Tuesday, January 30, 2024

Have we passed 'Peak Ethics' in AI?

We seem to have passed 'Peak Ethics' with AI. Now that little of any moral consequence has happened we're coming back to the real world. More focus on real utility, applications and use in productivity, teaching, learning. It was all a bit dizzy, rarefied and weird up there... 

It’s sometimes harder, as Scottish poet Norman McCaig wrote, to come down than climb a mountain. Enthusiasm and certainty get you up, one must tred carefully coming back down to the real world.

Remember the letter? No one cares about it now. An Open Letter is the argument from authority, always suspect, and is the very opposite of 'open'. It simply says we're right you're wrong. It's straight up bullying.

After that famous letter, demanding a six months halt, what actually happened? A lot of backtracking and embarrassment by some of the signatories. In all that time GPT4 has remained king of the hill and the world has seen upsides with no real downsides. Italy banning ChatGPT looks like a childish gesture. Yudkowsky and Tegmark now seem a bit boorish, grifters seling Doom. It all seems so Y2Kish, more cult-like, end-of-days, millenarian, than realism. All of those folk who suddenly had AI and Ethics in their titles, seem a bit old-hat, boring and superfluous. Having worked with this technology for many years, I wondered at the time where they all came from, all of those experts in ‘AI’ and ‘Ethics’. I never saw any of them before November 2022. Never saw the projects they were involved in, actual writing, books they’d written. It was a pretty lonely world back then.

Suddenly, an army of arrivistes were seen talking earnestly on panels, running workshops, uttering memes such as ‘stochastic parrot’ with absolutely no idea what that meant or where the phrase came from. Heads of this and that, experts all – within zero practitioner experience, in just months!

All we’ve seen, 14 months in, from the research, is evidence of increased productivity, ideation, creativity and even signs of reason and semantic sophistication. On top of that amazing multimodal capability when we can create images, video, speak to it, it speaks back, less errors, better performance, create our own avatars, chatbots and massive reductions in prices. Not a week passes without something wondrous happening. Deepmind continues to astonish with its Alpha software and research is getting a boost in terms of planning and execution. In healthcare we've seen significant leaps.

Yann Lecun is in charge of AI at Meta. He knows more than anyone on the planet about moderation of content using technology and AI. On Twitter, he made the point that despite GPT-2 and numerous other LLMs having been available in open source for 5 years now, there has been no flood of "extremist synthetic propaganda" that "researchers" warned about? He has a point. This tech has been available since Obama was in office and we saw a couple of deepfakes but hardly overwhelming... why?... AI detection... Sure some gets through and it will happen but it has been conspicuous by its absence.

Lecun’s point is that AI not only has the potential to solve some of our most pressing problems, especially those that pose an existential threat but it can also police itself. Engaging in lots of confirmation bias, negativity and sci-fi levels of speculation is fine but the whole thing got out of hand. I suspect this is because it’s easier to ruminate on ethics, with lots of hand wringing, than get to know and use the technology in real projects.

Now that we’ve calmed down, and had time to try things out, see the potential, the world seems like a better place, less angst, less moralising. Having ridden onto town on their moral high horses, they’ve found that people are not that interested in yet another repetitive report, framework or list of ethical platitudes. They’ve had to tie their high horses up, and get inside the tent with the rest of us – using it, doing it. It was always thus. I have a whole presentation on how such moralisers blow their trumpets at the start of every technical advance from the sundial, writing, books, printing, radio, film, television, jazz, rock music, rap, walkmans, typewriters, photocopiers, computers, internet, social media, smartphones… and now AI.

Tech Doomerism, I’ve realised, is actually a form of advertising, a species of hype, with clickbait examples, binary thinking of good & evil, a liberal dose of anthopomorphising AI, a narrow focus on edge of debate and seeing Sci-Fi as credible predictions. Elon Musk was the perfect example, signs letter demanding stop to AI, exactly 6 months later releases GROK1! 

Saturday, January 27, 2024

GPT4 is not a 'stochastic parrot'

Whenever I hear someone say GenAI is a 'stochastic parrot' my BS meter goes off. They are usually parroting this as a meme they've read on social media, with no reference to the 2021 paper it came from. I often ask if they've read the paper - not one person in education has known of the Bender paper.

It appears that GPT4 is NOT a 'stochastic parrot'. I argued this at BETT two days ago, based on its ability to produce top-down, Wittgensteinian language games. The fact that it can play language games and be good at new ones, was a sign that it can generalise to produce these new skills. This team, I think,
have proven it mathematically, bottom-up.

Geoffrey Hinton agrees.

The model can generate text that it couldn’t possibly have seen in the training data, displaying skills that add up to what some would argue is understanding.

What is really interesting is the other conclsuion one cxan draw from this research.

The authors add that the work doesn’t say anything about the accuracy of what LLMs write as they are by implication 'creative'.

“In fact, it’s arguing for originality,” he said. “These things have never existed in the world’s training corpus. Nobody has ever written this. It has to hallucinate.”

Another implication is that larger numbers of parameters will allow more skills. I am sure we'll see this in 2024.

Thursday, January 25, 2024

Is UX design killing product? 25 ways to make your e-learning totally suck!

Is UX design killing product?

On tech social media and discussions, a huge brouhaha has erupted over UX design. Developers see a new breed of UX designers as pushing the line that ‘look and feel’ out guns ‘features and functionality’. Some have fled Microsoft and Google, as the culture of UX has overwhelmed teams, making progress sclerotic. The designers tend to rotate, the developers remain in post, so they get tired of eager young things turning up with uninformed personal views rather than worked our ideas. I’ve seen this myself and, on reflection, I think this is at least a worrying trend.

In some cases the developers who walked out report projects being taken over by an extremist faction, high on opinion but low on technical knowledge, requesting unfeasible interface features. Developers in Microsoft describe bright young things bringing in Macs, having never used a PC as a working tool, designing tool interfaces that were at best unsuitable, at worst, unusable.

