0:03

Welcome to the show where we expose new perspectives on our ever evolving world through the lenses of various industries, cultures and backgrounds.

Our guests are disruptors, united by a common goal to bring their purpose to life.

0:21

Whether they’re from the commercial world or third sector, from the global North or the global S, expect an inspirational journey that will transform your perspective on just what is possible.

My name is Philippa White and welcome to Thai Unearthed.

0:42

For many of us, AI became an everyday reality in November 2022 when Open AI released ChatGPT as a public facing product, allowing us all the opportunity to interact with the AI model directly via a chat interface.

However, there are others who have been working closely with AI well before, understanding the opportunities, the challenges and the risks.

1:06

Hello and welcome to episode 96 of Thai Unearthed.

Today, we’re going to explore the intersection of technology, innovation, and ethics in a rapidly evolving world with Joe Fennell, who has been immersed in the study of AI ethics and safety for the last four years.

1:28

Joe’s passion for AI ethics began during his undergraduate studies and it has since evolved into a remarkable career path with a focus on ensuring AI works for the greater good.

Joe is part of the University of Cambridge’s first ever cohort of M Phil students studying the ethics of AI data and algorithms, and alongside his studies, he helps business teams upskill in generative AI and consults on AI ethics strategy.

1:58

He also works in partnership with the UN Mission in Kosovo on Safe Net, which is an ongoing project to help young people in the Balkans upskill in generative AI and become aware of threats and scams that bad actors increasingly use AI for.

2:17

So from exploring the global inequalities that AI might exacerbate to discussing the potential impact on human relationships, Joe brings a wealth of insight grounded in his extensive research at Cambridge.

Together, we’ll be tackling critical questions like, can AI be a force for good?

2:36

What ethical guardrails are necessary to prevent unintended consequences?

And how can we ensure that AI is used responsibly in areas like healthcare, education, and beyond?

So stay tuned for a thought provoking conversation that touches on the future of AI, its risks, and its promises for a more equitable world.

3:03

Joe, I am so happy that we finally made this happen.

Thank you so much for joining me here on Thai on Earth.

I’m thrilled to be here.

And thank you for doing this on a Sunday.

Normally I don’t record podcasts on a Sunday.

You’re a busy guy.

You’re a busy guy.

I’ve got a lot going on as well.

So we managed to make it work.

3:19

It sounds like life has been good to you post graduation.

Yeah, I’d say I’m got positive momentum on a few different dimensions.

I’m just trying to keep it up.

I’m in a lucky position, fresh out of university that I’m not really doing the brutal job application growing at the minute like lots of my peers are, because I’ve got a few disparate projects that if I just keep doing well, I think something good should come of it.

3:41

So that’s what’s taking out my time these days.

It sounds exactly what you should be doing in exactly that way.

I think that sounds fantastic.

So for our listeners, it would just be great to, well, first of all understand where are you?

So where are you sitting right now with that amazing image of the background?

I’m at home.

I’m at my family home in Berkshire in Wokingham, 10 minutes drive from my old school.

4:01

I suppose what’s brought me here is I graduated from Cambridge, first ever M Phil in the ethics of AI, data and algorithms.

I started philosophy undergraduate there as well before and I was massively interested in AI ethics and AI safety from from my first year and really my gap year before that.

4:18

I won an essay competition in my second year on black box medicine, which is comparing the lack of transparency in AI algorithms used for healthcare, but also used by humans and GPS who also can’t always explain their methodologies for how they come to certain diagnosis predictions and trying to make those standards equivalent or comparable.

4:37

At least not having a double standard that is.

Just finished my infill.

I’m still happy to be here, happy to talk about.

Great.

It would be great just to sort of know a little bit about you as Joe, who is Joe Fennell.

So you obviously went to Cambridge.

What got you to even wanting to study AI?

4:54

How did you get into that area?

Yeah, I mean, I was, I was paranoid about getting a job.

You go through this horrific university application grind in school, GCSE is a brutally competitive as well.

I sort of was trying to find safe career paths and I read this book in 2019 Life 3.0 by Max Tegmark.

5:12

That convinced me that my plan of all these my base option of becoming a lawyer might get compromised by large language models in 20-30.

Who at just the point I’ve come into the end of my grind years as an associate and maybe looking to make partner or do something more senior, will cull up to anywhere between 20 and 50% of the firm because of the how good they are at doing things like discovery and finding documents.

5:34

So I was like, wow, OK, if it’s this compromising and this effective to my career plans, then I should really try to find something that’s relatively.

I was already committed to philosophy at Cambridge.

So instead of switching to law, I doubled down on philosophy of AI and AI ethics.

5:49

I guess that tells you something about me.

Like, I’m sort of a anxious, worrying person and like, yeah, that’s what got me into this.

How has that sort of four years, I guess, right, four years at Cambridge focusing on AI, I mean, how has that journey been?

Well, I guess just on this idea of like was, you know, pure anxiety and fear of job automation.

6:09

What drove me into this?

