December 11, 2025 • 42 mins
Jon discusses his experiences and insights from attending a recent Society of Automotive Engineer (SAE) event focusing on AI and its applications in automotive design and racing. The episode details how AI is revolutionizing automotive engineering by improving design processes, enhancing efficiency, and enabling engineers to utilize their domain expertise better. Key examples include AI's role in Formula One aerodynamics, engineering intuition, and the development of more efficient electric motors. Summers emphasizes the potential of AI to exponentially speed up innovation while noting the importance of properly guiding AI with accurate data. The episode also examines the broader impact of technological advances on motoring history and the automotive industry.
===== (Oo---x---oO) ===== 00:00 Event Recap: Society of Automotive Engineers 02:43 AI in Automotive Design 04:17 Formula One and AI Applications 07:14 AI's Role in Engineering Intuition 09:38 Case Studies and Real-World Examples 17:13 The Future of AI in Engineering 24:56 Audience Interaction and Q&A 41:20 Concluding Thoughts and Sponsor Message====================
The Motoring Podcast Network : Years of racing, wrenching and Motorsports experience brings together a top notch collection of knowledge, stories and information. #everyonehasastory #gtmbreakfix - motoringpodcast.net
- More Information: Visit Our Website
- Become a VIP at: Patreon
- Online Magazine: Gran Touring
- Follow us on Social: Instagram
Jon Summers is the Motoring Historian. He was a company car thrashing technology sales rep that turned into a fairly inept sports bike rider. On his show he gets together with various co-hosts to talk about new and old cars, driving, motorbikes, motor racing, motoring travel. Copyright Jon Summers, The Motoring Historian. This content is also available via jonsummers.net. This episode is part of the Motoring Podcast Network and has been republished with permission.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
John Summers is the motoring historian.
He was a company car thrashing technologysales rep that turned into a fairly inept
sports bike rider hailing from California.
He collects cars and bikesbuilt with plenty of cheap and
fast and not much reliable.
On his show, he gets together withvarious co-hosts to talk about new
and old cars driving motorbikes,motor racing, and motoring travel.
(00:35):
Good day.
Good morning, good afternoon.
It is John Summers,the motoring historian.
I went to another one of thoseSociety of Automotive Engineer
events in the, uh, in the valley.
Sen. Once again, thank you for invitingme to, to the event and listener if you're
interested in, in this kind of technologyfutures niche that I'm falling down,
(00:57):
which as a car lover is a bit depressing,but if, if you are interested in it,
there are other episodes in my archive.
Cover Lidar and Vallejo, wherethe people that talked about that,
there was a German software housethat talked about the software
defined car they were involved in.
How can you make money off all the datastreaming off the car That's lurking in
(01:17):
my archive, the software defined car.
There was also the safety people,and that one's auto live with
a company that spoke to there.
And that's, uh, it's in thatheading in my, in my archive here.
I was also the keynote speaker at theAutomotive User Interface Conference
that was at Stanford in 2024.
(01:38):
And look, I'm, I'm mentioning allof this because I've been looking
at this future stuff thanks to phe.
I've sort of seen how it's happening, PHEand the Society of Automotive Engineers
and, and this is another one in thisseries of me trying to understand.
How Automobility is is changing, I think.
(01:58):
And, and if you are listeningto the pod, I hope I'm gonna be
able to try and convey that now.
In the past, it wasn't clear tome that the title that span, who's
the SAE rep, gives the events inorder to encourage people to come.
It's slightly different from.
Who the people actually are andwhat they may have actually crafted
the presentation originally to do.
(02:19):
So what do I mean by that?
I, I mean that when these peoplecome and present, usually they're
doing a presentation abouttheir company, what they do.
You know, I work for Vallejo andthis is the leader products that
we do, but what's so skillful aboutthe way that spend packages it up?
Is the, well, let me give youthe title of last night's event.
(02:43):
It was productive adoptionof AI in engineering design.
And the guy who was presentingto us was, uh, I think the CEO of
a company called Neural Concept.
His name was Thomas v Sharmaor v Sharma or something.
He was, it was a veryGerman name, but he was.
(03:05):
Had a French accent, very excellent,excellent English, but had a French accent
and had studied some college in Luanne,which is like well known as being like
basically Luanne Polytechnic, but it'slike, you know, world leading Swiss, like
clever engineering kind of a university.
So look, let me try and sum upwhat I learned first of all.
(03:27):
So if you've got a limited attentionspan, you can learn the important
bits and I can fill in all thedetails and share the notes later on.
So the important bits here is the, this issaying, okay, so how is AI actually being
used to develop the cars and the roadsthat we're actually going to experience
(03:48):
and have to encounter even if weourselves want nothing to do with ai and.
AI powered cars and all that.
It's, it's, you know, so on thereto understand for us, lay people to
understand how the world is changing.
Let's understand the mechanics of,of what the future of driving and
motoring is, is gonna look like here.
He had.
(04:09):
Three use cases, and I'll lookat my notes in, in a moment.
But really, I think this is thesort of takeaway a around it.
I mean, he had more examples, but oneexample and the, and and an example he
kept on coming back to was, uh, formulaOne aerodynamics and how when you use ai.
(04:30):
Instead of exploring one idea, youmight be able to explore five ideas and
therefore, you know, in, in the week thatyou've got or in the hour that you've got.
And he had a lot of metricsaround just how quickly you can
use AI for design iteration.
(04:52):
The other thing that I thought wasreally interesting was he talked about
at the moment, engineers need intuition.
Their research is guidedby intuition with ai.
Our AI tool will help themprobe in the right direction.
So in other words, if you'rea Formula One team, the data
(05:12):
streaming off the car, the AI.
Can help point you in the rightdirection rather than you needing
somebody like Adrian Newey to beable to say, aha, this is my vision.
This is where I think we should be going.
And then the data can be used to craft andhelp you move in that kind of direction.
(05:32):
So I think there's a subtle nuancethere, and I hope I've made that clear.
The third.
Element was ingest workflows.
This is, in other words, you know, youhave a bunch of different engineering
processes that lead to the finalprocess, and what you can do is you can
(05:52):
chuck it all into the AI and say, dude,how can you make this more efficient?
And then it will, one of the examplesthat he had, it may not have been
this specifically, but it illustrates.
Kind of complicated problems that he'sworking on in the automotive space.
All these EVs, theirbatteries need cooling.
There's these cooling pads and the coolingpads basically feature water moving
(06:17):
around like some kind of coolant movingaround under the cooling pad, right?
Well, the way in which youcan make the water move.
You know how it swigs and squawks,how it zigs and zags, what you do
to make it move in a particular way.
This has a very meaningful impact onhow effective the thing is at cooling.
If you think about the way that atraditional car radiator works and
(06:41):
how, if you know an old one that'sgot a few stone chips in it and.
Bashed and you know, it, itnoticeably doesn't work as well.
Right?
That might be why the caroverheats on that hot summer day
climbing a along hill, right?
So it's those kind of, uh, of ideas.
So once upon a time, what an engineerwould've done is be like, well, I know.
(07:03):
From my experience designingother cold pads that if I make
it do a figure of eight motion,say that worked well in the past.
So that will be thebeginning of my research.
So in the first instance, right, whatAI is gonna be able to do is help
me iterate on that five 10 x fasterby doing lots of different models.
(07:23):
But what he's saying also isthat I needed to be know that
figure of eight thing what?
We can do now is have the AI suggest,you know what, if you do zigzags
all the way up and down zigzag,crazy zigzag, that might be better.
Or, you know, how aboutPacman a Pacman system?
(07:44):
You know, where he goes around ina, in a Pacman kind of pattern.
That might be better.
Well now it's not it,it's not like this is.
Better than the figure of eight.
It's the beforehand you only had thefigure of eight method, now you've
got the figure of eight and thePacman and the zigzag and whatever
other stuff the AI's gonna pullout of its computer, silicon ass.
(08:06):
Um, so that's kind offascinating, isn't it?
So that was the, the important bit.
So let me give you some,some bullets around that.
So how did he call it?
He called the first one.
We can capture the physics.
For the Formula One team and somebodyin the audience was saying, you work
with, you know, Alfred Cent or whatever,Toro Rosso's called Racing Bulls,
(08:26):
whatever Toro Ross is called nowadays.
Anyway, he was saying they'vedone a lot better recently.
I don't know if they have anunfollow Formula one, but did
you have something to do with it?
And the guy was like quite smug,but obviously felt as if he did.
So that was really cool to experience.
So that's case one.
Capture the physics, right?
Case two, is this a, a designbecomes more efficient, right?
(08:49):
And the example here was Marla,the piston people who were now
developing electric motors.
