To better understand how Artificial Intelligence can impact the watch world in the near future I set out to discuss the topic with horological experts. Experts who further shared an interest in AI and the Tech & Science sector in general. Hereby, I've had very thought-provoking and insightful interactions with Giovanni Prigigallo (BioTech Entrepreneur & Co-Founder of EveryWatch), Erik Gustafson (Former Physics Engineer & Founder of Hairspring Watches), Ben Dunn (Former Product Designer at Facebook & Founder of Watch Brothers London), as well as Faheem Kingflum (ScrewDownCrown Newsletter). I further sat down with ChatGPT itself in a semi-serious interview that too led to several important insights.

January 17, 2024

Artificial Intelligence & Watchmaking - Applications, Dreams, Hopes & Fears

---

    Marcus Siems @siemswatches
    Collector, Author, Data Analyst

---

  

Rationale. To better understand how Artificial Intelligence can impact the watch world in the near future I set out to discuss the topic with horological experts. Experts who further shared an interest in AI and the Tech & Science sector in general. Hereby, I've had very thought-provoking and insightful interactions with Giovanni Prigigallo (BioTech Entrepreneur & Co-Founder of EveryWatch), Erik Gustafson (Former Physics Engineer & Founder of Hairspring Watches), Ben Dunn (Former Product Designer at Facebook & Founder of Watch Brothers London), as well as Faheem Kingflum (ScrewDownCrown Newsletter). I further sat down with ChatGPT itself in a semi-serious interview that too led to several important insights. (You can read the full transcript [here]).

The following will be an aggregation of these discussions. Thus, the writing comes from me but my brain was not the only one involved in the process this time.

 

Artificial Intelligence and Watchmaking:

- Introduction - ChatGPT Interview - Community Outlook -

 

What AI might influence?

Last week I've introduced the general functions of Artificial Intelligence and tried to demystify the concepts that have stunned all of us last year. Importantly, even though ChatGPT and similar open-access cloud services popped up in our collective minds just about 12 months ago its methodology has been around far longer.

 

You can find "A Brief History of Artificial Intelligence" covering the last 70 years of major developments by BootstrapLabs here.

 

While AI has been in the closet for about 40-50 years it just now came out into the open and took the public attention by storm. So as last year has been the ceremonial introduction we're hopefully looking into a more application centered exploration of its possibilities in 2024^[5]. And this is exactly the point I wanted to start examining how AI can impact our horological bubble.

Given the possibilities with AI I wanted to split this analysis up into three main parts: (A) Number crunching for example for watch pricing, (B) watch design, and (C) content creation. Hereby, I want to address our hopes, fears and biases when it comes to AI and hopefully get to a point where we can define manageable goals for future applications.

 

Meet the Experts: Associating faces with the grand insights I can gladly share here. (Left to Right) Erik Gustafson, Giovanni Prigigallo, Ben Dunn & (the shy) Faheem Kingflum.

 

(A) AI for Number Crunching

In a news-cycle that gets dominated by various chatbots we shall not forget that Artificial Intelligence is first and foremost a formidable set of methods to analyze and comprehend huge amounts of data. Taking this to the watch-verse there's probably one analytic all of us are looking for - experts included... The holy grail for watch enthusiasts is a tool to find the best price:

"Individual pricing of watches becomes extremely difficult when we're looking at rare variations. The market becomes stochastic." - Erik Gustafson (Hairspring)

Pricing - particularly on the vintage market - is both art and science. And it is an issue on all ends, the consumer, the manufacturer and the secondary market dealers... We all want to optimize and "feel good" about the monetary value we just assigned to these intricate pieces of ticking metal. If we think the price is right we might value the watch itself higher.

 

Erik's mission at Hairspring Watches is finding and pointing towards the coolest  and sometimes odd watches up for sale out there. He is also probably best known for his picturesque descriptions and the extensive background information on all these different listings. And he doesn't use AI for those yet (I asked). Photo Courtesy of Hairspring Watches and Menta Watches.

 

Understanding and educating about prices is probably the go-to application in watches and already in the mission statement of AI-savvy websites like EveryWatch. The interesting tidbit will be when such an app has been launched and AI becomes our lense onto the pricing landscape.