No, there was interest, but I think the order of my prioritization between sensibility of what should I be doing, but what I want to be doing was very much, I’m going to worry about this thing first.

And then among what’s left, I’ll pick what I’d like to do versus I feel like lots of young people, people who have a hobby they really like doing and then they try to monetize it, which was very much not me, right?

6:29

I went the other way through.

I decided that of all the things I could be doing that could make money, AI was one of the more interesting ones.

Was making my point there.

You know, what was it like doing AI before everyone became aware?

People have been aware and worried about AI since 1956 with the Dartmouth project.

And then sorry.

6:45

And by AI, I’m talking about machine learning, which is where that technique was really, really cool because lots of people, you know, consider Alan Turing’s work or see with the Enigma machine to also be sort of Seminole example of computer systems and things like the Turing test, which are all the ideas of how to identify standard for artificial general intelligence, which people don’t think it is very much any good anymore.

7:04

But there’s been AI summers and winters is how people describe it of when people really feel the hype versus when people feel really disillusioned.

I’m starting to see signs now we’re coming to the end of the summer of generative AI, or at least there’s the critics are getting more attention and more application than I think they were this time last year.

7:25

And there’s definitely dimensions of AI that have been overhyped on what it was like in my experience.

I mean, this, this is something I could talk a lot about.

I joined Cambridge in 2020.

You know, we’re in a pandemic and lots of schools online AI safety was not in any of the curriculums in so anywhere there wasn’t any courses, at least with the mainstream university.

7:46

I’m sure with the judge Business School and certain PhD courses there would have been, but there weren’t any masters that really focused on AI ethics or AI safety.

What there was was effective altruism with their Cambridge base, which is a movement about doing good more effectively.

And originally they were really focused on only global developments and eradicating extreme poverty.

8:05

But then there came considerations of animal welfare and then this really seismic considerations of future generations.

And the focus came about how can we prevent the world ending from various different likely causes and how they calculate what causes are the most important to work on and most likely to be world destroying as which has the most uncertainties.

8:23

Because if there are at least uncertainties regarding risk, you can manage that risk.

I’m.

So sorry, I’m just thinking for somebody who has problems with anxiety and worrying about things, this is like the worst area to be.

Yeah, no, that’s real.

That’s so, so real.

So, no.

So there’s effective altruism.

Global London happened fairly recently, and there was a whole workshop about existential risk and mental health because these people are naturally have an anxious disposition.

8:45

They care about people, you know, they care about life.

And they spend their lives, yeah, worrying about rogue supertalging systems taking over or biological warfare or nuclear warfare.

Well, much more likely, you know, a climate change catastrophic event that’s released, a bio weapon that creates a refugee crisis that causes like, states to militarize and enforce AI military systems successfully, Right.

9:05

In reality, you’ll get all of these things at the same time.

And that’s what these researchers have to deal with.

So yeah, I mean.

It’s not very true, but I mean, that’s becoming your focus.

It’s just a bit.

OK.

Yeah, that was a fun day.

For sure, like effective altruists I think are unpopular for a few reasons, but I don’t know they necessarily get enough sympathy for.

9:22

Yeah, thinking about this stuff seriously can be really, really taxing.

And I think what’s just really important for our listeners to also understand, if I’m not mistaken, your degree, it’s the only one focused on AI in the UK and it’s certainly a leader in global AI research.

9:41

How would you articulate that?

So I think there’s one in Northwestern Union London.

I think Oxford’s snuck in there as well the this year now.

So I think they’re on Twitter now.

I think Sydney has an AI ethics masters as well, University of Sydney in Australia.

So it’s not the only one.

But DFI, Cambridge Sense of Humour and Future Intelligence is definitely a very, very robust organization with some phenomenal researchers.

10:05

There’s aspects of the research that I think are really unique.

So my lecturer and one of the Co founders, Doctor Henry Shevlin seems to be coining this phrase social AI, which is AI fuse for social purposes, which are romantic partners and therapists and sometimes used to impersonate dead people for renews of like comforting and capacitors.

10:21

And there’s some really work and research have been learning in the reading group there on kinds of intelligence and social AI.

And Doctor Stephen Cave, our founder, has done lots of work on AI as the latest form of technology used to exacerbate like colonial powers and imperial powers and exacerbate the gap between the global North and the global South.

10:43

And there’s a lot of work on shedding light of ghost work in AI value chains with people doing very boring and rind based labour to label data sets in parts of the third world for next to no pay, who don’t get the acknowledgement for it and often don’t get the credit for work they’re really doing.

11:00

Because they want to give the impression that the algorithms is doing all it’s all itself when it’s really not.

You know, there’s, there’s lots of ways in which CFI is a really impressive a, a boundary pushing organization on those boundaries.

And I mean, it must have been an incredible experience because you were, I remember you saying that you’ve been, you know, having conversations with the people who are leading this in the world because your degree and your professor is one of those people.

11:26

So, I mean, you’ve had exposure to some incredible conversations, I imagine.

For sure, for sure.

