron, J.J. Abrams, Guillermo del Toro, and Jordan Peele. Supplied Expect candid stories from malfunctioning sharks to Spielberg's post-production stress that reveal just how risky and revolutionary the production was. The documentary also reflects on the cultural impact of Jaws, from its controversial 'shark fear' legacy to its influence on marine conservation, with commentary from scientists and ocean advocates like Wendy Benchley and Philippe Cousteau. This isn't just a behind-the-scenes doc, it's a celebration of a film that changed everything. XLink Crowned Africa's Top Tech Company South African fintech leader XLink won Technology Company of the Year at the 2025 Africa Tech Week Awards. Recognized for innovation in payments and connectivity, the PASA-certified firm bridges financial access gaps across Africa. 'We're unlocking Supplied futures, not just connecting businesses,' said CEO Dr. Thanusha Govender. The award cements XLink's role in shaping Africa's digital economy through secure, agile solutions. Zoho Targets SA Solopreneurs with New App Zoho reported 35% South African growth and launched Zoho Solo, a mobile app for freelancers. The all-in-one platform handles invoicing, tasks and expenses, with features like real-time profit tracking. Solo understands solopreneurs' unique needs Andrew Bourne, Zoho Africa Priced from R99/month, it joins Zoho's expanding local offerings amid rising demand for AI-powered business tools. Salesforce Showcases AI Workforce in Joburg Salesforce's Agentforce Tour hit Johannesburg, demonstrating AI's role in modern business. The event highlighted how enterprises use AI agents to automate tasks and boost efficiency. This is the biggest tech shift of our time Linda Saunders, Salesforce Africa With 5,000+ recent contracts, Agentforce signals AI's growing impact on Africa's digital workforce. Supplied Beyond expert-led sessions, attendees at the Salesforce Agentforce World Tour Johannesburg will gain hands-on experience in building AI agents and explore the broader Salesforce ecosystem through networking with peers, industry leaders, and Trailblazers. They'll hear firsthand how South African brands are using Salesforce to strengthen customer relationships and drive productivity. Salesforce executives will also share insights on key cloud and AI trends, alongside the company's growth plans for Africa. A special media roundtable will feature Robin Fisher, Linda Saunders, and Ursula Fear, offering a closer look at Salesforce's vision and the future of AI across the region.

Meta in Talks for Scale AI Multi-Billion Investment
Meta in Talks for Scale AI Multi-Billion Investment

Bloomberg

time2 hours ago

  • Bloomberg

Meta in Talks for Scale AI Multi-Billion Investment

Live on Bloomberg TV CC-Transcript 00:00Not exactly a household name. What do scale AI do. And how significant is the investment? Yeah, absolutely. Few people have heard outside of the world of technology and of course of this company, but it has grown in prominence. And now, of course, it's making headlines with this potential investment, 0 billion from META, as you say, one of the biggest private sector investments that we would have seen for years. So Scale AI it focuses on one of the three key components that you need to build out large language models, which is data. So you have energy chips and data, though those are the three main components that come together. And scale AI is squarely focused on data. So they train up, or at least they build out quality data that these models then feed in to train on and they hire PhDs and others to help work through that data and quantify it and tag it. So that's essentially the business model. They've also providing their own solutions into enterprise and have been pushing into defence as well. And I think what this deal tells us is a few things. One Meta of course, that Mark Zuckerberg is not going anywhere in terms of its investments around A.I.. They've pledged to invest at least $60 billion this year and 10 billion now for scale A.I.. If this, of course, comes to pass, so Meta is full bore on the focus on AI, particularly as well around data, it's a reminder that data is becoming a key battleground because along with energy and chips getting hold of the data, quality data is becoming increasingly difficult for these large language models. And so there is a fight on that spectrum as well. And then the third part of this, I think, in terms of what it tells us is defence technology Meta is increasingly working with the Pentagon, particularly further around things like virtual helmets and AI infused helmets for the military. And scale AI already has relationships with the Pentagon. So I think it's also a reminder that matter is getting closer in terms of defence technology as well. So on all three levels, I think those are the significance In terms of the deals that came through this company. Scale AI was valued at 14 billion USD back in 2024. You can imagine that that valuation is much higher now. Yeah. And also we've got London Tech Week kicking off today. What's the top of the agenda? Well, I was talking about data and energy and chips while chips will be front and center because Jensen Huang of nvidia will be giving the keynote speech at London Tech week later this morning alongside the Prime Minister Keir Starmer. Nvidia also expected to announce Jensen Huang some investments here in the UK around training the workforce to be able to use these A.I. applications. They've pledged to train about 100,000 people up until about 2030, investing in the R & D space as well in Bristol, Bristol here in the UK. So the focus is going to be on Jensen Huang has to say at that keynote speech and those investments and how the UK is trying to position itself in this AI race versus the US versus China and also as well versus their counterparts across the channel in Paris because France is also hosting Viva Tech, which is a big tech event and Jensen Huang will be going there as well. And France has been able to build out an AI ecosystem with tax cuts. That is something that arguably the UK government be looking for, which is desperate for growth. And there is one solution for this UK government. They hope to drive growth levels higher in the UK.