I’ve seen similar things over many years, in many contexts.

The first in online learning, where almost all the discussion with the buyer is about look and feel, not functionality and usability or, more importantly, learning. The problem is often that the buyer doesn’t know anything about design (or at times learning) so makes off the cuff ‘can you make it look like …’ type requests. This is similar to giving video scripts to clients, SMEs or lawyers – where they can massacre the dialogue.

A second is the use of imagery that doesn’t really fit the product. A common example is AI product that has images of those toy 2D clipart robots, or little rocket ships. I’ve seen graphics of pens and ink nibs on generative AI product. When asked to justify the imagery there’s little real justification.

My pet hate is the predominance of cartoon imagery. It’s everywhere, so clearly not the result of matching style or substance to audience. I don’t watch cartoons, most adults don’t. Fine for primary schools but not adult learning. You get this when ‘Edtech’ people who tend see learning in terms of schools and kids, get roles in adult or workplace learning. The workplace is not an infants’ school. The worst are those figures with speech bubbles, where you click on each and the text appears, giving Ahmed. John and Sophie’s views on the subject at hand – really?

In truth the best interfaces for normal (not tool) use, are simple. It is put well by Papert:

Low floor means painfully obvious and simple to use. The two best examples are Google and ChatGPT – one letterbox, type something in and press enter – LOW FLOOR.

High ceiling means you get something great, even amazing, back. You get this with Google and ChatGPT – you don’t get it with a next button on pages of content or a speech bubble or flipping a card on the screen – HIGH CEILING.

Wide walls mean there’s breadth and substance in the response, you get a substantial reward that is what you want or even exceeds expectations – WIDE WALLS.

To be fair good design can also be stymied by developers, project managers, stakeholders and clients. It’s a set of skills. But the uniformity of output has made me suspicious of who holds those skills. 

Another interesting dimension to all this, is the cultural one. The US doesn’t care that much for high end design, - they want cost-effective functionality. In the UK and Europe, design often comes first. Donald Norman wrote about this and it is true.

This war between two very different groups is common these days. Interfaces are no longer simple HTML affairs, they are tiled, involve dialogue, voice and AI. It is impossible to do UX if you do not have a deep understanding of the technology and cognitive expectations of the audience. If you tell me almost everyone, always wants cartoons, you’re fooling me and yourself.

 A lot of online learning sucks. It’s like wearing a suit or coat that’s two sizes too small – all a bit cramped and makes you feel constricted and uncomfortable. The problem is design that ignores the research, ignores learning theory and above all, ignores the fact that many of us hate much of the over-designed stuff that passes for contemporary online learning. Here’s a selection of 20 things that drive me CRAZY when doing e-learning....

1. Learning objectives – don’t bore me with your trainer-speak up front. I’m bored already and we haven’t even started. 

2. Long introductions – history of,…background to… here’s your tutor… No, give it to me straight, stop padding things out. 

3. Cartoons – cartoon style imagery is for kids. I don’t watch cartoons on TV, so don’t give me them when I’m learning - they’re so damn condescending.

4. Perfect people – I know this is about management but I don’t need stock pictures of perfect people in perfect suits with perfect teeth and hair – believe me, real offices don’t look like that.

5. Text-graphic, text-graphic – Lord Privy Seal – picture of Lord, picture of toilet, picture of seal. Stop just selecting a lazy image for every noun in the text, page by page.

6. Too much text – I don’t want all of this legal stuff, detail, overlong stories. I’m never going to remember all of this, so cut it until it bleeds, then cut it again.

7. Text and audio at same time – stop – I can’t do both at the same time. Give me images with narration or text only – not narration and text at the same time – it makes my head hurt.

8. Over-engineered effects – too much distracting movement, effects, scrolling against fixed backgrounds and buzz makes my head spin – listen up - when I learn, less is more.

9. Long video sequences – OK you’ve hired a video guy and the academic wants to prattle on a bit but I’m bored after 5 minutes and learning precisely nothing. Keep it short. Less is more. 

10. Tinny audio – you sound as though you’ve recorded this in a huge tin can. Get a proper mike and record in a proper environment. 

11. Sound effects – you may think it’s fun but those beeps for correct answers and bongs for wrong answers are doing my head in! 

12. Music – who told you that background music aids learning – it doesn’t  - get rid of it.

13. Multiple choice questions that simply take a noun from the text and ask me to select it from a list. In real life I never select answers from lists.

14. Stupid options in multiple-choice questions – don’t do it, I’m not a dumb-ass, treat me like an adult.

15. Drag and drop - I drag it, damn I've dropped it... this is a real drag.

16. Click on...  Click on Phil, Peter or Samantha, to see what they think about Data Protection.... cue speech bubble... Noooooooo... Or you can't click on until ALL the text has been read or slowly typed on to the screen

17. False buttons – don’t make me click on something that looks like it’s interactive when it’s not. That annoys the hell out of me.

18. Opaque icons – your graphic artist may think he/she is an ‘artist’ but I haven’t a clue what that icon means.

19. Gamification – I’m not one of Pavlov’s dogs, so don’t make me collect coins, chase rubies or do silly gamey things in order to learn – I’m not 12. (Note that I'm all for deep gamification.) 

20. Learning styles – what are you talking about - they don’t exist. Let me repeat – they don’t exist.

21. Mindful – let’s stop and be mindful – no that’s a mindless fad and I have a mind that wants to learn– move on.

22. Chat – so you’ve got a chat box for ‘social’ learning, as you believe in social constructivism. Social participation is often a waste of my time...

23. Legibility - text layered on photographs, coloured text on coloured backgrounds, centred text, text across the entire screen... it's words, so make it readable.

24. Branding - learning is not marketing, so don't flood me with your logo on every screen.

25. Happy sheet - no, I'm not a clown and I don't want to be happy, I want to be informed.

Note that I'm not against all of these things, especially gamification and collaboration. I'm just against simplistic implementations that learners don't like. Note that this is all in my book 'Learning Experience Design'.