I’m sorry, Just to circle, back when I joined in 2020, there was nowhere academically to really take care of safety seriously.

I was part of effective altruism and Cambridge’s fellowship, which an 8 week reading group on AI safety, governance fundamentals.

11:44

That then spun out into an organization called Blue dot Impact, which now runs like thousands of fellowships all around the world.

And I was sort of in their beta and that is like the go to place to get informed on AI governance.

And that was like in my second year.

By my third year, I was being surveyed on what should go into the masters on AI ethics, data and algorithms.

12:02

So genuinely within that three years, my timing was quite good about the gap year.

I saw AI safety and AI ethics go from this very fringe nerdy interests and Co curricular activity for some computer scientists, engineer and and me sort of the one philosopher who was trying to keep up and to yeah an accredited master’s degree, which was really quite something I’m.

12:20

Curious.

I mean, there’s obviously going to be many ways to answer this question, but for the listeners and perhaps in a way that the normal everyday person can understand, what was one of the most unforgettable things that you learned in your four years at Cambridge about AI?

12:35

That’s kind of stuck with you since.

Yeah, the the one I think about a lot is some research done by Whaley Chen in MIT who had these little fuzzy robots that used as tutors for children.

And they can be used either to tutor children or they can be used as peers, or they can even be a junior.

12:53

So the child is tutoring them and they will have positive learning outcomes.

I forget in which order is the best, but what’s really interesting is if you add a parent into this dynamics and the three of them, if you compare the interactions between the parent and the child when the robot’s there versus when the robot isn’t there, there’s more interactions between the parents and the child when the robot is there.

13:12

Now that really threw me because it is completely opposite to the narrative of AI and relationships.

Everyone feels like AI, particularly social AI, in some sense replaces human connection or lessens human connection.

But at the level of groups, it seems to be that I can function as a social catalyst and actually increase human connection same way maybe a gainer interactive thing can.

13:31

Now that study, like all the studies in the social area, literature was not done with enormous amount of power.

So not that many participants and not over an extensive long time.

I’m very keen to see this field get more resources that we can have a much longer version of that study to see if that’s a recency effect.

13:47

You know, maybe you get more talking with your child when you’ve got this call new toy, but when the toy stops being new, maybe that stops, right?

But for now, it’s an open question.

And, you know, I think that could be a really positive vision for AI and, you know, human to human connection if that relation and that dynamic could exist.

Yeah.

14:02

I mean, it’s super interesting and polemic, right?

Any parents who are listening to this instead of thinking of, I mean, gosh, even just with mobile phones and screens and technology and trying really hard to just take away that exposure as much as possible for so many reasons, right?

14:20

And then listening to this, you can kind of see the resistance as well from parents wanting to kind of introduce kind of your relationship with your child will be better if you introduce a robot.

And it’s like, Oh my God.

So I mean, it’ll be interesting to see not only, you know, if people continue to want to facilitate that kind of relationship because it’s not new and if it is still working from a from a research point of view, but just how from a behavioral point of view of people will embrace it.

14:47

But it’s fascinating, right?

You wouldn’t imagine that.

No, not at all.

Not at all.

And as I say, it’s becoming mainstream.

So when I said the accredited course on AI ethics at Cambridge and what’s a real change from what I was used to, there was actually quite a different flavor of AI ethics in the Masters than what effective altruism Cambridge was concerned about.

15:03

Effective altruism Cambridge was concerned about existential risk and like AI creating bio weapons and AI ethics Masters is really about the Gray areas.

So it’s about these questions about when, for example, marketing tool for romantic partners becomes extractive, like say you get someone fall in love with your system and then put the romantic beach.

15:20

Minor paywall is something that Replica did and figure out the works of that.

I’m reading something in a minute on this big copyright dispute for an AI as well.

And there’s ethical debates on either side of that for the freedom that open web versus compensating artists.

But the other massively vindicating thing for effective altruist that happened in 23 was the founding or the establishment of the UKAI Safety Institute AC.

15:41

And this was so, so crazy.

This was crazier to me than trap GPT because genuinely in 2020, if you talked about super intelligent AI systems taking over the world, people would look at you like you’re an utter and like, you know, watch maybe too many Terminator movies or something.

15:56

Then ChatGPT comes out and suddenly everyone gets really spooked and everyone’s really worried it’s going to come up.

And the people who are familiar with systems going, look, ChatGPT isn’t super intelligent.

That’s particularly not 3.5 when it came out in November 2022.

But what was so strange is I’d be going to these effective altruists meetings and events and there was no set career path for careers in AI safety because this was a fringe interest that existed largely at the universities, predominantly within this community.

16:21

And now I’ve been to since and you’ve got career guidance workshops of people working at the AI Safety Institute in the UK, right?

Because just in the past year is so dramatic, like we went mainstream and it’s really weird to clock into that.

It’s so crazy and to that actually that’s a question that I had for you because things have changed so dramatically to the sort of fringe area that you’re studying to now.