Of course, the motor that they developedusing AI was 15% more efficient and
three decibels quieter thanks to to ai.
You also talked a lot about how theengineers then went back to see what
(09:10):
the AI had done that made it 15%more efficient and three, so in other
words, there, oh, an engineeringintuition was being informed by the ai.
This symbiosis is really weird, isn't it?
We've seen it before, right?
Initially, the car wasjust a way to get around.
Then suddenly it gave us the suburbs andsuddenly it gave us mass consumer credit.
(09:33):
Didn't it?
Suddenly all that stuffhappened in the twenties.
Right?
But it's just fascinating stuff, isn't it?
The example for the ingestion ofworkflows, and I thought this was amusing
given her the lead, our presenter,um, was Vallejo and the idea of, of
their product is to unite these sortof three efficiencies, uh, together.
(09:55):
So that was the sort of cliffnotes, I guess a lot of people.
But I mean, yeah, you can turn off now.
I mean, I does say a lot of peoplewould've turned off 'cause of my
useless prattling before then,but that's my, uh, my attempt to
do, uh, to, to do a sum up there.
Let me move through my notes now.
Flashing out what we just talkedabout, let me also mention that.
(10:16):
Span whilst we were there, had a coupleof giveaways sa, international Edge
Artificial Intelligence in Connected,cooperative, and Automated Mobility.
I haven't even had a look atthat yet, but it's like this
is like a pamphlet kind of guy.
I guess it's more of that softwaredefined car stuff that I've talked
(10:38):
about in other presentations.
This is definitely this materialedges right towards my know
the enemy kind of thing.
Understand how they're deve inventing.
Cars and driving.
Now that's interesting, right?
I that I funny that I thought ofthat because they laid on nice food.
Thanks for that, Stan.
And, uh, after I scarfed the nice food,I was, uh, hovering about to, uh, to
(11:04):
sit down and, uh, somebody struck upwith me and I thought I recognized
him and he was like, I, I know you.
And it was one of the people thatI presented to when I went and
presented five great BMWs to BMWs.
Group of designers in, um, Silicon Valleythere, he was one of the people that had
had attended, and in fact, I was thinkingof him just recently because he had been
(11:28):
involved in the development of the BMWS1000 RR Uber Sports Bike, and we talked
about how his insight for me was how.
MW Faithful were all aboutthe tele leave of suspension.
And of course the S 1000RR was away from that.
(11:48):
It was, we're just gonna do thingsthe way, basically the Japanese do
it with the upside down conventionalforks and you know, BMW will.
Engineers felt that their telelever solution was better.
So there was resistancearound that internally.
So, uh, the S 1000 RR had to be aknockout because just internally
(12:10):
it would've crippled the careersthe people involved had it.
So he's passionate about the S 1000.
RR was really pleased that I'd included itin my like list of like five great BMWs.
He showed me he'd recently been to Munich.
There was a BMW Design classic studio,and he was scrolling through the photos
(12:30):
that he had until he landed on the.
PK Era Bra and BMW Formula One Car,which had been another one of my
great cars that I just felt showedlike BMW's immense design superiority
twice in Formula One with thatbra and BMW of the early eighties.
This was a. Basically A BMW 2002,you know, 1,602 block with a special
(12:56):
head on it that could take the turbo.
And those crazy statistics, like1500 horsepower statistic that you
always hear being bandied aboutthe Turbo Era Formula One cast.
That was the BMW in line four.
That produced that statistic.
My understanding is as a flashreading, so you know, certainly
(13:16):
centers Lotus Reno of the mid eighties.
They reckoned 1300.
1350 in qualified peoplemay know more now.
I've not been on the forums.
That's a, a different kind of rat hole.
But yeah, so I was chatting with thisBMW fellow, he was showing me these
pictures of these awesome classicBMWs and saying that he'd thought
of me and I was musing on my, uh.
(13:38):
Role as, as influencer and I, aswe were stood there span was like,
you can take all these books away.
Well, people rushed up andtook away all the books.
So the Edge Artificial Intelligencepamphlet was the only thing that I
got from the pamphlets, and the onlybook that I got was this Biblical
looking tome, which is entitledTransport Transitions, advancing,
(14:03):
sustainable and Inclusive Mobility.
Proceedings of the 10th TRA conference,2024 Dublin Island volume two.
Sustainable transport development, andnow this is full on no thine enemy stuff.
This is, if you're familiar with theway that modern cities are creating like
(14:25):
exclusion zones by putting bollards and.
Corralling the car corrallingcars into, uh, lanes that
are wide enough for the cars.
Just, but you know, that extra spacethat they used to be for you to
like cut the corner a little bit.
Now there's room for pedestrians tostand in the road or in the case of San
Francisco, FAL attics to fall in the roadin that space, and they put these annoying
(14:49):
bollards around and, and all of that.
That's, these are the people.
They were creating the theory.
But behind that, another one of thesepods that I did when I was asked to do
a keynote at the Auto User InterfaceConference, uh, in 2024 last year,
I, a number of the sessions that Iwent to focused on, on this kind of
(15:13):
sustainability and, and, and so on.
It really feels to me a lotlike, well, think about the way
that cars of the 1960s were.
And then compare them with the way thecars of the 1970s had those big ugly
bumpers and big pillars, and were heavyand were unresponsive and de-tuned
(15:38):
and, and really the whole freedom ofdesign and freedom of expression and
the sheer joy of motoring it left.
Didn't it at that time?
Well, that I feel like is what's happeningwith these people who are doing this.
It's interesting.
That should sit at thenexus of both of 'em.
(15:59):
'cause it's all traffic technology.
It's interesting that the, uh, AIsystems that have been developed on
Formula One or that, one of the otherapplications that this guy talked about
last night was this SP 80, which is the.
Waterland speed record for unpoweredvehicles, so for sale vehicles.
(16:20):
So this is like a clay foil thing.
I'll put a link in for it.
This was learning about, this is one ofthe most awesome things that I've learned
and, and seen about for, for some time.
So it's interesting.
There's this exciting sort ofmotoring adventure that sits at the
same time as this really dystopiansuffocating, the freedom of motoring.
(16:44):
I might even review this properly.
I might have a proper look through it.
I've not done more than flickthrough it, so there, I've,
I've prejudged it, haven't I?
What kind of an asshole YouTube,but am I to prejudge the, uh,
transport transitions, advancingsustainability and inclusive mobility?
My god,
(17:05):
enough for the preamble and thesetting up and the setting the
stage and, uh, and, and all of that.
Productive adoption of AIand engineering design.
So it's AI workflow processes.
So his, so the speaker's backgroundis that he developed some AI work
processes when he was at that lowsand college that I talked about,
(17:27):
which became the industry standard asneuro concepts about a hundred people.
Now, automotive is their strongest sector.
I've got a photo.
I mean, it's basically like an openbox and all the flaps of the box
are like all of you know the waysthat the product can connect and
the partnerships that they have.
And then the inside of the box is theirAI shit stuff out of the flaps of the box
(17:52):
with stuff like, you know, the Nvidia.
Net core network infrastructure,which I assume is something like,
you know, the Microsoft vendornetwork that I sold through when,
back in the days when I was a, a techsalesperson 25 years ago, I could be
wrong, could be talking outta my ass.
I've included the slide.
If you can be asked, go tothe notes and click on it.
(18:14):
11 OEMs in automotive,18 tier one suppliers.
They're also managing a growing expertcommunity, whatever the fuck that May.
The next was where we, we talked aboutthese three main pillars, and that was
the language that he used AI to capture.
The physics IE, what arrow package shouldI use In Formula one is the defined.
(18:38):
Requirement and you know, itcan really help with that.
And the example that I gaveearlier about the figure of eight
flow or the PackMan flow of wateraround a battery cooling pack.
You know, the same if youthink of it, aero swirl.
It's a similar kind ofsuper complex modeling.
In fact, another use case theyhad was for an architecture firm
(19:02):
who used to build these modelsto create the wind tunnel effect.
So that they could forecast a wind tunneleffect on pedestrians 'cause otherwise
in San Francisco's a bit like this.
And you know, a lot of American citiesare, the streets become, when they
have the high rises, it makes like awind tunnel kind of bacteria effect.
So it's really horrible walking as apedestrian, especially if it's raining.
(19:24):
'cause it blows the bad weatherright up in your face kind of thing.
Well.
Nowadays, rather than having tolike make the models and take two
weeks to like get results, nowadaysthe AI can do it from in 10 minutes
flat, so that's another application.
It seemed to me that was pretty muchthe same as how do you shape the
error foil on a formula of one card.
Not that different from, I'll come ontothe water speed record one in a minute.