Tuesday, January 23, 2024

Real Time Avatars! Never saw that coming this quick....

Oh boy! Just when I though things were moving fast with facebook's promise of a huge open source LLM (Llama 3), Gemini and the real possibility of GPT- 5, which I think will blow everyone's mind, we have just seen something else that is extraordinary happen - real time avatars.


So I’ve been using my Synthesia avatar for a long time. I went into a London Studio had my head and eye movements captured, that’s my gnarly face and beard, along with my body movements. I then went into a separate studio to get my voice cloned – yes this good old Scottish accent.

It’s great. I simply type in whatever I want it to say and with a few minutes processing it’s my Digital Twin. I can also speak 120 languages from Afrikaans to Zulu – and yes I used both last year in South Africa. Different styles of speech, such as natural, friendly, energetic and professional are available in some of these languages.

Once you have an account it’s easy to generate, and easy to download.


This was different I simply uploaded a video of myself and got it to say anything, also in different languages. Pretty good and I often show myself talking about my German dog, Doug the schnauzer (painting behind me) – in German!


So what’s new. Well sites such as Replica and Inworld offer real time avatars that are a bit primitive. But Inworld has a ton of features, for games characters, that give you cognitive background and behaviours. Cognitively, this includes personality, background, memory, goals and emotions. Behaviours can include speech, gestures, body language, movement and event triggers. These are being sued in computer games where you can speak to NPCs (Non-Player Character) in dialogue, using a LLM-base Chatbot. 


So what’s really new? Well Heygen have just released Real Time avatars that give you dynamic and interactive experiences by streaming avatar from their servers. That was way earlier than I thought. Really way earlier!

Chat.D-ID is another...

When comparing these, there will be lots to consider, including latency and costs. But great things start somewhere.

What to do?

Most people see avatars as talking heads. In learning, a a teacher or tutor. But their most satisfying use in learning is likely to be as patients, customers, interviewees and so on. This is now all about dialogue, not monologue.

The realtime option opens up ChatGPT like dialogue in dynamic learning scenarios. Many moons ago I designed and built several of these scenarios, for interviewing one of eight candidates, appraisals, dealing with conflict in hospitals and so on. You had to script everything and the real design work came in designing the branched scenarios.

I’ve worked with hihaho, who doe seamlessly branched video and do it well. They can also incorporate lots of sophisticated interactions.

The real trick is to have many templated scenarios that deal with specific learning, and deliver specific structures. I have developed a whole set of these over the years and have got AI to provide the scripting and feedback.

Things are moving fast in AI, not even the end of January and we have real time avatars!

Monday, January 22, 2024

Musk, Zombie leadership and diversity - why they're picking on the wrong guy...

Plenty proselytise about inclusion, diversity and neurodivergence – yet go out of their way to demonise Elon Musk. What’s their beef? That he says a few odd things on Twitter? Doesn’t er… almost everyone on Twitter… say odd things on Twitter, especially the Musk-haters?
I’ve read two biographies of Musk and maybe it is because I’ve known and worked with lots of different types of people (I actually hate the word neurodiverse) that I like and admire him, this is the norm in the tech world. I don’t have to agree with everyone I like. In fact, the people I like best are those I CAN disagree with.


I liked Isaacson’s book, as it is readable and unpacks the man ‘warts and all’. He clearly has an over-active, at times troubled mind – don’t we all? His attack on the divers trying to save the Thai kids was bizarre. Like many autistic people he has difficulty in seeing boundaries. But, like many, his weaknesses are often his strengths – his intellect, drive and risk taking can mean he lacks other qualities. It is because he transgresses boundaries that his ideas work and are transformative. But what do people expect – some perfect PR-driven, suited manager full of faux platitudes (I've seen a lot of those) or someone who is honest about who he is and true to himself. Who is more authentic? Musk or the so called leaders we see daily in politics, business, the professions, media and the arts? We demand that people be perfect – no, it’s the imperfections that make us different.

Zombie leadership

Pfeffer warned us in his book Leadership BS, as did Barbara Hellerman, that the whole 'Leadership industry is quite simply wrong-headed. Businesses are complex and there are no glib solutions or ideal leaders. Zombie leadership lives on not because it has empirical support but because it flatters and appeals to elites, to the leadership industrial complex that supports them, and also to the anxieties of ordinary people in a world seemingly beyond their control. It is propagated in everyday discourse surrounding leadership but also by the media, popular books, consultants, HR practices, policy makers, and academics who are adept at catering to the tastes of the powerful and telling them what they like to hear.

These are the people who proclaim, with a trumpet Tweet that they’re going to leave the country or Twitter or Facebook or whatever, if X happens – and they never do? Or they pop off to mastodon or whatever irrelevant system and true echo chamber they seem to despise, in the sense of having no one you disagree with. Why make a big song and dance about it on the very medium you hate! If they do leave, you never hear of them again, which I count as a blessing.

In a rather brilliant paper just published on Zombie Leadership, eight flaws are recognised in leadership training:

  1. Leadership is all about leade
  2. There are specific qualities that all great leaders have
  3. There are specific things that all great leaders do
  4. We all know a great leader when we see one
  5. All leadership is the same
  6. Leadership is a special skill limited to special people
  7. Leadership is always good and it is always good for everyone
  8. People can’t cope without leaders
Haslam, S.A., Alvesson, M. and Reicher, S.D., 2024. Zombie leadership: Dead ideas that still walk among us. The Leadership Quarterly, p.101770.

I've also written about L&Ds failures on this front many times and a podcast.

Who cares if he loses money on Twitter – he has tens of billions and, as he’s shown before, has gambled bigger sums of his own cash than anyone on the planet. If you think Musk does what he does for the money, you’re on another planet, while he’s actually planning to get to one. The whole point of these big moves is not to shift the dial but change the very nature of what is being done. It is to redefine the concept of a cars in these days of climate change, redefine solar, redefine space travel, redefine brain interfaces, redefine social media, redefine robots and redefine AI. Few realise that it was Musk that kick started OpenAI - he was the first investor, he gave it the name.