16:43

I mean, you said there’s now departments within the government that are hiring people because they now recognize that this is a real risk.

I’m just curious, how can AI destroy the world and what should we all be aware of in this context?

The two examples that I asked, yeah, there’s lots of but OK, 333.

17:03

Oh, this is like 3.

Yeah, for sure.

Now 33 examples to keep you up at night, for sure.

Biological weapons formula concoction is something people worry about.

So people at the AI State Institute are working on guard rails to make sure that foundation models cannot tell you how to concoct a biological weapon at home and synthesize them because that will enable widespread bioterrorism, which would be terrible.

17:23

Cyber attacks is another big worry because if you can hack the cyber attacks of, say, power grids, you can shut down the energy in a city.

And that can be enabled for all sorts of other terrorists and all sorts of uses.

You know, another one is deception.

You can imagine large language was doing before, but you pair these things together and you can imagine a deep say of a Russian general official who might be commanding their subordinates to attack a region.

17:48

That’s incentive, right?

Can imagine terrorists hacks that and you can escalate conflicts in the war and you great and they can be done at very high speeds and if you have them wearable to do so, this can be really catastrophic.

These are all just actually bad actors with something less than super intelligent.

But the real thing that effective altruists tend to talk more about is a super intelligent system with a relatively benign goal.

18:08

So I think the example is paper clip maximizes where you just tell something to make as many paper clips as you can.

And in making as many paper clips as it can with its super intelligence, it ends up setting up security systems and locking down all the regions of forests and taking over government bodies to ensure that no one can infringe on them, and just turning everything it physically can into paper clips that.

18:28

Is a thing.

This is more speculative because we’re talking about a much more intelligent system that will ruthlessly operate it’s very stupid goal, whatever it is.

And the truth is, it doesn’t really matter what the goal is.

There will be some things that a super intelligent system will want to pursue as an instrumental goal to that goal.

18:45

So one of them is self preservation.

If you give an algorithm any goal and super intelligent, it will recognize that a big barrier to it fulfilling that goal will be it being turned off.

So it will take measures to ensure that it can’t be turned off.

And then the greater like spatial awareness it will have, it would use to ensure its security and ensure its power to, you know, ruthlessly or indifferently.

19:05

Yeah, indifferently.

I think it’s the best see through what to do now you put guardrails in there to prevent the worst outcomes.

But there’s an awesome, you know, sci-fi writers have been aware of a long time.

I, Robot by Isaac Asmos puts an article on the call there is called the the three laws of robotics is what’s called Yeah, of like do no harm, protect yourself.

19:25

And then on the third one is do whatever you’re instructed to do and you can’t do the best.

But that whole book is listed with examples of how you have catastrophes and crises while obeying those three laws, right?

Totally, Yeah.

Yeah, This is why, you know, philosophers seem to be attracted to this space because it’s very philosophical challenges.

19:42

How do you write all the conditions?

Yeah.

And that’s kind of my next question because I mean, you’re not going to be without a job, obviously, because you listen, you listen to this.

The challenges are infinite.

And we have hacking computers is one thing.

This is a whole other level of hacking, right?

19:58

The fabric of society, basically, and how you can just get machines to just completely destroy.

You need to have things in place that teaches AI what is wrong.

But then The thing is that the system believes whatever feeds the system.

20:14

So you can also have people saying that this is all right.

And so then who who is who owns the truth?

And so then that’s where ethics comes in, right?

So that’s.

Where ethics comes in, this is where the whole degree that you’ve now finished is so unbelievably important because it’s finding the fabric that can then put order to this.

20:33

And I just wonder how do you put order to this?

Yeah, that’s an interesting question.

So one way to classify these concerns is near term and long term risks.

So I want to reassure your viewers a little bit that these paper clip maximizes and super intelligence systems and general intelligence systems is not happening next year.

20:48

I think Elon Musk posted that AGI is happening next year.

That’s not true.

And commonly say like there are still barriers to what we call artificial general intelligence that are pretty significant.

And the big one is from this benchmark arc, which people can look up if they’d like to, It’s it’s much easier shown visually than explained sort of in words.

21:08

There are definitely nearer term safety risks.

And the way I classify safety versus ethics risk is safety is the stuff that everyone sort of agrees on controversially would be a terrible outcome, like a energy grid being shut down or by a weapon being simulated independently or mass unemployment, Right.

21:26

Actually, I should I take that back.

There are a bunch of libertarians who would say mass unemployment is just a free market having fun, right?

So that’s already drifting into ethic for that reason, because I wouldn’t even consider that safety.

But and then ethics is where the debates happen.

You know, where the grey areas is of should this be a harm?

Is this group that’s being harmed the group that’s in need of the most protection?

21:44

Or are there other groups who’s protecting?

It would come come at the cost of the the social difficult stuff is in the ethics and the technical difficult stuff is in the safety because in safety, the standard approach is mechanistic interpretability.

Well, mechinter, which is just you’ve got this massive machine learning system.

22:00

If it’s generative, I then it will be a transformed model and and you don’t know how it works.