(19:49):
'cause that is more complicated.
Well, I'll talk about it now.
It is more complicated, right?
Because it's, it's not justthe shape of the foil, it's
the flexibility of the foil.
It's not just the shapeand physics of the water.
It's also the pull ofthe kite on the foil.
Dragging it along.
And then there's the wind as well.
(20:10):
So that's a really supercomplex piece of modeling.
It's also the single coolestapplication and four oh one's pretty
cool, but this foil, SP 80 waterlandspeed record thing is so cool.
So just to be clear on that, Waterlandspeed record at the moment is 65 knots.
They're trying to do.
80. He was saying lastnight they've got up to 55.
(20:31):
And the little bit offootage of it, my God.
I mean, 'cause you all know 55 knotson water that's gonna feel like a
hundred miles an hour on land and it'srough, you know, anyway, whatever.
So I was pretty taken with, uh, with that.
Wind and foil and allof that kind of stuff.
That to me was coolerthan the formula one.
Anyway, that's pillar one,capturing the physics pillar two,
(20:53):
you know, capturing the physics,what should I do, kind of thing.
Pure use case.
Pillar two is the MLA electric motorexample where it's, you know, where they
built a more efficient motor more quickly.
And the measure of it was 15% moreefficient and three decibels quieter.
And then we talked also about,because the way their software works,
(21:15):
you can drill into what the AI did.
Your engineers know what it did, sothey can learn better next time and
have better ideas next time so theycan inform the AI better next time.
So, and, and you know, that'skind of obvious, right?
If you've used any ai, as you well know,if you put shit in, you get shit out.
If you asked it, who wonthe Battle of Gettysburg?
(21:36):
And why?
If you a good example, if you ask it,who won the Battle of Berrys burg?
It gives you some total bullshit 'causeit doesn't know what you're talking about.
It takes a guess.
It might be right, it might be wrong.
You have to be informed to understandthat I was to illustrate this and
this is where AI is at the moment.
To illustrate this, I was researchingearlier today a particularly
(22:00):
gruesome shooting incident inLondon's history where three
policemen were shot dead by gang.
Members, basically whatwe call now gang members.
1966, it was Harry Roberts was the mainvillain, but one of the other villains
was, uh, was this guy John Duddy.
(22:22):
Well, I was googling up John Duddy, butAI kept on wanting to give me answers
about John Duddy, who's some Irish boxer.
So that illustrates how, if I hadn'thave been on the case to say, no,
not John Duddy, the Irish boxer.
I mean, John Duddy and Harry Roberts,who were involved in this rather,
(22:43):
uh, unpleasant shooting incidentin, uh, in London in the 1960s.
The bad guys.
I don't.
They they, it was a standardvanguard involved in the shooting.
If you picture up Googleup a standard vanguard.
I mean, the cars as ugly as the, uh, asthe story and the people involved in it.
And it gives you a real feelingof this sort of guy Richie
(23:08):
Craze kind of underworld London.
But it was still shockingbecause it was a shooting.
I guess at that time, villains used likeknives or you know, one of the three,
the three that went to jail that didn'tactually shoot any of the cops, but
went to jail anyway, when he came outof jail, he was released in the 1980s
and uh, he was like living in a flat inBristol and his flatmate, who was a heroin
(23:32):
addict, beat him to death with a hammer.
I was like, God damnit,there's some retribution there.
But there isn't, right?
Because the main villain, it wasa really nasty piece of work.
This Harry Roberts, he's still alive.
They released him in 2014as far as I can make out.
The bastard is still alive.
All of the accounts talk abouthow he gloried in killing.
(23:55):
Apparently he learned to do it when he wasin the British Army on national service.
There were some likerebellion and he records.
He killed four people bragsin the Wikipedia entry about
killing four people there.
That's when he really.
Decided he enjoyed killing people.
My God, how did people become so twisted?
(24:15):
But anyway, back to where we were going.
And the third pillar, so oneis capture the physics two is
more efficient, more effective.
The Marlo electric motor.
Three is the Valeo exampleof ingesting workflows.
So the overall process of developingwhatever widget it is from cradle
(24:38):
to grave becomes faster and andAI informed, and then that offers
the same kind of circular flow.
Of learning that we talked about,in example, number two, Mar electric
motor, and this is where I'vereally gotta take my hat off to
spend for the formatting spend.
Then asked a question and this ledto a backwards and forwards with the
(25:00):
audience, which was really cool becausewhat I'm trying to understand and
what I'm trying to share with you, thelistener, is where are we with this shit?
Like, is this bloke talking aboutvaporware or is this something which
is actually gonna be on the next BMW?
Like, 'cause there's three guards fromBMW sitting right over there and the
(25:22):
campus were on like opposite acrossthe road from the building we were
in was Ford's base in Silicon Valley.
So there were probablyfour people there as well.
I mean, I don't know, nobodywas wearing a Ford badge.
I mean, everyone's just wearing like.
Engineer Silicon Valleyneutrals, aren't they?
You know, the Rohan pulloverand the uh, jeans kind of thing.
(25:44):
But anyway, yeah, fanhad a question about.
AI versus traditional simulationsand the IE. Why would I use AI
instead of a traditional simulation?
And the answer is thatAI can be predictive.
So it turns days of work intothe work of minutes or seconds.
(26:05):
It's an evolution of currentprocesses and makes them faster.
So it's not like you're chuckingout everything you did before.
Fuck on that bollocks.
We're using the AI now.
No, it's not that.
Rather it is looking atyour current products.
Allowing the AI to ingest theworkflow and then looking at what
(26:27):
efficiencies it might suggest, andI find myself thinking as I'm saying
this, the evolution of sports bikes.
And there's a jigsaw video, whichI'll add a link to, which shows
the evolution of the Suki S xr.
When you were living here, it didn'tseem like it had changed very much.
You know, 'cause each year they onlylooked a little bit different and
(26:48):
yeah, they were evolving generations.
But even when the generations evolved,you know, when they went to like water
cooling and fuel injection, you know,that seemed like an enormous change.
And I guess it was, but it'sstill an evolution, right?
It was still basically a G six R if yousat on the previous generation and you
get on, you know, the all new water coolfuel injected, when it still felt like
(27:12):
a big, heavy, scary G six R. A few yearslater, they'd evolved to kind of feel
like 600 had a few years ago, right?
So, so what I'm saying is that frommy like nineties, eighties, nineties,
geo six R through the early watercold, you know, it's away from the oil
cold ones to the early fuel injected.
(27:34):
Water.
Cool ones through to the K five, youknow, and the last one that I have,
which is arguably, you know, whicha lot of people feel, is peak analog
sports bike design, which does feellike a 600 easy to ride, like a 600.
Not as intimidating as the earlier bikes,although it's just as powerful, right?
That is a long explanation about howa particular design thing evolved.
(28:01):
Right.
What we're saying is that thatevolutionary process can be
incredibly powerful, right?
It seems like it's just smallchanges incrementally, but when they
compound, even if it's just from a1993 jigsaw to a 2001 jigsaw, from
a 2001 jigsaw to a 2005 jigsaw.
(28:24):
The changes are enormoustaken all together.
They're enormous.
The thing in the 2005 jigsawis nothing like the 1993.
I mean, it's still a sports bike.
It still has like the GSXR genes, but ohmy word, the thing has evolved an an awful
lot, so this is my way of underlining.
I guess the other example of continuingevolution is the Porsche nine 11.
(28:48):
And it's had itsrevolutionary changes, right?
But even then it's been like,oh, it's water cold now.
Like, oh, it's the 9 9 1.
Now I can barely look at it.
It's so big and fat.
Yet now you're looking at it andbeing like, actually it's all right.
And when you look at like an LG 50 car,you're like, wow, it's the same size.
There's a Volkswagen Beetle.
It's so titchy.
Right?
You know, you, you get used to thechange and the thing evolves whilst
(29:12):
the DNA re remains the, the same.
That is the evolutionary process canhappen much, much more quickly with ai.
Let's be really real about that, and thatI think is something that I hadn't fully
wrapped my head around until last night.
But to answer Spence's question, AIis not a replacement for simulation.
(29:35):
Instead, it's a way to make the simulationmore effective and happen faster.
Oh, this is interesting.
So like all AI clarity aroundunderstanding your goals in using
the AI delivers much better results.
So just like we know, the prompt is key.
If you ask who won the la, the Battle ofLake Berry Sea, the AI's not got a clue.
(29:59):
If you ask it who won Gettysburg, itcan give you a really good answer.
Now, the example that the people sittingnext to me were asking about was.
If I like, forget to tell it, tell theAI that the car's got a crumple zone.