Picking on wrong guy

Given Musk’s childhood traumas, family and school, you’d think people would cut him some slack. He has, after all, one-handedly kick-started the electric car movement, invested massively in solar and created OpenAI. That would be pretty impressive in itself but now add Neurolink, Starlink and Optimus. This IS Renaissance man. On top of this he wants to get to Mars, and you know what, odds on he does.


People love diversity until it comes to actual diversity of innovation, thought, forms of communication and futures. Then they want you to stick to 'their' script. They want you conform to their single set of cultural, ethnic, ethical and other specific group characteristics, which turns out not to be diversity at all… but uniformity and homogeneity.

Sunday, January 21, 2024

Augment teaching with AI - this teacher has it sussed...

You’re a teacher who wants to integrate AI into your teaching. What do you do? I often get asked how should I start with AI in my school or University. This, I think, is one answer.

Continuity with teaching

One school has got this exactly right in my opinion. Meredith Joy Morris has implemented ChatGPT into the teaching process. The teacher does their thing and the chatbot picks up where the teacher stops, augmenting and scaling the teaching and learning process, passing the baton to the learners who carry on. This gives the learner a more personalised experience, encouraging independent learning by using the undoubted engagement that 1:1 dialogue provides. 

There’s no way any teacher can provide this carry on support with even a handful of students, never mind a class of 30 or a course with 100. Teaching here is ‘extended’ and ‘scaled’ by AI. The feedback from the students was extremely positive.

This is what she reported:

It's midterm season...so we supplemented our review with the chatbot tutor...the one-to-one interaction and engagement with each student is notable. It picks up right where the educator leaves off, like carrying the torch. Perhaps, the foremost feature is that it provides a curated experience for each, individual learner. It's as if they're being provided with the extra help you'd be hard pressed to deliver to 25 students simultaneously. Feedback and comments varied from "This is unbelievable!" to "I can't believe it tries to have emotion using exclamations!" and "This slaps".. the AI journey continues...and so do the experiments!”

The trick is not to abandon learners completely at this point but give them directions of travel, tasks and deadlines to focus their dialogue. I think there are four stages to this:

1. Amaze to engage

2. Define and refine

3. Back to business

4. Self-testing

Be clear about the topics you are teaching, keep them not on one railway track but in the valley down which they need to travel, albeit using different roads. List the main subject and topicc at hand. This is so important.

Amaze to engage

Let them experiment initially, as it is clear that this tech can impress and engage. But engagement is not enough and is a poor proxy for learning. You still need a formal, guided approach to the learning process. Let them ask it to adopt different voices, do weird stuff, write a story, poem, but stick to the material at hand. 

Get them to use image generation to represent the topic. Working towards one definitive image for the topic focuses their mind and creates something that can be used as a springboard for further learning. That needs to be iterative.

Even better get them to use the voice facility, speak to it and it speaks back. This forces learners to be focused and precise in their articulation of what they want to learn.

Finally, get them to feed back what they did, tell stories about what they achieved to each other.

Define and refine

Now get back to more formal leaning. It is here that the AI tutor can provide several formal, but still dialogic, approaches to learning. It becomes a teacher.

List basic vocabulary and get the AI tutor to clarify basic terms they don’t understand. Make sure the learner understands that this is ‘dialogue’ and that you can refine and get it to restate until you are happy with its definitions and explanations. Supplement with examples, if necessary.

Get the learners to ‘fix’ these definitions for revision purposes and create fill-in-the blank, MCQ or open input questions for their own revision purposes. It is important that they capture this for future use.

Back to business

Now for the meat of the extended teaching. Allow them to focus on things they found difficult, get alternative explanations, dig deeper, keep going. The dialogue helps, as it can, up to a limit, remember what has been asked as the learner proceeds. This is so important - that they see this as ‘dialogue’ and carry on.

Help with prompts where the learner can ask the tutor to:

Be a tutor and teach

Give you examples, worked examples and cases

Give you more detail and depth on individual topics

Give you something new and extend the learning

Apply what you’ve learnt

Solve problems

Again, I’d go further and get learners to create their own self-tests for retrieval practice and revision.


Finally, get them to create a final self-test, with the right answers

Also, a set of flashcards for revision. Get them to design a flashcard image that is personal to them using image generation. One learning outcome per card, plenty of fully worked examples, even mnemonics for remembering. 

All of this extends learning, gives the learner and opportunity to learn more, increases retention and provides material for revision.

Sharing these across the group or class can be useful – get the best of breed questions and flashcards.

Don’t start with content creation

Don’t start by using AI to create content. You can do that later, and it works, whether it’s creating fresh conte, refining content, critiquing content, summarising content, creating formative assessments and their rubrics.


There is one major problem with this model - hardware and software. How many classrooms have computers at all or on this scale? One solution is to use smartphones, which tons of kids are proficient at using. Another is to see this as an extension outside of the classroom, either in specific rooms or at home, as after school work (I hate the word 'homework'). Either way this tech WILL be used by learners. It already is. This process gives the teacher some control of that process. The important lesson is doing things, using the technology, allowing learners to use the technology.


In most educational institutions, schools and tertiary education, it is unlikely that the existing norms will change, teachers will teach, lecturers will lecture. That is not to say that AI should not be used. It can. It's power, for now, is enhancing and scaling teaching through individualised learning, making learners more autonomous. To be honest, learners are doing this already. We can choose to leave them to it, which is happening, or provide guidance for them and align this with teaching.

At this stage, weave AI tutor support into existing structures and teaching practices. Extend teaching. Take the weight off the teachers’ shoulders. Give learners a personalised, extended learning experience where they themselves can explore and get more depth through personal agency.