Like you can open source it.

You can see a whole bunch of code, but there’s an enormous amount of layers of neurons that you don’t know what those layers are doing.

In particular, the hidden layers are considered the black box layers.

22:16

So there are techniques like heat mapping where you try to visualize what they might be doing in a way that’s more interpretable to a programmer.

And you can have breakthroughs of this sort explain ability is like the level above interpretability.

And this is quite popular with large language model where you actually have these systems explain their reasoning as they go.

22:35

There’s challenges about how reliable these self generated tools are, not because we worry that it might be lying in a human sense, but we worry that the way it’s generating language isn’t the same as the way a human would.

And you can worry about meaning not being in the loop in in the right sense.

22:51

But to classify AI ethics throughout.

So you’ve got long termist and near termist.

I also make a distinction between using AI versus building AI.

So one of the things I’ve been working on with companies is prompt engineering, doing workshops and sessions where you try to use ChatGPT called 3.5 perplexity.

23:09

Some of the image generators better and better typically means with a lower hallucination rate, which means they’re less likely to make things up.

Because of course, if it makes things up less, that is better both your own performance and ethically, because you can imagine all sorts of ethical compromises and problems can arise with an algorithm that isn’t doing what it’s meant to be doing.

23:26

And then for building AI, there’s this big question around participation and compensation.

I was talking about these artists who and you know, unions, they’re very keen to get them compensated and a lot of shadow and ghost work happening that doesn’t get credited.

But also there’s dimensions in which these systems aren’t included.

23:44

So I think mid journey this image generator in its initial outings couldn’t generate images of elderly black women.

They were fascinating, yeah.

Because they were a minority on in the training data set.

So there was a cool organization called We and AI which had an artist who I met who was taking photos of herself and uploading it to mid journey.

24:05

So to increase representation manually.

But of course, this is where the ethics come in that stop being a problem eventually, probably partly due to personally met, but now you have the capacity potentially for misrepresentation unless people are really on it with the guardrails, which is another human task in manually operating because you can create all sorts of stereotypes of harmful representation of this person.

24:27

So, you know, there’s a sense in which ethics involves more trade-offs than solutions.

But you know, there is a perfect world where you have really robust guard rails and representative data sets.

It just takes enormous at work.

And you know, that’s the, that’s the job.

So at that point actually, because you can, you can see so many just even from our conversation and clearly this is a snapshot, tiny little grain of sand in everything that’s going on in the world of AI.

24:54

But you know, you’ve, you’ve provided a bit of a picture as to just the different avenues and the different areas that we need to be concerned about and can work in.

And I’m curious, just from your point of view, what gets you excited and what are you passionate about?

25:09

You know, what do you see yourself wanting to do?

And of course, you’re on a journey, you know, you’ve just come out of university.

What your life looks like in 10 years is going to be different in 20 years.

And I don’t even think that you need to know what your life is going to look like at this stage.

25:24

But today, Joe Fennell, what do you see yourself wanting to do?

Like what gets you excited in this area?

I I try to do everything to be honest.

I have a list of things I think that I’m trying to pursue all at once.

So I know your organization does trips for business people to do meaningful work and build teamship behaviours.

25:44

I just come back from Kosovo, actually in the Balkans, and I want some UN funding to do upskilling in prompt engineering, which is using generative very better and also scam awareness, just making young people aware that deep fakes can do very compelling impressions of people’s faces now and also synthetic voices of celebrities and if they have access to your voice online, increasingly you, your loved ones.

26:06

And what was so cool about that project was in that boot camp, you had young Kosovans and Serbians and Albanians who are of a generation that can’t remember the war.

So they don’t have any resentment or blood against each other directly.

26:22

And actually, because there’s been so much development and money sent into it, they think of themselves as very modern and they think of themselves as wanting to move on from the past.

They’ve got real appetite.

So to be able to facilitate what felt like a future investment, peace building operation with these young people was really special, was really, really special.

26:41

I’m so I’m still in contact with them.

You know, I share as much resources from Cambridge on the stuff that I can.

It’s, well, I just, I fly over the entirety of the EU from the UK to Kosovo to, to do this.

I’ve been twice now.

So that gets me massively excited.

The other thing that gets me interested and excited is the degree to which AI is disruptive enough to level the playing field between the global North and global S Yeah, I’ve been told by 2050 the population of Africa is going to be enormous.

27:11

And there’s, you know, signs that they are undergoing doing industrialization and, you know, technological innovation on a scale that’s comparable to China in the 1960s relative to, you know, other global forming powers.

You know, Nigeria, for example, is increasingly becoming a really important global superpower.

27:27

They’re also a very, very young population.

I am very interested in investing in AI based tutors and education.

No resources for parts of Africa for a few years.

One is there’s sometimes less bureaucracy than there is is based in the UK.

27:44

There’s a lot of appetite and a lot of ambition, but I’m very nervous about this ambition because there’s hints of white saviorism and I would be doing it, but more more pressing than that.