Analogously we can look at the stock market. A lot of people "playing" the day-trading game trying to beat the market and making some money on the side. However, most of the actual trades are made by bots or at least informed by Artificial Intelligence^[6-7]. In the U.S., Europe and the major Asian capital markets algorithmic trading accounts for over 2/3 of the entire trading volume. A trend that has been stable for quite some years by now.

 

Growth rate for algorithmic trading on respective stock markets by region. From light to dark blue areas indicate tendencies for faster growth. Thus, the Asian market shows the strongest relative development (Forecasts 2024 - 2029). Data and Figure Courtesy of Mordor Intelligence.

 

In other words, if we would adopt this (dystopic?) strategy to the secondary watch market we might soon find that not only pricing but also trading might become largely automated. Dealers and financially well situated individuals and companies might change the market to become faster paced*... Or the opposite happens as prices will be better defined and "steals" become obsolete. That's actually an interesting line of thought we will be able to follow in the future.

 

(B) AI-Supported Watch Designs

On to the next big application: Design. The important bit is that the design of a watch is not just its appearance but also functionality is implemented. Can the case structure hold the movement? How does the dial need to be integrated to stay in place? How can you attach the lugs to integrate the strap or a bracelet as stylish and comfortable as possible. All these dimensions are part of designing a watch.

 

Design is more than a quick sketch on a napkin... Where each bolt, screw and piece of material sits is an intricate process to perfection. Photo of the Audemars Piguet Royal Oak sketch from Gerald Genta vs. its first execution (ref. 5402), Courtesy of Audemars Piguet Chronicles.

 

In other words design is all of the way from a rough idea to a final product. And the ways to get there are manifold. Ben Dunn, founder of Watch Brothers London, has been working in Product Design before he started his own watch business and experienced several generations of software applications come and go:

"It's tough to accept change... but typically you adapt and become better in the process" - Benn Dunn (Watch Brothers London)

So we can be hopeful that AI can be of use to come up with design solutions that might otherwise go beyond our imagination. Particularly the development of new movements is a costly and time-consuming endeavor as it incorporates intricate relationships between physics and aesthetics.

"AI can make this cheaper, and my dream and hope is for leaps in innovation from a mechanical perspective – where we see increasingly thinner movements and complicated watches being fitted into smaller cases, by virtue of AI designing the impossible" - Kingflum (ScrewDownCrown)

But Kingflum also cautions these ideas:

"My fear is that the watch industry is too traditional, and shuns AI as being too modern." [because] "For example, Patek would not suddenly start using F.P. Journe style watch hands, but an AI model might end up producing such a design anyway." - Kingflum (ScrewDownCrown)

We got a potential contradiction at our hands - on the one hand an industry that is proud and embracing its heritage and tradition to even a potentially anachronistic level and on the other hand a technological sector that taps into Science Fiction.

I personally believe that there is one example of a great symbiosis between human and artificial watch design already in existence - namely the Casio G-Shock GD-001. It's the 40th anniversary G-Shock that was just auctioned at Phillips New York.

 

The Casio G-Shock 40th Anniversary watch (ref. G-D001). Photo Courtesy of Casio & Phillips New York.

 

Aesthetically it's not my cup of tea and we can all argue how easy it is on the eye - it's an 18k full golden, 45mm behemoth of a watch and pretty much looks like several other fever-dreamish AI-generated watch designs. But the beauty of it lies in the development process, which has been unique so far and serves as a proof of principle how AI can influence and enhance design.

 

You can find the "G-SHOCK 40th Anniversary Model - G-D001 Development Story" on YouTube and Phillips Watches.

 

What Casio did was first training a deep-neural-network (see Methods) on all Casio G-Shock models of the last 40 years to generate a "classic G-Shock". That by itself wouldn't be much of an impressive feature. It's what modern AI does and can do best. It is on a meta-level the same as taking an aesthetic mean over the G-Shock lineup.

However, what makes this project so inspirational is that the AI output was further instructed to optimize the watch by physical constraints! The generated 3D-concept had the looks of a classic G-Shock and was also optimized for weight-reduction, structural strength, material characteristics and processing methods - it is thus physically as shock-absorbing as possible and aesthetically as G-Shock as possible!

 

The side view of the Casio G-Shock G-D001 with its skeletonesque case design. Photo Courtesy of Casio & Phillips New York.