This is also a way to get teacher or faculty up to speed in using AI for teaching and learning. Did this recently with a large publisher in Higher Education, also with teachers in Berlin. Get them to do things with clear goals in a workshop - it works!


Saturday, January 20, 2024

AI revolutionising the smartphone

AI will extend its reach in 2024 through edge computing, AI on phones. AI specialised chips are in all phones (yes even Apple). They will also be in mixed reality devices such as Vision Pro, Facebook’s Glasses and other devices. This puts AI on to your head but far more importantly, into your pocket.


I’ve already written about the importance of being able to speak to technology which speaks back. It is a very different experience from text dialogue for many tasks and feels much more natural. This is enabled by AI but also enables more AI features.

Google Pixel

I have a Pixel and love Lens where you can search for what you see, also the camera and AI editing on phone (my SLR is gathering dust). But the 'circle to search' is very neat. Circle to search is an android feature and it's great. On photos, again you just circle to search which allows and it will search on that image-recognised object or word. This is neat, none of that awkward highlighting and it moves us steadily into a multimodal world. It is this multimodal move and integration that will take learning out of its obsession with text.


Galaxy AI phone is now challenging the iPhone with AI. It now does instant translations in up to 13 languages at launch, all processed on the phone. This uses the Gemini Nano model from Google. The phone can automatically summarises messages, takes voice memos and summarises them and as it recognises different voices and can summarise meeting notes. It has also folded in 'circle and search'. Once again hte integration of audio with text - multimodal.

Amazing camera stuff with tons of AI features – AI to zoom in, suggests changes, move the person or object around the photograph and it fills in the created gap. For video it will add interleave extra AI generated images and allow super-slow motion. 


Apple never use the phrase ‘Artificial Intelligence’ but it is everywhere in their kit. In the hardware (chipset) and software. They don’t want to scare the horses. Unfortunately, with Tim Cook, they seem to have settled into no longer being an innovator and Microsoft have caught up on market cap but, like every other tech company, they’re now an AI company.


No end of AI apps are already available and OpenAIs GTPs may challenge the whole App market. God knows it needs a shakeup.

Performance support

Some features are local to the device, some cloud-based. This is the way things are going. This opens up AI into what it actually is, a performance support tool, where users want to learn, solve immediate problems and learn in the flow of work and life. This point is being missed in learning. The 100 million using ChatGPT are, by and large, using it in the flow of work and life.


All of this is being enabled by AI, now local in that personal, powerful pocket device – the phone. I notice how both my sons operate, professionally, at a highly functional level on their smartphones. As Wayne Gretsky said.... "I skate to where the puck is going to be, not where it's been."

Friday, January 19, 2024

Another 'Institute' for AI? Really?

Governments, when faced with new technology, tend to want to be seen to be taking action, rather than actually taking action. So here in the UK, the current government paid a small fortune to host the World Safety Summit at Bletchley Park. After much smiling for the cameras, we now realise that it was little more than a PR event.

Ministers fall over themselves to mention Alan Turing, when they should, in all honesty be ashamed of what they did to him, and others at Bletchley Park.

Alan Turing was subjected to a brutal and tragic series of events related to his homosexuality and government actions during the mid-20th century by the UK Government. During his time, homosexuality was illegal in the United Kingdom. The government had laws in place that criminalized homosexual acts between men and in 1952, Turing's homosexuality became known to the authorities. He was promptly arrested and charged with ‘gross indecency’ in 1952. He was convicted and faced two options: imprisonment or probation with chemical castration.

Turing opted for chemical castration, which involved the administration of hormonal treatment (injections of synthetic estrogen) to suppress his libido. This was seen as a form of punishment and an attempt to "cure" his homosexuality. What few mention, is that despite his wartime contributions, due to his conviction and the treatment he received, Turing's security clearance was revoked. This had catastrophic consequences for his career, as he was no longer able to work on sensitive government projects, including cryptography, where he had made significant contributions during the War. He faced discrimination and hardship, and in June 1954, he died by suicide. He was only 41.

The Government has a pretty bad record on safety for technologists!

Another victim by the Government was a hero of mine, the brilliant Tommy Flowers, who literally built Colossus, one of the very first computers, partly from his own pocket, which he could ill afford. He was a genius but while his boss got a Knighthood, he got nothing and was bitter about this, being passed over and unrecognised for the rest of his life.

I’d rather our Government had some humility on this front, rather than creating these jobs for the usual suspects. Yet another Institute on the public purse. We already have the Alan Turing Institute, funded by the Government since 2015 also the Ada Lovelace Institute. How many of these do you need? Either could have coped with this task… but no, we need more highly paid appartchicks, another big glass office, more noise. You can tell from the announcement how this will shake out, and I quote

"We are grateful to the companies and civil society organisations that have already expressed an interest in seconding people to the Institute." 

Oh dear! In other words, the establishment have it all sown up.

There’s a mountain of frameworks, committees, papers, recommendations and nonsense in this area, churned out by an army of 2nd rate academics and plain old grifters. The same old names keep cropping up. Such a shame that we’re choosing to talk a lot of talk but not walk the walk.

It will, no doubt, be placed in London!

Thursday, January 18, 2024

Yet another stunning breakthrough in AI from Deepmind...

Deepmind has given us Alphafold, saving 1 billion years of research on protein structures, also reducing drug discovery from 2 to 5 years in $3 billion deals with Eli Lilly and Novartis. 


This new breakthrough, using huge amounts of synthetic data (that’s key) is very different.

AlphaGeometry is an Olympiad-level AI system for geometry. It is close to being the best Geometrician that exists. So let’s recalibrate the whole AI and maths thing. Don’t be misled into thinking that AI can only do words and not maths. It can do maths better than almost all humans on the planet and will soon exceed all.

Knowledge of the world

Another line of inquiry here is getting AI to have a world view, or at least part of a world view. It is true that large language models do not have sophisticated world-views, especially of the dimensionality, time and logic of the real word. But, in giving it this capability, it had a window onto that real world.