Can it figure out after the crashhas happened by measuring the
(30:22):
data, can it figure out that ithad a crumple zone, or do I need to
tell it about the whole concept ofcrumple zones in the first place?
And, and the answer to that was.
AI cannot invent a phenomenonit hasn't been trained on.
So in other words, if you don't tellit about the concept of crumple zones,
(30:43):
it can't pull that shit out of its assand just make up the concept of that.
You have to say, well come back towhat I said before is you can't ask it.
Who won the Battle of Lake Berry Sea?
You have to know like Berry Seed's,a good place to do power boating
and Gettysburg was where there was abig important battle that determined
(31:05):
the outcome of the Americans of War.
The purity of F1 means that it is theperfect example and illustration of the
application of this ar, and it really was.
He kept, I would say he cameback to it three or four times
just because it's such a pure.
Illustration of ways.
Think about that.
There's a race in two weeks time.
(31:26):
You can't wait three weeksfor the data to come back.
You need to be modeling somethingnow, and it literally, if you've got
three ideas for what a wing mightlook like once upon a time, you only
add time to work on one wing design.
You have to like engineering intuition.
Take a guess what youare gonna work on now.
You can work on all fiveideas that you've got.
(31:46):
It's really, the impact is exponential.
That's what you end up feeling about.
It's really, really fascinating.
So on that theme of aiding engineers,intuition, the AI is capable of
looking at the data and then makingsuggestions about where to go.
(32:11):
And what it can't do is come up witha phenomenon like a crumple zone.
That he didn't know.
But what it can say to you is, oh,if you wanna avoid crashes, maybe
work on Thai technology ratherthan working on crumple zones.
That's not a very good example.
But you understand what I'm saying isthat the notes that I made last night was,
(32:31):
um, evolving from intuition led designevolution to data led design evolution.
AI can show the best direction to go.
It could prevent dead ends.
And I was laughing because one ofthe other cars that the chap from
BMW had been looking at at BMWClassic was something that he was
(32:52):
like, I've ever seen this before.
And I was like, good lord.
I remember it.
It was the Twin KeelWilliams BMW Formula one car.
I might be wrong on this.
Google it, find out more.
But I remember they fundamentally an ideato design a chassis in a different way.
It was a huge gamble, and if ithad worked, it would've given them
(33:13):
massive competitive advantage, but itdidn't, and it just put them behind.
You might even be able to make a case.
It was.
One of the things, if not the thing thatwas like Williams Spiral downward from
a front of grid team to a midfield team.
But I dunno, my contemporaryFormula one history enough to be
able to, uh, to comment on that.
(33:35):
But at least the Twin Keelcar is a case in point.
AI can do a bunch of work.
AI could have potentially told theWilliams team, guys, this is a dead end.
You don't wanna do this.
This is interesting.
Right.
He wanted to emphasize what his productdoes is enables your engineers to
(33:55):
utilize their domain experience better.
So he's saying, look, you guys know betterthan I do what prompts to put in to my ai.
I'm just saying my AI is gonna help youguys get better if you use it properly.
So I, but I thought that was interesting.
Uh, enable your engineers to.
(34:17):
Utilize their domain expertise better.
There were a number of mentionsof sort of scope of performance
games and, and 10 X seemed to bethe number that kept cropping up.
Now, I dunno how you cankind of measure that.
Maybe it's like, you know, MalcolmGladwell's 10,000 hours before you
(34:38):
achieve like symbiosis with the machineand what you're doing or whatever is.
10,000 hour theory is, Inever read the book sadly.
The point is that that's, they'renot suggesting that it's gonna
like, make you 50% more efficient.
They're suggesting really, uh,orders of magnitude and, you know,
(34:58):
I think it's a recognition of thatpotential that might explain why.
The tech companies are falling overthemselves to have the lead there
because they see the productivitygains that are are gonna be had.
I mean, the rest of us just seemass unemployment, don't we?
But there you go.
At least when people are unemployed, theyalways do productive things like phishing.
(35:18):
They don't do things like getinvolved in stupid revolutionary
politics, but we won't go there.
Uh, we talked a lot aboutthis concept of the black box.
What he said was, customersthink they want a black box.
You know, they just push it andtheir shit's more efficient.
That's what they think they want.
But actually when we work with them,they realize that what they want
(35:41):
is why I described when we weretalking about the M Electric motor
example right at the beginning, wasthat what they're looking for is,
yes, great, it's 15% more efficient.
But I want to dig in and see whatprocesses it did to get to be.
15% more efficient.
I wanna understand how it gotthere because I'm gonna be able
(36:02):
to apply that to other projects.
And if I, as an engineer and a teamof engineers can understand that,
that's gonna improve my engineeringintuition next time around.
So my prompts for the AI aregonna be that much better.
So it's gonna be this fly.
Effect, isn't it that, I mean, when youthink about what that means sort of down
(36:24):
the line for the speed and efficiencyof any and every engineering project,
it's kind of mind boggling, isn't it?
And there were a bunch of questionsthat were kicking around the topic
of, do you tell the model of physics?
Or do you let the AI look at thedata and work out the physics itself?
(36:45):
And I think if you come back to the rumplezone example, well, I think the answer to
that would be that you have to tell it.
If you built a rumple zone on thecar, you have to tell it about that.
If you are trying to set thewater speed record, you have to.
Tell it that.
Yeah.
The craft itself can bend a little bit.
It's not just aboutthe wind and the water.
(37:07):
The craft itself might bend a little bit.
You have to tell it allthe parameters, don't you?
In fact, right?
If you think about my example of JohnDuddy, when I first surfed on John
Duddy, I was using Google's Gemini.
What it gave me was the Irishboxer I, I then put it with
Harry Roberts in 1966, and then.
(37:30):
It was able to delivera meaningful result.
Uh, it was a question alsoabout how to prepare and clean
the data before feeding the ai.
And of course, this was a general,there was the general feeling in the
audience that, that was sort of thebread and butter of any AI project was
how you feed and, and clean the data.
(37:52):
Properly.
Then, uh, he wrapped up with two examples.
The first one we've talked a little bitabout already was the, we'll talk about
both of 'em a little bit already, but thefirst one, these Canadian architects that
do highrises and they use AI modelings toprove that the new building won't create
horrible wind tunnels for pedestrians.
So I talked about that as well.
(38:13):
The additional wrinkle that I forgot tomention earlier was that originally they
would just model for the wind tunnel.
But now they're able to add sun andshadow modeling and then they were able
to blend that and some other measurethey had that I didn't remember to
create a total pedestrian comfort.
Kind of measure.
(38:33):
It's quite interesting, isn't it?
You see the way the AI is able to notjust improve the wind tunnel modeling, but
also have all these additional dimensionsthat overall develop and deliver a much
more holistic, well for out solution,but none of it's possible without an
(38:54):
engineer giving it the right prompt.
Pretty soon it's gonna be, isn't it?
Let's, let's be clear.
I mean, we're all engineering our wayout of a job here, apart from these SP
80 CH water speed record fellows, right?
That's gonna be the future, isn't it?
Everyone's gonna be busy doingwater speed records and fixing
vintage motorcycles, aren't they?
(39:14):
They're not all gonna beinvolved in pointless politics.
I just wanna say as well, just withthis thought, right, the way the Romans
dealt with this was the corn doll.
And I thought of that whenAndrew Yang came along and
proposed his like minimum wage.
On the face of it, you are like,Andrew Yang passed me the bomb, man.
What kind of hippie bullshit is that?
And then when you stop and think aboutit, it's like it is the way the Romans
(39:37):
controlled their unruly populous.
So it's the Unpowered water speed recordthat these SP 80 CH guys are doing.
As I said, the, the recordnow stands at 65 knots.
Their goal is 80 knots.
It's a hydrofoil kind of design.
If you've not had a little lookat the clip or the thumbnail
(39:58):
image here, the physics of thefield is extremely complicated.
Because of the multiple factors,the kites pull, the resistance of
the water, the shape of the water,wind blowing across the foil and
the surface of the water itself.
The shape, stiffness and friction,coefficients of the foil itself.
(40:21):
So an interesting project.
And then to wrap up.
A really interesting thought that it justleaves your jaw like dragging on the floor
a little bit because I just hadn't thoughtof it in these kind of terms before.
But this is a good way actually to justwrap up my whole little session here.
(40:43):
It's that the productworks mostly off graphics.
Two dimensional or three dimensional.
It doesn't use words and numbers so much.
And in fact, when somebody asked aboutthat, he made the point that there
were many other applications that coulddo words and numbers better than his.
(41:04):
Like it was a bit of a, like, wasn'tit obvious that was was what we do.
Like we can do these two andthree dimensional objects instead
of just words and numbers.