The early versions of these technologies always have bugs and you don’t want to expose vulnerable groups for one reason or another to bug proof software is the worst with disabled people and transcription systems.

28:07

When people who had say, for one reason or another couldn’t take notes, maybe had a hearing impairment or a learning impairment, they weren’t able to take notes and they were just dished out these transcription systems.

I think it was in University of Georgetown.

And they ended up having to do way more work where previously they had human assistance note takers with them.

28:24

And they were those human note takers were replaced with AI systems way too soon.

I just meant these people had a much, much harder top line doing electric.

You would never want to recreate that sort of dynamic in the effort to use technology to bridge the gap between the global North and global S.

28:40

But still, I think there’s another side of that managed tension that could be could be really exciting.

Yeah, definitely.

It’s interesting.

I’m going as you know actually I’m going to be going to Denmark for running a a workshop there.

It’s with an innovation team for a large company and what I will be talking about a lot because it’s something that we do talk about a lot with Ty.

29:01

There’s a quote from an Ace Nim that I use a lot, which is we don’t see things as they are.

We see things as we are.

And that says a lot because what happens in any work in any other place.

So if you come from one place and you’re implementing work in another place, you of course will be seeing those challenges through your experiences in your eyes.

29:24

And until you actually truly understand the reality of those other individuals and work with them rather than doing to them, you’ll never be able to find the right solutions.

And so in order to be able to innovate, in order to be able to Co create, you do need to work very, very closely with whoever it is that you’re working with.

29:43

Then we’ve got so many examples, but one we worked with a an organization in Malawi and they work with fuel efficient stoves.

And one of the biggest problems in the energy discussion globally, even at, you know, the energy meetings at the UN, everyone’s talking about solar and everyone’s talking about electricity and everyone’s talking about gas or, you know, natural gas or whatever and creating stoves, for example, for people in the global S so that they can cook, but they don’t actually understand the reality on the ground for these individuals.

30:14

And so solutions that are actually what these individuals need are never being talked about because people are looking at the solutions for the global S through the eyes of the global north.

And so this what you’re talking about is, I mean, gosh, there’s so many people in Africa that I spoke to that would not have the COVID vaccine because of the history of people giving vaccines to people in Africa as Guinea pigs and testing vaccines.

30:42

And so they didn’t take the vaccine.

There were so many vaccines that were just sitting in fridges because the people just wouldn’t wouldn’t take it.

Having that understanding of one white savior.

Yeah.

I guess, you know, I think at the end of the day, it’s just if you don’t have that attitude and you work with people and you sort of are in the same levels, and that’s that’s not really a problem.

31:01

The problem is truly being able to take away that baggage and be able to immerse yourself into the reality and actually understand what people actually need and what are the solutions that they need.

And yeah, I mean, fascinating because obviously there probably are solutions that can be found that can help, but it’s only working with the individuals that you’ll be able to find it, right?

31:23

Yeah, usually.

And on the issue of reflecting what people really need, AI has been, I think really misused.

I mean, AI has been applied to lots of problems where it’s really not well suited.

Often soon, definitely.

Companies don’t.

Have the capabilities to be able to use it.

I mean, that’s the other thing, you know, you implement processes or you implement systems, but at the end of the day, people need to be able to manage those systems.

31:46

Can they continue it?

You implement it, but then can they continue it?

Do they have the education?

Yeah, If we should stay on large language models, I think the area that I’ve been looking through recently is like the effective GitHub Co pilots and these coding assistants on the like software development landscape.

32:02

And it just seems like lots of the demos for in this industry have been really misleading from companies present.

And actually software engineering as a vocation is really safe, and it’s much more safe than I think people with investments incentives would make out that these like coding assistants can replace good software engineers and companies.

32:22

So we are coming to the end, but I did want to ask you about the work that you’ve been doing with AI and medicine and the medical profession.

Now, my family, obviously, as many people know, my dad was a doctor.

My, you know, so many people in my family were physicians.

32:39

I mean, if my dad was still alive, Oh my God, he would be wanting to listen to this conversation because he was such a tech guy and loved, you know, he was an early adopter for everything.

If he was still with us and knowing just how AI can be used in the medical profession and, and from a safety point of view, etcetera, he would have been all over this.

32:59

And I’m just curious, what does this even mean?

If you could just give me and our listeners just a little bit of an idea as to what does AI look like in the medical profession.

Okay.

Would you like a more positive take on that question or a more concerning take on that question?

33:15

And you do both.

The positive take has to do with efficiency and outsourcing.

So for example, mental health and lowness crises are something that’s pervasive across the country.

If there was a version of a chat bot that could be licensed within a particular branch of therapy, maybe it’s cognitive behavioral therapy or another type of therapy.

33:38

The interesting thing about therapy is you don’t need qualification of the same way to be a doctor because you’ve wanted different schools and different philosophies of therapy that could be replicated that was meaningful to people at scale.

That might maybe, maybe maybe this is in sorry, even though this the optimistic take, this is a somewhat speculative optimistic take.