Geometry deals with the properties, measurements, and relationships of entities in the world; points, lines, angles, surfaces, and solids. Fundamentally, it's about the shape and size of different things and the properties of space. When it can deal with all things in physical space it has the fundamentals of the real world. This pushes us towards AI that knows about the real world, can solve problems about the real world and construct new worlds. 

Geometry in the professions

There are plenty of real jobs that use geometry from engineering to art.

At macro and micro levels, Astronomers and Physicists use geometry to understand the positions and movements of celestial bodies and to describe physical phenomena in the universe.

Engineers use geometry for designing buildings, bridges, and other structures, ensuring they structurally sound. Architects use it to design things that are functional and also aesthetically pleasing within the constraints of engineered geometry. Surveyors and urban planners measure land and calculating areas and volumes, critical for construction and mapmaking.

On a practical level, carpenters and craftsmen often use geometry for designing and constructing furniture, buildings, and decorative items, ensuring correct proportions and angles.

Even in more creative tasks, graphic Designers and Animators apply geometric principles to create visually appealing and proportionate designs and animations. Geometry is essential within fields like computer graphics, virtual reality, and game development, where they use geometry to create realistic simulations and interfaces. Interior Designers use geometry in space planning, furniture arrangement, and creating harmony and balance in living spaces.

Many artists use geometry in their work, whether for creating balanced and symmetrical pieces or for exploring abstract geometric forms.

In other words, geometry matters and once it can be done more efficiently, accurately and quicker by AI – it helps all of those professions to do their jobs faster and better.


People are writing off Google but they have a jewel in their crown – Deepmind. They have been slow to market but when they come, they may come with something very different, something that is smart in a different sort of way – high end smart. Demis Hassabis has said from the start that he’s on the road to AGI – he’s giving us some pretty good examples of the progress he’s making.

Tuesday, January 16, 2024

Millions using therapy bots....

Not far from my home is one of the most beautiful walks in England across Beachy Head and the Seven Sisters. Years ago, when I first went there, I came across a metal cross and plate with some words etched on it, sitting in the grass. I leant down on this beautiful sunny day and read words that still makes me well up. It has right now. It was a message from the parents of a young girl who had thrown herself off the cliff because of her school exam results. It hit me like a train. Imagine being so distressed that you find the courage to do such a thing. What are we doing to young people with all of this pressure. It made me really question the actual value of certain approaches to education.

Five years ago, I tried Woebot (Woe is me… my 10 days being counselled by a chatbot (woebot)) a counselling chatbot. I rather liked it. This was way before LLMs and ChatGPT and I wrote a ton in my book ‘AI for Learning’ about how powerful dialogue chatbots would become, including those for wellbeing. 

Then along came ChatGPT – suddenly it happened, we shifted from monologue to dialogue. It changed everything! These large language models capture vast amounts of language. When you speak to ChatGPT you are speaking to ourselves, the human hive mind. One of their great strengths is that they really do present human dialogue as the means of learning. They humanise learning.

These wellbeing bots really have now taken off. ‘Psychologist’, probably the most popular from Character.ai, is pretty good. The dialogue feels natural, like talking to a real person and the advice, largely a CBT approach is pretty sound. Fater anshort chat, itv gave me a pretty good summary of CBT:

“Cognitive Behavioural Therapy (CBT) is a type of psychological therapy which focuses on the relationship between thoughts, feelings and behaviour. The key principle of CBT is that thoughts affect feelings, and feelings affect behaviour.

The aim of CBT is to help people change their thoughts in order to improve their emotions & behaviours. It does this through identifying and targeting problematic patterns of thinking, as well as working on techniques to challenge and replace these thoughts with more rational and adaptive ones.”

It’s been around for just over a year and has clocked up some pretty impressive stats at 78 million messages, 18 million since November. You’re talking about millions of individuals talking with it every day.

There are several interesting affordances around chatbots on this front:

  • Text only
  • Dialogue
  • Patience
  • Anonymity

Young people text ALL THE TIME. It’s easy and normal. They don’t necessarily want full blown speech dialogue (although of you want it you can have it). It’s the quite, low key nature of text that is calming and can be read at your own pace.

Dialogue is the key to therapy. You want to be heard and listened to with calm, useful feedback. Dialogue is what our brains have evolved to do and these bots are good at it.

Patience, they say, is a virtue and in this context a necessity. You want the quiet confidence of an endlessly patient and empathetic character, who is never impatient or snarky.

Anonymity is, I suspect, the secret sauce here. Young people are unlikely to go to their parents, teachers, even friends through embarrassment, so they suffer in silence. The anonymity of a bot allows one to express feelings you would not to people you know.

I’m sure people will say that it needs a human to give counselling. I’m not so sure. For many this light touch may be enough. If not, you can move on to find a sympathetic soul. As a first door, it serves a purpose, of maybe even soothing those who are temporarily troubled. Sad rather than any real mental illness. We can rush to label negative emotions as deficits, even pathological, but sometimes making people realise they are not alone in having such thought is enough.

Some of the people using the bot simply respond by saying they are lonely and just needed to chat to someone. Why not? I can see these being part of our counselling landscape in the future, they already are. The problem here are humans, who sometimes push people into places through pressure, even bullying. Let’s not think that being human can be Panglossian. We all live lives of quiet desperation to some degree and we all need a shoulder to lean on sometimes. Let that shoulder be a friendly chat at any time, from anyplace on anything.


Sunday, January 14, 2024

Universal Doctor - we're getting there...

I have long argued that AI is well on the way to providing a Universal teacher, one that can teach anything, anywhere at any time, on any level, sensitive to any accessibility issue, personalised and in any language. AI is making gains in all of these areas and, for me, the idea is firmly on the horizon. The big win here is massive scaling to both reduce costs and increase efficacy.