 

With this co-creation it is of course hard to say where AI ends and human intervention starts but I'd argue that the final design and product is fundamentally new, different, and by definition aligned with its purpose - namely shock-absorbing and accurate time-telling.

Now the interpretation starts with why the design looks like it does and not differently. To me for example the case structures do look almost organic: a skeleton. And on the semantic level this makes a lot of sense. Evolution equipped us with similarly open structures - bones - to save our intestines from shock and harm. In the same way the case does with the movement. 

However, this guessing game unfortunately is always part of these modern AI applications. The network and its connections can't be understood with human semantics. We simply don't know which aspects of the data were weighted such that they gave rise to a certain design...

"You would have to break down the model further and further to understand how it is doing its job" - Giovanni Prigigallo (EveryWatch)

  

Another example from the automotive world: The Czinger 21C is an AI co-created fully 3D-printable supercar. Photo Courtesy of Czinger & Forbes.

 

Independent of its underlying reasoning, the human-AI co-creation concept can pretty much grant constrained-unconstrained design results. Constrained because you as a designer define what data your AI is trained on and what other for example physical constraints you wanna add. Unconstrained because it neglects influences that subconsciously go into the human design process - an AI is itself free from social norms and the history of design we as humans get exposed to on a daily basis.

It might take some time until more established/historical/traditional brands take up the human-AI co-creation approach but look out for the Independents...

 

(C) AI-Powered Content Creation

With the growing popularity and potency of Chatbots and generative Artificial Intelligence models content creation appears to be one the most discussed areas to be affected by AI. It appears that AI content is met with a mixture of three feelings: Fear, devaluation and mockery. All very normal human emotions.

 

One interesting - and slightly absurd - deep dive into the power of generative AI and automatic information processing has been performed recently by @awatchcritic as guest on Kingflum's ScrewDownCrown Newsletter. Photo Courtesy of awatchcritic & Kingflum.

 

What we've seen so far are exactly these examples as above or this Hodinkee story that all appear to follow the simple narrative: "look how AI imagines watches". These examples are funny and may, if put into the right context, be able to give us some insights into how we perceive watches on an abstract level - because AI is learned on websites and human-made content. On the surface these applications are futile but most likely only the beginning.

It's easy to mock about this content mainly because we've seen little advanced AI that was applied - openly - in the watch landscape. This gets us to our second emotion - Devaluation. When we learn that a text has been written by a robot we tend to value it less then a human-written text. That's somewhat of an inherent bias that many of us - including me and some of my interview partners - have.

 

"Watches Growing Out Of Plants". Another (early) example applying generative AI for watch content creation. Photo Courtesy of Hodinkee.

 

We associate that whatever is "robot" or machine-made is also cold, emotionless. For example in a direct interaction with a chatbot we're still pretty good in dissociating it from human conversation^[8]. Further, after a model is trained it can generate an output in the matter of seconds while humans have to ponder about their writing for several hours. These are potentially some of the factors that go into the assessment of value.

"[It] is much nicer to hear a story about a man who toiled for decades to conceive something, rather than an AI model which produced an idea overnight." - Kingflum (ScrewDownCrown)

But if you think passing the Turing test is scary I can get a lot more dystopic and poke into every content creator's deepest fears (and bank accounts)... Imagine a world where most of the content, written and visual, is AI-generated! That's still way ahead in the future, right? Well, already in February 2019 the NY Times reported that about every Third(!) article published on Bloomberg News was AI generated^[9-10]!

 

"The Rise of the Robot Reporter" NY Times article from February 5th 2019 - 5 years ago(!) - on how one third of Bloomberg News articles were written by AI. Photo Courtesy of NY Times.

 

That's news already from 5 years ago! We may not perceive it yet but it is happening! So are we all doomed? Are the machines taking over, am I and so many others soon out of our jobs, made redundant by AI? I guess (and hope) not and here's why.

"Content creation won't go out of business anytime soon" - Giovanni Prigigallo (EveryWatch)

 

(C*) Detour: AI and creativity

I tapped into the "Intelligence" wording and its aspects and fears already last week and this goes into a similar direction. I ask the question: Can AI be truly creative? Can it create something completely new?