33:57

Alleviate the mental health crisis that we’re in now.

The immediate flip side negative of this positive is in reality, the majority of social I bought some development are private and they’re for profit organizations, which means that you have these romantic partner systems that are designed to be maximally attractive and attentive and addictive to you.

34:14

And they are not trying to improve your mental health.

They are trying to make money out of you and, you know, get you hooked on certain services they then put behind pay walls.

So I would be really fascinated as a project in a nationalized AI therapist chat bots that was designed at least to treat you as a patient or a citizen rather than as a customer, because it’s at least missing in the markets.

34:36

And I think it would just be good to have a horse in this race of this kind, right?

And it might maybe it’s useless, you know, maybe chat bots, as long as you know they’re not conscious and you know, you have sense, you’re not really being listened to, aren’t any good to anyone.

But on the off chance that they are, I would appreciate a bet on a chip on on that tab maybe right.

34:55

And also on the spirit of efficiency.

There’s hopes that there can be like leaner things.

So one thing’s I went to a forum Jesus College in Cambridge recently all about social AI and I sat on the table talking about social AI and healthcare.

I spoke with healthcare professionals and one thing they told me is they spent a lot of time discharging patients and like paperwork.

35:14

So there’s a hope that large language models can do a lot of that paperwork if it can do it reliably in restricted formats and just free up doctors times to patients and do things that really matter.

There’s hopes that you could maybe speed up waitingness with this as well.

Of course, that’s all a lovely vision, but there’s no shortage of ways it could go horribly wrong because early technologies have habits of making life more complicated and horrible for everybody, right?

35:40

So one of them is accountability.

So you can imagine doctors are typically given indemnity insurance, as are hospitals in case things go wrong, at least in the UK.

In the American system, it’s a bit nastier where they can really like go after in US to like preserve security of the of the company.

35:57

But you can imagine that AI agents, if they end up effectively getting indemnity assurance in some way, might compromise accountability in case some version of healthcare treatments would go wrong.

Or if something becomes overly cheap.

You can imagine how you know, making the whole system more complex might dilute accountability, which is something that’s, you know, not unfamiliar problem in like massive healthcare systems or massive companies.

36:23

You know, when you have massive complex systems, accountability becomes harder.

But you can imagine how thousands of AI agents added into this port might complicate the matter further.

And particularly if everyone’s suing everyone is seems to be the case in the American.

So the other one is in the same way the unguard railed AI can people can use to create bio weapons, they can also use them to do biohacking and administer their own healthcare advice, right?

36:44

And you can imagine people being drawn, thrown atoms and silos and deciding that their chat bots, particularly if it’s a company that’s marketing itself as extremely trustworthy, knows their health just as well as their doctor does, but without the waiting times, how you can have this terrible misinformed medicine all the time and or at least it’s biohacking.

37:02

Then you can imagine people really able to do biological interventions on themselves in the cells in a way that’s, you know, not maybe as pernicious as what a terrorist might do, but certainly dangerous.

AI is an interesting thing to talk about because you switch which metaphor you’re using to describe it.

37:18

A lot of time when you’re talking about jobs, people tend to describe it as like the creation of cars again, and they talk about how it affects jobs the same way cars affected, you know, the horse riding industry versus the mechanic industry.

When you’re talking about like super intelligent systems, they get compared to nuclear weapons.

Sometimes they get compared to the dawn of electricity with biohacking.

37:35

It’s it’s really quite novel.

It’s hard to imagine like a tool that can empower people to alter like, I suppose the Internet.

The dawn of the Internet is the nearest example is when you people have access to enormous amounts of just information in theory, but the difference is going to be AI systems can personalize and make it practical for you relative to what you want to do.

37:55

What previously you had was briefly smart people had to do on Yeah, or everybody.

It is, unsurprisingly, highly distributed intelligence.

Now, there’s lots of ways in which it’s very unintelligent.

So this can be really, really compromising in its own way.

But if people overly trust unintelligent systems, then they have horrible consequences relative to their own goals.

38:14

If they trust appropriately very intelligent systems, but have, say, horrific ambitions with doing harm to their own bodies, and that’s also going to be terrible for all a matter of reasons.

But it’s a it’s a paradigm shifting technology either way.

Yeah.

It’s fascinating.

Is there anything that I haven’t asked you that you’d like to tell our listeners?

38:35

Yeah, I think I want business leaders to know why AI ethics is important for individual organisations.

Yeah, nice.

The immediate reason for you as an organization is because if you’re on top of AI ethics, you are by definition ahead of AI compliance.

38:51

And I am expecting no shortage of lawsuits and copyright infringements and regulatory headaches for people in the world in adopting AI overly hastily or in properly.

And I really feel that you can save yourself a lot of compliance and legal strife if you have some serious ethical philosophical proactivity with what you’re doing.

39:12

That’s partly because lots of the regulations are still coming down the pipe.

In the UK we have a regulation light approach.

In the EU that’s not the case.

The EUAI Act, they have actually quite a lot going on with compliance.