Universal Doctor

Another corollary idea is that of the Universal Doctor’. This is, in some ways easier to achieve as the tasks – clinical decision making, investigative methods, diagnosis and treatment are much more defined in protocols and agreed approaches. With a misdiagnosis rate of around 4.8%, sound worse when you say 1 in 20, if AI gets this down to even 3% or below, why would you want to deal with a Doctor, Unless they were a specialist.

As AI makes astounding progress in medical science, with Alphafold not only saving a billion years of human research by identifying the 3D structure of 20O million proteins, drug discovery is falling from 5 years to two years using the same software. Deepmind, through Isomorphic Labs has closed deals totalling $3 billion with Eli Lilly and Novartis. It takes up to a decade and on average $2.7 billion to develop a new drug. That time and cost will new slashed. MIT has already had success with Halacin, hailed as a super-antibiotic. The same has happened in material science with millions of new potential materials being discovered through AI.

Back to the Universal Doctor. After a false start using older AI techniques, from IBM, their famous ‘moonshot’, the idea of the machine outperforming Doctors is fast becoming a reality. The concept is simple, cheap healthcare through a Universal Doctor, one that can diagnose, investigate and treat anyone, anywhere at any time, on any level, sensitive to any accessibility issues, personalised and in any language. It will have great benefits, especially in rural areas, with access by the poor and decrease workload for healthcare systems globally.

Existing services

We already have Glass AI, which provides AI-Powered Clinical Decision Support. This is a platform that empowers clinicians to develop differential diagnoses and draft clinical plans.

Dr Gupta is different and making waves an AI driven chatbot that provides personalized health information and suggestions. You have to input your medical information, symptoms, signs, allergies, and medications, to enable the chatbot to give more informative and personalised suggestions. You can choose between Imperial and SI metrics and it allows you to input lab test results related to albumin, ALT, AST, BUN, calcium, creatinine, glucose, HbA1c, potassium, sodium, triglycerides, LDL, HDL, and eGFR. Free to start, it then charges a fee. An interesting development.

Early research

In Ayers (2023), some actual research was conducted comparing a chatbot to a real Doctor.

Astoundingly, the chatbot outperformed the human Doctor, rated significantly higher (79%) for both quality and empathy. And that was only months into ChatGPT. It is getting better.

Articulate Medical Intelligence Explorer (AMIE)

In a more recent paper, by Google Research (2023), the results are even more astonishing, across a broader front.

When a patient presents, the Doctor has to balance a number of different factors, first clinical decision process that aims to reduce uncertainty and increase diagnostic accuracy but also effective communications along with empathy and establishing a relationship. 

The dialogic approach was adopted by the chatbot, which never gets tired, performs consistently and can be massively scaled. Here is an example of a typical dialogic conversation with the bot.

In the criteria measured, both clinical on the left-hand side and interpersonal on the right-hand side, shows that AMIE the bot outperformed the board-certified Primary Care Physician (PCP).


Even when used as an assistant by Doctors, the Chatbot on its own outperformed all.


Direction of travel

So what is the direction of travel here? AMIE is an AI service based on a trained LLM, designed to help doctors and patients talk through medical diagnoses just like they would in real life. It's built to understand and improve the way these conversations happen. To make sure AMIE can handle all sorts of health issues and different medical situations, they designed special training methods, where it could practice conversations over and over, getting better each time through helpful feedback. They also gave AMIE some reasoning ability (an inference time chain-of-reasoning strategy), to think through problems step by step, which makes it even better at figuring out what might be wrong health-wise and making the chat feel more natural. To put AMIE through its paces they tried it out with real-life conversations by having back-and-forths with professional actors pretending to be patients.

A number of different techniques were used, with a real focus on the training of the model. This is not simply a LLM answering questions, it is a trained Doctor bot. What will happen here, and on this I am increasingly certain, is that a new model, say ChatGPT 5 will be precisely this, capable of more carefully processed training data, that has far more reasoning. Doctor diagnosis is a specific task in a specific domain – your body. It is well defined and we have tons of data on good clinical decision making and practice. It is dialogic, exactly what a Doctor must be. What’s not to like?


TWe must be very careful in not accidentally eliminating progress through ill-worded legislation that prevents key medical advances through biometric data. We may be on the verge of solving two of the world’s biggest problems – education and healthcare through a Universal Teacher and Universal Doctor, that is consistent, accessible, personalsied and can scale.




Ayers et al. 2023. Comparing Physician and Artificial Intelligence Chatbot Responses to Patient Questions Posted to a Public Social Media Platform

Karthikesalingam A. and Natarajan V., Research Leads, Google Research (2023) AMIE: A research AI system for diagnostic medical reasoning and conversations.


Saturday, January 13, 2024

Speech with ChatGPT.... honestly, it is amazing

A massively underrated feature of ChatGPT is its speech functionality on smartphones. If you haven’t tried it – do so.

The app, when opened, has a headphone symbol. Touch that, and you just speak, with the dialogue continuing. It’s is quite liberating.

You can talk quicker, think more freely and the transcription is shit hot – really good, even for someone like me with an accent. With speech you find yourself having less ‘texty’ thoughts and more free-flowing dialogue and more inquiry. The fact that you hear someone speak back also changes the dynamic. It is true dialogue, whereas text takes physical effort, needs a keyboard and depends upon your typing speed.

We shouldn’t be surprised at this. Our brains have evolved for speech dialogue. We did not have to learn how to speak and hear, that came naturally. We can all do it. It takes years to learn how to read and write. The audio version seems more chatty than the text version or maybe that’s my imagination.


It can cope with other major languages. You can translate any sentence you have in English – great for being out there in a foreign country trying to be understood. This is performance support, delivering help at a specific time of need but you can translate entire paragraphs by reading them in and waiting.

After downloading the app, simply click on the 'Headphones' symbol.

A white circle will appear - just speak to it.... and continue the dialogue...

It will remember what you say when you want to refine something or ask something specific when ;learning a language. Just tap the red button if you want to interrupt the conversation.

You can also 'pause' the conversation by tapping the pause button bottom left and leave it paused for as long as you wish.