The future, my word, the future isa mind boggling place, isn't it?
Thank you.
Drive through.
(41:32):
This episode has been brought to youby Grand Touring Motorsports as part
of our Motoring Podcast network.
For more episodes like this, tune in eachweek for more exciting and educational
content from organizations like TheExotic Car Marketplace, the Motoring
Historian, break Fix, and many others.
If you'd like to support GrandTouring Motor Sports and the Motoring
Podcast Network, sign up for oneof our many sponsorship tiers at
(41:54):
www.patreon.com/gt Motorsports.
Please note that the content,opinions and materials presented and
expressed in this episode are thoseof its creator, and this episode has
been published with their consent.
If you have any inquiries about thisprogram, please contact the creators
of this episode via email or socialmedia as mentioned in the episode.
John Summers is the motoring historian.
He was a company car thrashing technologysales rep that turned into a fairly inept
sports bike rider hailing from California.
He collects cars and bikesbuilt with plenty of cheap and
fast and not much reliable.
On his show, he gets together withvarious co-hosts to talk about new
and old cars driving motorbikes,motor racing, and motoring travel.
(00:35):
Good day.
Good morning, good afternoon.
It is John Summers,the motoring historian.
I went to another one of thoseSociety of Automotive Engineer
events in the, uh, in the valley.
Sen. Once again, thank you for invitingme to, to the event and listener if you're
interested in, in this kind of technologyfutures niche that I'm falling down,
(00:57):
which as a car lover is a bit depressing,but if, if you are interested in it,
there are other episodes in my archive.
Cover Lidar and Vallejo, wherethe people that talked about that,
there was a German software housethat talked about the software
defined car they were involved in.
How can you make money off all the datastreaming off the car That's lurking in
(01:17):
my archive, the software defined car.
There was also the safety people,and that one's auto live with
a company that spoke to there.
And that's, uh, it's in thatheading in my, in my archive here.
I was also the keynote speaker at theAutomotive User Interface Conference
that was at Stanford in 2024.
(01:38):
And look, I'm, I'm mentioning allof this because I've been looking
at this future stuff thanks to phe.
I've sort of seen how it's happening, PHEand the Society of Automotive Engineers
and, and this is another one in thisseries of me trying to understand.
How Automobility is is changing, I think.
(01:58):
And, and if you are listeningto the pod, I hope I'm gonna be
able to try and convey that now.
In the past, it wasn't clear tome that the title that span, who's
the SAE rep, gives the events inorder to encourage people to come.
It's slightly different from.
Who the people actually are andwhat they may have actually crafted
the presentation originally to do.
(02:19):
So what do I mean by that?
I, I mean that when these peoplecome and present, usually they're
doing a presentation abouttheir company, what they do.
You know, I work for Vallejo andthis is the leader products that
we do, but what's so skillful aboutthe way that spend packages it up?
Is the, well, let me give youthe title of last night's event.
(02:43):
It was productive adoptionof AI in engineering design.
And the guy who was presentingto us was, uh, I think the CEO of
a company called Neural Concept.
His name was Thomas v Sharmaor v Sharma or something.
He was, it was a veryGerman name, but he was.
(03:05):
Had a French accent, very excellent,excellent English, but had a French accent
and had studied some college in Luanne,which is like well known as being like
basically Luanne Polytechnic, but it'slike, you know, world leading Swiss, like
clever engineering kind of a university.
So look, let me try and sum upwhat I learned first of all.
(03:27):
So if you've got a limited attentionspan, you can learn the important
bits and I can fill in all thedetails and share the notes later on.
So the important bits here is the, this issaying, okay, so how is AI actually being
used to develop the cars and the roadsthat we're actually going to experience
(03:48):
and have to encounter even if weourselves want nothing to do with ai and.
AI powered cars and all that.
It's, it's, you know, so on thereto understand for us, lay people to
understand how the world is changing.
Let's understand the mechanics of,of what the future of driving and
motoring is, is gonna look like here.
He had.
(04:09):
Three use cases, and I'll lookat my notes in, in a moment.
But really, I think this is thesort of takeaway a around it.
I mean, he had more examples, but oneexample and the, and and an example he
kept on coming back to was, uh, formulaOne aerodynamics and how when you use ai.
(04:30):
Instead of exploring one idea, youmight be able to explore five ideas and
therefore, you know, in, in the week thatyou've got or in the hour that you've got.
And he had a lot of metricsaround just how quickly you can
use AI for design iteration.
(04:52):
The other thing that I thought wasreally interesting was he talked about
at the moment, engineers need intuition.
Their research is guidedby intuition with ai.
Our AI tool will help themprobe in the right direction.
So in other words, if you'rea Formula One team, the data
(05:12):
streaming off the car, the AI.
Can help point you in the rightdirection rather than you needing
somebody like Adrian Newey to beable to say, aha, this is my vision.
This is where I think we should be going.
And then the data can be used to craft andhelp you move in that kind of direction.
(05:32):
So I think there's a subtle nuancethere, and I hope I've made that clear.
The third.
Element was ingest workflows.
This is, in other words, you know, youhave a bunch of different engineering
processes that lead to the finalprocess, and what you can do is you can
(05:52):
chuck it all into the AI and say, dude,how can you make this more efficient?
And then it will, one of the examplesthat he had, it may not have been
this specifically, but it illustrates.
Kind of complicated problems that he'sworking on in the automotive space.
All these EVs, theirbatteries need cooling.
There's these cooling pads and the coolingpads basically feature water moving
(06:17):
around like some kind of coolant movingaround under the cooling pad, right?
Well, the way in which youcan make the water move.
You know how it swigs and squawks,how it zigs and zags, what you do
to make it move in a particular way.
This has a very meaningful impact onhow effective the thing is at cooling.
If you think about the way that atraditional car radiator works and
(06:41):
how, if you know an old one that'sgot a few stone chips in it and.
Bashed and you know, it, itnoticeably doesn't work as well.
Right?
That might be why the caroverheats on that hot summer day
climbing a along hill, right?
So it's those kind of, uh, of ideas.
So once upon a time, what an engineerwould've done is be like, well, I know.
(07:03):
From my experience designingother cold pads that if I make
it do a figure of eight motion,say that worked well in the past.
So that will be thebeginning of my research.
So in the first instance, right, whatAI is gonna be able to do is help
me iterate on that five 10 x fasterby doing lots of different models.
(07:23):
But what he's saying also isthat I needed to be know that
figure of eight thing what?
We can do now is have the AI suggest,you know what, if you do zigzags
all the way up and down zigzag,crazy zigzag, that might be better.
Or, you know, how aboutPacman a Pacman system?
(07:44):
You know, where he goes around ina, in a Pacman kind of pattern.
That might be better.
Well now it's not it,it's not like this is.
Better than the figure of eight.
It's the beforehand you only had thefigure of eight method, now you've
got the figure of eight and thePacman and the zigzag and whatever
other stuff the AI's gonna pullout of its computer, silicon ass.
(08:06):
Um, so that's kind offascinating, isn't it?
So that was the, the important bit.
So let me give you some,some bullets around that.
So how did he call it?
He called the first one.
We can capture the physics.
For the Formula One team and somebodyin the audience was saying, you work
with, you know, Alfred Cent or whatever,Toro Rosso's called Racing Bulls,
(08:26):
whatever Toro Ross is called nowadays.
Anyway, he was saying they'vedone a lot better recently.
I don't know if they have anunfollow Formula one, but did
you have something to do with it?
And the guy was like quite smug,but obviously felt as if he did.
So that was really cool to experience.
So that's case one.
Capture the physics, right?
Case two, is this a, a designbecomes more efficient, right?
(08:49):
And the example here was Marla,the piston people who were now
developing electric motors.
Of course, the motor that they developedusing AI was 15% more efficient and
three decibels quieter thanks to to ai.
You also talked a lot about how theengineers then went back to see what
(09:10):
the AI had done that made it 15%more efficient and three, so in other
words, there, oh, an engineeringintuition was being informed by the ai.
This symbiosis is really weird, isn't it?
We've seen it before, right?
Initially, the car wasjust a way to get around.
Then suddenly it gave us the suburbs andsuddenly it gave us mass consumer credit.
(09:33):
Didn't it?
Suddenly all that stuffhappened in the twenties.
Right?
But it's just fascinating stuff, isn't it?
The example for the ingestion ofworkflows, and I thought this was amusing
given her the lead, our presenter,um, was Vallejo and the idea of, of
their product is to unite these sortof three efficiencies, uh, together.
(09:55):
So that was the sort of cliffnotes, I guess a lot of people.
But I mean, yeah, you can turn off now.