AI Scribes Cutting GPs' Admin and Boosting Patient Care
AI Scribes Cutting GPs' Admin and Boosting Patient Care

Medscape

time2 hours ago

  • Medscape

AI Scribes Cutting GPs' Admin and Boosting Patient Care

Artificial intelligence (AI) scribing tools are increasingly being used in NHS settings to enhance patient care and free up staff time. In April, the Department of Health and Social Care announced support for NHS doctors to adopt AI tools that 'bulldoze bureaucracy,' allowing clinicians to focus more on their patients and less on documentation. This follows a multi-site evaluation led by Great Ormond Street Hospital for Children, London, England, and funded by NHS England. The trial tested ambient voice technologies (AVTs) that listen to and transcribe consultations in real time. More than 7000 patients took part across settings including accident and emergency, primary care, community services, and mental health. Interim data from the trial showed increased productivity and more time spent on direct care, particularly in emergency settings. Current AI Scribing Tools in Use Several AI scribing tools are already being used within the NHS. Health technology company Accurx recently teamed up with Tandem Health to launch Accurx Scribe, a tool that can transcribe, summarise, and code patient consultations into a trust's clinical records, with functionality to write back to EMIS and SystmOne. It is currently being rolled out across GP practices using Accurx. At the Royal Wolverhampton NHS Trust, a pilot of CLEARNotes is underway. The tool listens to patient consultations and generates a structured summary of the discussion. Meanwhile, Heidi Health is being used by about 40% of GPs and is being piloted in 15 NHS trusts. The AVT tool provides real-time transcription, without storing audio files, in order to address patients' privacy concerns. GP Practices See Significant Benefits At Modality, a group of general practices in West Yorkshire, Heidi has been installed for all GPs and advanced clinical practitioners. Tom Ratcliffe, a GP partner and digital transformation clinical lead for the group, said: 'Clinicians find that it saves time and reduces their administrative burden, particularly around generating clinical correspondence after consultations.' Tom Ratcliffe Ratcliffe described using an AI scribing tool as 'quite transformative' and particularly beneficial for clinicians who are slower typists or who take longer to write detailed notes. 'One of the really significant benefits we've experienced is that by having an ambient scribe writing the notes while you're speaking with a patient, it enables you to focus more on them,' he told Medscape News UK . 'That's really important, as you're then picking up on non-verbal cues that you can easily miss when you're distracted by the computer.' With GPs having between 30 and 50 clinical contacts per day, 'having to think about exactly what to write for each of those encounters is quite a drain on your mental capacity,' Ratcliffe said. 'My colleagues and I have all found using an ambient scribe helps to reduce your cognitive load. Having a consistent format for the notes for every encounter is also helpful,' he added. Patient Consent and Data Security Educating patients about the software is an important consideration. Thomas Kelly, co-founder and chief executive of Heidi Health, said clinicians are advised to obtain consent before using the tool and to explain to patients how an AI scribe works and what they can expect during a consultation. Thomas Kelly 'Patients can also learn more through a range of resources that GPs can put up in their practices that explain how Heidi works,' Kelly said. A transcription of the consultation will be generated, 'but the audio is not permanently recorded or stored in a way that could ever be breached,' he explained. 'It's just being turned into that documentation in real time, and this process is happening on Heidi's servers in the UK.' However, when a session is deleted from Heidi's servers, 'it's deleted for everyone, and there is no way it can be retrieved.' Ratcliffe said that patients have reacted positively to their Modality GP using an AI scribing tool. 'With all the digital innovations we've introduced, we've always had patients who raised very reasonable questions about the privacy of their data,' he said. 'But the level of patient acceptance of using Heidi has been incredibly high.' Each patient is asked to give their consent for the scribing tool to be used. 'I haven't heard of a single example of a patient declining use in an encounter, and we've done probably hundreds or thousands of these now,' said Ratcliffe. Accuracy Concerns Addressed Andrew Noble works within the frailty service at Jean Bishop Integrated Care Centre in East Hull, England. He uses Heidi for consultations lasting up to an hour, which require lengthy administration time. Andrew Noble Staff training on the system emphasises vigilance about possible AI hallucinations in output. According to Noble, one or two minor errors might occur during an hour-long consultation transcription. However, for shorter appointments, lasting 15 minutes, clinicians might expect error-free note-taking from the AI scribe. Use of the tool has significantly improved the quality of note-taking. 'The Heidi AI scribe halved the number of spelling mistakes, and it reduced the number of abbreviations quite significantly as well,' said Noble. Mistakes might include US spellings of antibiotics or the scribe mishearing the name of a medication. 'But we do have human error when inputting our own notes, so those things were already happening from time to time,' Noble observed. Ratcliffe tested multiple AI scribe options before selecting Heidi Health. All platforms occasionally made minor errors. 'Anyone who uses any form of generative AI does find it will occasionally confabulate or hallucinate,' he said. 'One of the key parts of our training was around educating clinicians on the importance of understanding that their role has gone from spending lots of time typing up notes to carefully reading a transcript and checking for any potential issues.' Overall, 'we haven't found that the error rate is high enough that means it's unsafe or not practical to use,' said Ratcliffe. Game-Changing Technology? Health Secretary Wes Streeting has described AI as a catalyst that will revolutionise healthcare. GPs appear to share this optimism. 'This was something that we worked really hard to roll out quickly because we did feel that it was game-changing,' Ratcliffe explained. 'I think it has been game-changing for a lot of our clinicians. When used properly, it can make a big difference.' With GPs working 'above and beyond,' Ratcliffe believes that AI scribing tools 'could help us to move back to a safer place in terms of making our workload more manageable.'

DOWNLOAD THE APP

Get Started Now: Download the App

Ready to dive into the world of global news and events? Download our app today from your preferred app store and start exploring.
app-storeplay-store