But also this related to a second reason.

It makes the project one more striving for a positive vision rather than catastrophe management or catching up, which I think is more pleasant just to do and maybe more inspiring rather than a more panic inducing.

39:37

And the third reason is AI ethicists.

Ethics is a type of philosophy and AI safety is something shared between mathematician, computer science and philosophers.

I think like particularly because AI systems aren’t close to general intelligence yet.

Philosophical thinking and philosophical practice is a good thing to get in the habit of doing, especially at the culture level, because it should mean that you’re more resilient and adaptive to these like paradigm shifting events like.

40:02

Technologies, but it also means that you might be a bit more innovative and creative.

And I think there’s there’s enough subject matter in AI ethics to really get those juices flowing.

And it’s really, it’s really interesting.

I’m just wondering from the point of view of what you’ve just said, what could that look like?

So, for example, you are you a consultant?

40:19

For, for, for sure, for your ideas, I really focus on practical AI in the sense of using ChatGPT in your workflow more effectively, which means lowering the hallucination rate to as low as you can, knowing what it’s good for and what it’s not good.

I’m quite rare in that respect in terms of almost the entirety AI industry is it’s on building AI systems that are more bespoke and I’m trying to target the neglected problem of using them in effective ways.

40:41

For your workflow, I think it’s best to work on a few different levels at the same time. 1 is, as I say, performance and ethics because hallucination has that overlap.

The transparency with the systems and interpretability also has that overlap.

Compliance with the regular scene also has that overlap.

Another big one that I like to work at the cultural level is anti complacency strategies because there’s a feeling as soon as you introduce something like tragic either, oh, I can just let this thing do my job for me.

41:05

And then performance can slip when actually there’s a massive opportunity cost to that.

Because what something like a large language model can do is help you get to a bad first draft of a template you might be working on.

And one of the tricks of human psychology is it’s much easier to edit a bad draft or correct someone’s wrong answer to a question than it is to fill a blank page or answer a question yourself, right?

41:29

So it can be very stimulating in that respect.

And I really like to say idea of generative AI as a productivity hacked by generating stimulus, But again, that varies depending on what you’re doing.

So I’m really enjoying 1 to one practically AI sessions where we use GPT of 4034 these purposes.

41:45

And as I say, I’ve got the prompt engineering techniques to to memory to know how to get results that are just better in general, as I say, which the standard metric for that is low hallucination rates, but often it’s more bespoke.

Lots of people don’t know that with G ChatGPT Pro, you have custom instructions, which means you can improve a little blub to give you personalized results in an ongoing fashion for every chat you open.

42:07

Of course, you can sort of train your own GPTS with more advanced models to give you like hyper personalized effects.

But suppose people who aren’t willing to pay for that just to the free model.

You can recreate little personalized responses with every conversation you open.

And again, one thing I’m really enjoying is making that AI hype real to people by merging those prompting techniques to their like actual workflows in their day-to-day.

42:26

So that’s what I’m.

That’s why I’m really.

Great.

So you do workshops and you you can go into a company and do a session to help people understand about this kind of stuff.

Yeah, I’ve done both actually.

So I’ve done workshops for people at the group level and I’ve brought mentors in as well to do like Gemini tasks.

But recently I’ve actually just been 1 to one sessions.

42:42

What I’ve just been saying is, OK, we’ve got this hour, what have you got to do today?

And I’m just like looking over your shoulder saying, OK, that’s great, here’s this tool.

Perplexity is better for this like research based task.

And your in house ChatGPT system will be better for recreating this internal comms document, right?

42:59

And again, always Fact Check everything because it has app a bit of hallucinating.

But you know, that’s something I’ve I’ve really been liking.

Well, it’s something where I feel like I’ve really grown a lot of quite unique expertise.

Yeah, fantastic.

I mean, I’ll include your e-mail address.

Do you have a site?

43:15

Yeah.

It’s Joe Fennel.

It’s Joe fennel.com.

Joefennel.com so great.

We’ll include your your website and for the listeners, do reach out.

It’d be hard to find somebody as experienced as Joe.

Joe, thank you so much for joining me.

This has been amazing.

I’ve absolutely loved it.

So thank you for your time and we will definitely speak soon.

43:33

Thank you, Philippa.

See you soon.

Bye.

Take care, bye.

Hey everyone, this is Philippa again.

I hope you enjoyed listening.

Now this is your chance to get involved with Ty.

If you’re looking to create better leaders, better companies, and a better world, that’s just what we do by helping leaders tap into their greatest asset, their humanity.

43:56

We have a number of corporate programs that impact a range of people, from individuals at a company to 500 people around a business.

Or check out my book, Return on Humanity Leadership Lessons from All Corners of the World.

You’ll find the answers to how business can truly become a positive force while remaining at the forefront of competition.

44:18

You can find all the information you need on all of this at thaileadership.com.

Get in touch and I can explain more.

A huge thanks to Bejan Naviera for Co producing this with me and for creating the music.

I hope we’ll meet up again soon.