I mean, I does say a lot of peoplewould've turned off 'cause of my
useless prattling before then,but that's my, uh, my attempt to
do, uh, to, to do a sum up there.
Let me move through my notes now.
Flashing out what we just talkedabout, let me also mention that.
(10:16):
Span whilst we were there, had a coupleof giveaways sa, international Edge
Artificial Intelligence in Connected,cooperative, and Automated Mobility.
I haven't even had a look atthat yet, but it's like this
is like a pamphlet kind of guy.
I guess it's more of that softwaredefined car stuff that I've talked
(10:38):
about in other presentations.
This is definitely this materialedges right towards my know
the enemy kind of thing.
Understand how they're deve inventing.
Cars and driving.
Now that's interesting, right?
I that I funny that I thought ofthat because they laid on nice food.
Thanks for that, Stan.
And, uh, after I scarfed the nice food,I was, uh, hovering about to, uh, to
(11:04):
sit down and, uh, somebody struck upwith me and I thought I recognized
him and he was like, I, I know you.
And it was one of the people thatI presented to when I went and
presented five great BMWs to BMWs.
Group of designers in, um, Silicon Valleythere, he was one of the people that had
had attended, and in fact, I was thinkingof him just recently because he had been
(11:28):
involved in the development of the BMWS1000 RR Uber Sports Bike, and we talked
about how his insight for me was how.
MW Faithful were all aboutthe tele leave of suspension.
And of course the S 1000RR was away from that.
(11:48):
It was, we're just gonna do thingsthe way, basically the Japanese do
it with the upside down conventionalforks and you know, BMW will.
Engineers felt that their telelever solution was better.
So there was resistancearound that internally.
So, uh, the S 1000 RR had to be aknockout because just internally
(12:10):
it would've crippled the careersthe people involved had it.
So he's passionate about the S 1000.
RR was really pleased that I'd included itin my like list of like five great BMWs.
He showed me he'd recently been to Munich.
There was a BMW Design classic studio,and he was scrolling through the photos
(12:30):
that he had until he landed on the.
PK Era Bra and BMW Formula One Car,which had been another one of my
great cars that I just felt showedlike BMW's immense design superiority
twice in Formula One with thatbra and BMW of the early eighties.
This was a. Basically A BMW 2002,you know, 1,602 block with a special
(12:56):
head on it that could take the turbo.
And those crazy statistics, like1500 horsepower statistic that you
always hear being bandied aboutthe Turbo Era Formula One cast.
That was the BMW in line four.
That produced that statistic.
My understanding is as a flashreading, so you know, certainly
(13:16):
centers Lotus Reno of the mid eighties.
They reckoned 1300.
1350 in qualified peoplemay know more now.
I've not been on the forums.
That's a, a different kind of rat hole.
But yeah, so I was chatting with thisBMW fellow, he was showing me these
pictures of these awesome classicBMWs and saying that he'd thought
of me and I was musing on my, uh.
(13:38):
Role as, as influencer and I, aswe were stood there span was like,
you can take all these books away.
Well, people rushed up andtook away all the books.
So the Edge Artificial Intelligencepamphlet was the only thing that I
got from the pamphlets, and the onlybook that I got was this Biblical
looking tome, which is entitledTransport Transitions, advancing,
(14:03):
sustainable and Inclusive Mobility.
Proceedings of the 10th TRA conference,2024 Dublin Island volume two.
Sustainable transport development, andnow this is full on no thine enemy stuff.
This is, if you're familiar with theway that modern cities are creating like
(14:25):
exclusion zones by putting bollards and.
Corralling the car corrallingcars into, uh, lanes that
are wide enough for the cars.
Just, but you know, that extra spacethat they used to be for you to
like cut the corner a little bit.
Now there's room for pedestrians tostand in the road or in the case of San
Francisco, FAL attics to fall in the roadin that space, and they put these annoying
(14:49):
bollards around and, and all of that.
That's, these are the people.
They were creating the theory.
But behind that, another one of thesepods that I did when I was asked to do
a keynote at the Auto User InterfaceConference, uh, in 2024 last year,
I, a number of the sessions that Iwent to focused on, on this kind of
(15:13):
sustainability and, and, and so on.
It really feels to me a lotlike, well, think about the way
that cars of the 1960s were.
And then compare them with the way thecars of the 1970s had those big ugly
bumpers and big pillars, and were heavyand were unresponsive and de-tuned
(15:38):
and, and really the whole freedom ofdesign and freedom of expression and
the sheer joy of motoring it left.
Didn't it at that time?
Well, that I feel like is what's happeningwith these people who are doing this.
It's interesting.
That should sit at thenexus of both of 'em.
(15:59):
'cause it's all traffic technology.
It's interesting that the, uh, AIsystems that have been developed on
Formula One or that, one of the otherapplications that this guy talked about
last night was this SP 80, which is the.
Waterland speed record for unpoweredvehicles, so for sale vehicles.
(16:20):
So this is like a clay foil thing.
I'll put a link in for it.
This was learning about, this is one ofthe most awesome things that I've learned
and, and seen about for, for some time.
So it's interesting.
There's this exciting sort ofmotoring adventure that sits at the
same time as this really dystopiansuffocating, the freedom of motoring.
(16:44):
I might even review this properly.
I might have a proper look through it.
I've not done more than flickthrough it, so there, I've,
I've prejudged it, haven't I?
What kind of an asshole YouTube,but am I to prejudge the, uh,
transport transitions, advancingsustainability and inclusive mobility?
My god,
(17:05):
enough for the preamble and thesetting up and the setting the
stage and, uh, and, and all of that.
Productive adoption of AIand engineering design.
So it's AI workflow processes.
So his, so the speaker's backgroundis that he developed some AI work
processes when he was at that lowsand college that I talked about,
(17:27):
which became the industry standard asneuro concepts about a hundred people.
Now, automotive is their strongest sector.
I've got a photo.
I mean, it's basically like an openbox and all the flaps of the box
are like all of you know the waysthat the product can connect and
the partnerships that they have.
And then the inside of the box is theirAI shit stuff out of the flaps of the box
(17:52):
with stuff like, you know, the Nvidia.
Net core network infrastructure,which I assume is something like,
you know, the Microsoft vendornetwork that I sold through when,
back in the days when I was a, a techsalesperson 25 years ago, I could be
wrong, could be talking outta my ass.
I've included the slide.
If you can be asked, go tothe notes and click on it.
(18:14):
11 OEMs in automotive,18 tier one suppliers.
They're also managing a growing expertcommunity, whatever the fuck that May.
The next was where we, we talked aboutthese three main pillars, and that was
the language that he used AI to capture.
The physics IE, what arrow package shouldI use In Formula one is the defined.
(18:38):
Requirement and you know, itcan really help with that.
And the example that I gaveearlier about the figure of eight
flow or the PackMan flow of wateraround a battery cooling pack.
You know, the same if youthink of it, aero swirl.
It's a similar kind ofsuper complex modeling.
In fact, another use case theyhad was for an architecture firm
(19:02):
who used to build these modelsto create the wind tunnel effect.
So that they could forecast a wind tunneleffect on pedestrians 'cause otherwise
in San Francisco's a bit like this.
And you know, a lot of American citiesare, the streets become, when they
have the high rises, it makes like awind tunnel kind of bacteria effect.
So it's really horrible walking as apedestrian, especially if it's raining.
(19:24):
'cause it blows the bad weatherright up in your face kind of thing.
Well.
Nowadays, rather than having tolike make the models and take two
weeks to like get results, nowadaysthe AI can do it from in 10 minutes
flat, so that's another application.
It seemed to me that was pretty muchthe same as how do you shape the
error foil on a formula of one card.
Not that different from, I'll come ontothe water speed record one in a minute.
(19:49):
'cause that is more complicated.
Well, I'll talk about it now.
It is more complicated, right?
Because it's, it's not justthe shape of the foil, it's
the flexibility of the foil.
It's not just the shapeand physics of the water.
It's also the pull ofthe kite on the foil.
Dragging it along.
And then there's the wind as well.
(20:10):
So that's a really supercomplex piece of modeling.
It's also the single coolestapplication and four oh one's pretty
cool, but this foil, SP 80 waterlandspeed record thing is so cool.
So just to be clear on that, Waterlandspeed record at the moment is 65 knots.
They're trying to do.
80. He was saying lastnight they've got up to 55.
(20:31):
And the little bit offootage of it, my God.
I mean, 'cause you all know 55 knotson water that's gonna feel like a
hundred miles an hour on land and it'srough, you know, anyway, whatever.
So I was pretty taken with, uh, with that.
Wind and foil and allof that kind of stuff.