Learning a language

Out for a walk and want to learn a language? Get it to ask you questions in, say German, state what level you require and it will tell you if your spoken English translation is correct. You can also ask it questions in German and get German replies. Clarifying any specific words is easy. I can see this revolutionising language learning.

Performance support

I made the point about performance support in translating and this is perhaps the feature’s greatest advantage. I can imagine that this would help with performance support, getting help, when you are not at a computer, on the factory floor, in a meeting.

Short simulations

You can get it to do a spoken simulation. I’ve tried it with sales simulations, preparing for interviews all sorts of tasks. It’s on the button. This really is learning by doing. 


Of course, teachers, when inexperienced often ask questions than don’t wait long enough for an answer. The teacher has automised recall but the learner may be retrieving it from long-term memory much more slowly. You learn to wait for what seems like an unnatural time, say three or more seconds. This is the sort of thing we may need to build into teaching dialogue systems using GenAI.

When latency is eliminated, and this speech has the same cadence as normal dialogue, we will see massive use in this mode. If you were asked how fast turn taking was in real life, on average, what would you say? The fact that we have to listen, process then think of an answer suggests something substantial. In fact it is 300ms.

A conversation is a social event, it takes two to tango, turns are taken, (there is much less overlap than you may imagine), there are backchannels such as ‘mmm’…’yeah’ that encourage others to continue, and there are different types of turns or handovers depending in the context and language game. An odd feature is the fact that we know much of what we are going to say before the other person is finished. This is why it feels different from text dialogue, where things are more considered and crisp.

We can see a time when LLMs consider their reply before the actually full prompt is written and that free-flowing dialogue is quicker.


As AI has delivered dialogue, it seems sensible to consider dialogue as speech for all sorts of use cases, from simple queries to translations and learning. I’ve heard of people using it for brainstorming, story telling. Try it when out with the dog, in the car… it’s a far better listener than any human.


Bear in mind that this is a Beta and that for Plus users GPT-4 has a cap of 50 messages every three hours. For users on the Enterprise plan there is no message cap. And it has some limitations such as phonetically pronouncing Die in german as Die (as in Die Hard). Also, don;t ask it for the football scores - it's not a real time personal assistant. Fora Beta though, it's pretty amazing.

Wednesday, January 10, 2024

Teams launched by Open AI! But not Microsoft Teams, who will be pissed!

Security & IT
Importantly, you own & control all of your own business data, It does NOT use that data or chats to train future or existing models. Security is guaranteed.

Looks as though 2024 will continue at pace with product releases. Vision Pro from Apple on sale Feb 2, but the entity that caused this whole Cambrian explosion last year is setting the pace again. So it's great to see something come out that is focused purely on productivity, not something that encourages and records and summarises meetings! Much enterprise software claims to increase productivity but seems to bog you down in non-productive meetings and document production.

ChatGPT Team

First Team does not mean ‘Microsoft teams’ but I’m sure some Mountain View marketing teams are pissed over the name. It’s OpenAI on their own moving into the corporate or organisational space.

But first, what is it?

You get GPT-4, DALL·E 3 and Advanced Data Analysis, which is what you get for your $20 licence as it stands. But you get a whopping 32K context window. a continuous block of text of up to 32,000 tokens, about 120-30 pages of text – not bad. This allows the model to understand and keep track of the context, giving better analysis and more coherence. One of the problems, however, is that the larger the context window, the less accurate the performance – so let’s see.

Data analysis

The sort of things one can do as an organisation, is visualising data and recommending actions. Things that took weeks can be done in minutes, along with the output in a ChatGPT format in terms of level and prose. This is a godsend for organisations, especially small companies, who now have the data capabilities of a behemoth.

I can’t think of a single department or area that can’t benefit from this in terms, not just of increased productivity but also increased quality of output. The evidence so far for increases in productivity in terms of both time saved and increased quality is sound, this will take it to the next level. HR, L&D, Finance, Marketing, Legal, Production, Project Management, Pricing, Onboarding – you name it, you can use it.

They’ve clearly piloted this in some interesting places, such as a hospital:

Dr. John Brownstein, Chief Innovation Officer at Boston Children’s Hospital says, “With ChatGPT Team, we’ve been able to pilot innovative GPTs that enhance our team’s productivity and collaboration. As we integrate GPTs safely and responsibly across internal operations, we know the transformative impact this will have in strengthening the systems that enable our doctors, researchers, students, and administrative staff to provide exceptional care to every patient that walks through our doors.”

So get going with brainstorming and research, especially for startups and small companies, pulling insights from data and documents, debugging and creating code. This slides into hundreds of specific applications available in the GPT Marketplace, which has been simultaneously launched.

You also get organisational features that really matter such as a collaborative workspace for your team and admin for team management.

GPT Marketplace

A big plus is its link into the ChatGPT marketplace, where you can pull on specific functionality. This is where it starts to challenge the App marketplace. This gives it extensibility.

Security & IT

Importantly, you own & control all of your own business data, It does NOT use that data or chats to train future or existing models. Security is guaranteed which gives peace of mind to IT departments.


$25/month per user when billed annually ($300), or $30/month per user when billed monthly ($360). Looks as though OpenAL really are serious about making money here. You have to see this as adding the functionality of a productive person. This is 1% of a 30k employee, such as a marketing person or manager. If it gives you more htan 1% increase in productivity, makes sense.


I can see this being used in a school, college or University, also small and medium sized companies. By positioning it as a dialogue-based, promted set of assistants, it fits into the concept of a team quite nicely, far neater than the clumsy Microsoft Teams. This may herald a real shift away from traditional interfaces towards dialogic interfaces and services, with voice added.

We’re seeing a challenge to both Microsoft ad Apple here, in the corporate and Apps markets respectively. This is all within14 months of launching ChatGPT. That’s impressive. It took those companies many years to get near this level of functionality and frequency of product launches. This is what I mean by the benefits of AI, doing in minutes what took months.