That to me was coolerthan the formula one.
Anyway, that's pillar one,capturing the physics pillar two,
(20:53):
you know, capturing the physics,what should I do, kind of thing.
Pure use case.
Pillar two is the MLA electric motorexample where it's, you know, where they
built a more efficient motor more quickly.
And the measure of it was 15% moreefficient and three decibels quieter.
And then we talked also about,because the way their software works,
(21:15):
you can drill into what the AI did.
Your engineers know what it did, sothey can learn better next time and
have better ideas next time so theycan inform the AI better next time.
So, and, and you know, that'skind of obvious, right?
If you've used any ai, as you well know,if you put shit in, you get shit out.
If you asked it, who wonthe Battle of Gettysburg?
(21:36):
And why?
If you a good example, if you ask it,who won the Battle of Berrys burg?
It gives you some total bullshit 'causeit doesn't know what you're talking about.
It takes a guess.
It might be right, it might be wrong.
You have to be informed to understandthat I was to illustrate this and
this is where AI is at the moment.
To illustrate this, I was researchingearlier today a particularly
(22:00):
gruesome shooting incident inLondon's history where three
policemen were shot dead by gang.
Members, basically whatwe call now gang members.
1966, it was Harry Roberts was the mainvillain, but one of the other villains
was, uh, was this guy John Duddy.
(22:22):
Well, I was googling up John Duddy, butAI kept on wanting to give me answers
about John Duddy, who's some Irish boxer.
So that illustrates how, if I hadn'thave been on the case to say, no,
not John Duddy, the Irish boxer.
I mean, John Duddy and Harry Roberts,who were involved in this rather,
(22:43):
uh, unpleasant shooting incidentin, uh, in London in the 1960s.
The bad guys.
I don't.
They they, it was a standardvanguard involved in the shooting.
If you picture up Googleup a standard vanguard.
I mean, the cars as ugly as the, uh, asthe story and the people involved in it.
And it gives you a real feelingof this sort of guy Richie
(23:08):
Craze kind of underworld London.
But it was still shockingbecause it was a shooting.
I guess at that time, villains used likeknives or you know, one of the three,
the three that went to jail that didn'tactually shoot any of the cops, but
went to jail anyway, when he came outof jail, he was released in the 1980s
and uh, he was like living in a flat inBristol and his flatmate, who was a heroin
(23:32):
addict, beat him to death with a hammer.
I was like, God damnit,there's some retribution there.
But there isn't, right?
Because the main villain, it wasa really nasty piece of work.
This Harry Roberts, he's still alive.
They released him in 2014as far as I can make out.
The bastard is still alive.
All of the accounts talk abouthow he gloried in killing.
(23:55):
Apparently he learned to do it when he wasin the British Army on national service.
There were some likerebellion and he records.
He killed four people bragsin the Wikipedia entry about
killing four people there.
That's when he really.
Decided he enjoyed killing people.
My God, how did people become so twisted?
(24:15):
But anyway, back to where we were going.
And the third pillar, so oneis capture the physics two is
more efficient, more effective.
The Marlo electric motor.
Three is the Valeo exampleof ingesting workflows.
So the overall process of developingwhatever widget it is from cradle
(24:38):
to grave becomes faster and andAI informed, and then that offers
the same kind of circular flow.
Of learning that we talked about,in example, number two, Mar electric
motor, and this is where I'vereally gotta take my hat off to
spend for the formatting spend.
Then asked a question and this ledto a backwards and forwards with the
(25:00):
audience, which was really cool becausewhat I'm trying to understand and
what I'm trying to share with you, thelistener, is where are we with this shit?
Like, is this bloke talking aboutvaporware or is this something which
is actually gonna be on the next BMW?
Like, 'cause there's three guards fromBMW sitting right over there and the
(25:22):
campus were on like opposite acrossthe road from the building we were
in was Ford's base in Silicon Valley.
So there were probablyfour people there as well.
I mean, I don't know, nobodywas wearing a Ford badge.
I mean, everyone's just wearing like.
Engineer Silicon Valleyneutrals, aren't they?
You know, the Rohan pulloverand the uh, jeans kind of thing.
(25:44):
But anyway, yeah, fanhad a question about.
AI versus traditional simulationsand the IE. Why would I use AI
instead of a traditional simulation?
And the answer is thatAI can be predictive.
So it turns days of work intothe work of minutes or seconds.
(26:05):
It's an evolution of currentprocesses and makes them faster.
So it's not like you're chuckingout everything you did before.
Fuck on that bollocks.
We're using the AI now.
No, it's not that.
Rather it is looking atyour current products.
Allowing the AI to ingest theworkflow and then looking at what
(26:27):
efficiencies it might suggest, andI find myself thinking as I'm saying
this, the evolution of sports bikes.
And there's a jigsaw video, whichI'll add a link to, which shows
the evolution of the Suki S xr.
When you were living here, it didn'tseem like it had changed very much.
You know, 'cause each year they onlylooked a little bit different and
(26:48):
yeah, they were evolving generations.
But even when the generations evolved,you know, when they went to like water
cooling and fuel injection, you know,that seemed like an enormous change.
And I guess it was, but it'sstill an evolution, right?
It was still basically a G six R if yousat on the previous generation and you
get on, you know, the all new water coolfuel injected, when it still felt like
(27:12):
a big, heavy, scary G six R. A few yearslater, they'd evolved to kind of feel
like 600 had a few years ago, right?
So, so what I'm saying is that frommy like nineties, eighties, nineties,
geo six R through the early watercold, you know, it's away from the oil
cold ones to the early fuel injected.
(27:34):
Water.
Cool ones through to the K five, youknow, and the last one that I have,
which is arguably, you know, whicha lot of people feel, is peak analog
sports bike design, which does feellike a 600 easy to ride, like a 600.
Not as intimidating as the earlier bikes,although it's just as powerful, right?
That is a long explanation about howa particular design thing evolved.
(28:01):
Right.
What we're saying is that thatevolutionary process can be
incredibly powerful, right?
It seems like it's just smallchanges incrementally, but when they
compound, even if it's just from a1993 jigsaw to a 2001 jigsaw, from
a 2001 jigsaw to a 2005 jigsaw.
(28:24):
The changes are enormoustaken all together.
They're enormous.
The thing in the 2005 jigsawis nothing like the 1993.
I mean, it's still a sports bike.
It still has like the GSXR genes, but ohmy word, the thing has evolved an an awful
lot, so this is my way of underlining.
I guess the other example of continuingevolution is the Porsche nine 11.
(28:48):
And it's had itsrevolutionary changes, right?
But even then it's been like,oh, it's water cold now.
Like, oh, it's the 9 9 1.
Now I can barely look at it.
It's so big and fat.
Yet now you're looking at it andbeing like, actually it's all right.
And when you look at like an LG 50 car,you're like, wow, it's the same size.
There's a Volkswagen Beetle.
It's so titchy.
Right?
You know, you, you get used to thechange and the thing evolves whilst
(29:12):
the DNA re remains the, the same.
That is the evolutionary process canhappen much, much more quickly with ai.
Let's be really real about that, and thatI think is something that I hadn't fully
wrapped my head around until last night.
But to answer Spence's question, AIis not a replacement for simulation.
(29:35):
Instead, it's a way to make the simulationmore effective and happen faster.
Oh, this is interesting.
So like all AI clarity aroundunderstanding your goals in using
the AI delivers much better results.
So just like we know, the prompt is key.
If you ask who won the la, the Battle ofLake Berry Sea, the AI's not got a clue.
(29:59):
If you ask it who won Gettysburg, itcan give you a really good answer.
Now, the example that the people sittingnext to me were asking about was.
If I like, forget to tell it, tell theAI that the car's got a crumple zone.
Can it figure out after the crashhas happened by measuring the
(30:22):
data, can it figure out that ithad a crumple zone, or do I need to
tell it about the whole concept ofcrumple zones in the first place?
And, and the answer to that was.
AI cannot invent a phenomenonit hasn't been trained on.
So in other words, if you don't tellit about the concept of crumple zones,
(30:43):
it can't pull that shit out of its assand just make up the concept of that.
You have to say, well come back towhat I said before is you can't ask it.
Who won the Battle of Lake Berry Sea?
You have to know like Berry Seed's,a good place to do power boating
and Gettysburg was where there was abig important battle that determined
(31:05):
the outcome of the Americans of War.
The purity of F1 means that it is theperfect example and illustration of the
application of this ar, and it really was.
He kept, I would say he cameback to it three or four times
just because it's such a pure.
Illustration of ways.
Think about that.
There's a race in two weeks time.
(31:26):
You can't wait three weeksfor the data to come back.
You need to be modeling somethingnow, and it literally, if you've got
three ideas for what a wing mightlook like once upon a time, you only
add time to work on one wing design.
You have to like engineering intuition.
Take a guess what youare gonna work on now.
You can work on all fiveideas that you've got.
(31:46):
It's really, the impact is exponential.
That's what you end up feeling about.
It's really, really fascinating.
So on that theme of aiding engineers,intuition, the AI is capable of
looking at the data and then makingsuggestions about where to go.
(32:11):
And what it can't do is come up witha phenomenon like a crumple zone.
That he didn't know.
But what it can say to you is, oh,if you wanna avoid crashes, maybe
work on Thai technology ratherthan working on crumple zones.
That's not a very good example.
But you understand what I'm saying isthat the notes that I made last night was,
(32:31):
um, evolving from intuition led designevolution to data led design evolution.
AI can show the best direction to go.
It could prevent dead ends.
And I was laughing because one ofthe other cars that the chap from
BMW had been looking at at BMWClassic was something that he was
(32:52):
like, I've ever seen this before.
And I was like, good lord.
I remember it.
It was the Twin KeelWilliams BMW Formula one car.
I might be wrong on this.
Google it, find out more.
But I remember they fundamentally an ideato design a chassis in a different way.
It was a huge gamble, and if ithad worked, it would've given them
(33:13):
massive competitive advantage, but itdidn't, and it just put them behind.
You might even be able to make a case.
It was.
One of the things, if not the thing thatwas like Williams Spiral downward from
a front of grid team to a midfield team.
But I dunno, my contemporaryFormula one history enough to be
able to, uh, to comment on that.
(33:35):
But at least the Twin Keelcar is a case in point.
AI can do a bunch of work.
AI could have potentially told theWilliams team, guys, this is a dead end.
You don't wanna do this.
This is interesting.
Right.
He wanted to emphasize what his productdoes is enables your engineers to
(33:55):
utilize their domain experience better.
So he's saying, look, you guys know betterthan I do what prompts to put in to my ai.
I'm just saying my AI is gonna help youguys get better if you use it properly.
So I, but I thought that was interesting.
Uh, enable your engineers to.
(34:17):
Utilize their domain expertise better.
There were a number of mentionsof sort of scope of performance
games and, and 10 X seemed to bethe number that kept cropping up.
Now, I dunno how you cankind of measure that.
Maybe it's like, you know, MalcolmGladwell's 10,000 hours before you
(34:38):
achieve like symbiosis with the machineand what you're doing or whatever is.
10,000 hour theory is, Inever read the book sadly.
The point is that that's, they'renot suggesting that it's gonna
like, make you 50% more efficient.
They're suggesting really, uh,orders of magnitude and, you know,
(34:58):
I think it's a recognition of thatpotential that might explain why.
The tech companies are falling overthemselves to have the lead there
because they see the productivitygains that are are gonna be had.
I mean, the rest of us just seemass unemployment, don't we?
But there you go.
At least when people are unemployed, theyalways do productive things like phishing.
(35:18):
They don't do things like getinvolved in stupid revolutionary
politics, but we won't go there.
Uh, we talked a lot aboutthis concept of the black box.
What he said was, customersthink they want a black box.
You know, they just push it andtheir shit's more efficient.
That's what they think they want.
But actually when we work with them,they realize that what they want
(35:41):
is why I described when we weretalking about the M Electric motor
example right at the beginning, wasthat what they're looking for is,
yes, great, it's 15% more efficient.
But I want to dig in and see whatprocesses it did to get to be.
15% more efficient.
I wanna understand how it gotthere because I'm gonna be able
(36:02):
to apply that to other projects.
And if I, as an engineer and a teamof engineers can understand that,
that's gonna improve my engineeringintuition next time around.
So my prompts for the AI aregonna be that much better.
So it's gonna be this fly.
Effect, isn't it that, I mean, when youthink about what that means sort of down
(36:24):
the line for the speed and efficiencyof any and every engineering project,
it's kind of mind boggling, isn't it?
And there were a bunch of questionsthat were kicking around the topic
of, do you tell the model of physics?
Or do you let the AI look at thedata and work out the physics itself?
(36:45):
And I think if you come back to the rumplezone example, well, I think the answer to
that would be that you have to tell it.
If you built a rumple zone on thecar, you have to tell it about that.
If you are trying to set thewater speed record, you have to.
Tell it that.
Yeah.
The craft itself can bend a little bit.
It's not just aboutthe wind and the water.
(37:07):
The craft itself might bend a little bit.
You have to tell it allthe parameters, don't you?
In fact, right?
If you think about my example of JohnDuddy, when I first surfed on John
Duddy, I was using Google's Gemini.
What it gave me was the Irishboxer I, I then put it with
Harry Roberts in 1966, and then.
(37:30):
It was able to delivera meaningful result.
Uh, it was a question alsoabout how to prepare and clean
the data before feeding the ai.
And of course, this was a general,there was the general feeling in the
audience that, that was sort of thebread and butter of any AI project was
how you feed and, and clean the data.
(37:52):
Properly.
Then, uh, he wrapped up with two examples.
The first one we've talked a little bitabout already was the, we'll talk about
both of 'em a little bit already, but thefirst one, these Canadian architects that
do highrises and they use AI modelings toprove that the new building won't create
horrible wind tunnels for pedestrians.
So I talked about that as well.
(38:13):
The additional wrinkle that I forgot tomention earlier was that originally they
would just model for the wind tunnel.
But now they're able to add sun andshadow modeling and then they were able
to blend that and some other measurethey had that I didn't remember to
create a total pedestrian comfort.
Kind of measure.
(38:33):
It's quite interesting, isn't it?
You see the way the AI is able to notjust improve the wind tunnel modeling, but
also have all these additional dimensionsthat overall develop and deliver a much
more holistic, well for out solution,but none of it's possible without an
(38:54):
engineer giving it the right prompt.
Pretty soon it's gonna be, isn't it?
Let's, let's be clear.
I mean, we're all engineering our wayout of a job here, apart from these SP
80 CH water speed record fellows, right?
That's gonna be the future, isn't it?
Everyone's gonna be busy doingwater speed records and fixing
vintage motorcycles, aren't they?
(39:14):
They're not all gonna beinvolved in pointless politics.
I just wanna say as well, just withthis thought, right, the way the Romans
dealt with this was the corn doll.
And I thought of that whenAndrew Yang came along and
proposed his like minimum wage.
On the face of it, you are like,Andrew Yang passed me the bomb, man.
What kind of hippie bullshit is that?
And then when you stop and think aboutit, it's like it is the way the Romans
(39:37):
controlled their unruly populous.
So it's the Unpowered water speed recordthat these SP 80 CH guys are doing.
As I said, the, the recordnow stands at 65 knots.
Their goal is 80 knots.
It's a hydrofoil kind of design.
If you've not had a little lookat the clip or the thumbnail
(39:58):
image here, the physics of thefield is extremely complicated.
Because of the multiple factors,the kites pull, the resistance of
the water, the shape of the water,wind blowing across the foil and
the surface of the water itself.
The shape, stiffness and friction,coefficients of the foil itself.
(40:21):
So an interesting project.
And then to wrap up.
A really interesting thought that it justleaves your jaw like dragging on the floor
a little bit because I just hadn't thoughtof it in these kind of terms before.
But this is a good way actually to justwrap up my whole little session here.
(40:43):
It's that the productworks mostly off graphics.
Two dimensional or three dimensional.
It doesn't use words and numbers so much.
And in fact, when somebody asked aboutthat, he made the point that there
were many other applications that coulddo words and numbers better than his.
(41:04):
Like it was a bit of a, like, wasn'tit obvious that was was what we do.
Like we can do these two andthree dimensional objects instead
of just words and numbers.
The future, my word, the future isa mind boggling place, isn't it?
Thank you.
Drive through.
(41:32):
This episode has been brought to youby Grand Touring Motorsports as part
of our Motoring Podcast network.
For more episodes like this, tune in eachweek for more exciting and educational
content from organizations like TheExotic Car Marketplace, the Motoring
Historian, break Fix, and many others.
If you'd like to support GrandTouring Motor Sports and the Motoring
Podcast Network, sign up for oneof our many sponsorship tiers at
(41:54):
www.patreon.com/gt Motorsports.
Please note that the content,opinions and materials presented and
expressed in this episode are thoseof its creator, and this episode has
been published with their consent.
If you have any inquiries about thisprogram, please contact the creators
of this episode via email or socialmedia as mentioned in the episode.
Popular Podcasts
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
The Bobby Bones Show
Listen to 'The Bobby Bones Show' by downloading the daily full replay.