As an example I've appended a clip from a relatively recent documentary on AlphaGo, the deep-neural-network that was trained to defeat Go grandmaster Lee Sedol in 2016. It did. And on its way in one of the games it made an astonishing move. Move 37 was not anticipated by Lee Sedol, not even anticipated to be a probable move by AlphaGo, thought to be a mistake by most at first... But it proved to open up the board late in the game and secured the win for AlphaGo.

 

You can find "Lee Sedol vs AlphaGo Move 37 reactions and analysis" here.

 

So what happened? AlphaGo created a strategy that was not played by humans, not predicted by humans, not seen by humans. That must be creativity, right? Yet, I would say no and Giovanni Prigigallo adds:

"I still think that creativity in this case comes from calculated risk." - Giovanni Prigigallo (EveryWatch)

Let's play the devil's advocate for a second. The Japanese game of Go is on a cultural and historical level compatible with the game of Chess. Both have close-to immeasurable degrees of freedom and thus strategies and heuristics have evolved over Centuries. If you'd want to master one of these games and play on a competitive level you would read books, get a coach, play against other skilled players, etc.

 

The two ancient games Chess (left) and Go (right) are played on different boards, have different rules and distinct strategies... But their impact and reputation on the historical and cultural level are compatible.

 

In other words, you would go down beaten paths, follow the game as it has traditionally been played. AlphaGo on the other hand learns it from scratch. It plays itself, and not only (skilled) human players. By playing itself thousands of times, optimizing its own moves, it walks through Centuries of strategy developments on its own, developing its own beaten paths. And thus can beat humans by playing "unpredictable" moves, which actually might be very classic moves in the AlphaGo play-book. That's not creativity but being unbiased from tradition.

 

(C) AI-Powered Content Creation - The Sunny Outlook

I do believe that we can spin all of this into a positive for the human side. First, we don't need to be afraid because we won't be out of human content creation anytime soon. AI is as of right now not creative in the classic sense and lacks human level emotions. You won't take my word for it? Here's ChatGPT on the matter:

"[W]hile AI can simulate certain aspects of emotional expression and recognition, it does not possess the subjective, experiential, and contextual depth that defines human emotions. The emphasis on emotions being a significant distinction between AI and humans arises from the fundamental differences in how emotions are generated, experienced, and understood by these entities." - ChatGPT3.5

The sensual experience is an important factor in this discussion. Good content of any sorts lives not only through vision but graps us on several levels.

"A really good description of a watch incorporates the feeling of it, the haptics, how it weighs in your hand and on your wrist. This adds value to [content]" - Ben Dunn (Watch Brothers London)

And here's the catch: If we don't need this deep layer of connection to the information provided, why not using AI to help? For example when you'd have to write a press releases on new watches that no journalist has touched or even seen in the metal... You are as depleted from the sensual information as AI would be. Why not letting it help you? The devaluation of the timing aspect becomes re-evaluated as a positive asset - the human-AI co-creation approach again.

 

"Biology vs Robotics" by Randall Munroe (xkcd).

 

AI Outlook - How it can help?

I honestly believe that the AI boom will benefit most if not all of our lives. Artificial Intelligence will not render Human Intelligence obsolete. Even the contrary is true and many synergies evolve. Let's not forget that in the end AI is a human-made tool. If for example you don't need a deep (emotional) understanding of a matter AI will save you hours of your work and will make you ultimately more productive. You can have more time for the truly creative projects.

Additionally, AI can process a lot more information at once, incorporate them and at the same time search-engine optimize the text. The holy trinity of data-driven content if you want. Similarly, you can generate completely new designs with physically or biologically plausible constraints in a fraction of the time. But there's a learning curve, too.

"You need to know how the system works to get good output" - Giovanni Prigigallo (EveryWatch)

You need a good AI application but you also need to know how to query it. Take ChatGPT and other chatbots for example. If you'd know how they're wired, you could adapt your inquiries accordingly, prime the system on how you want it to behave^[11]. Thus, the AI automation might cost a few jobs but it also generates others in the form of AI experts - be it engineers constructing networks or AI operators trained for the specifics of their model(-type).

 

Proportion of people working in manufacturing jobs between 1945-2020 in the U.S. - This effectively shows how automation has historically impacted the job market. Yet, unemployment is not increasing in the same way as jobs are shifted from one area to the next. Data Courtesy of the U.S. Bureau of Labor Statistics.

 

I'd like to recite Ben Dunn (Watch Brothers London) here:

"It's tough to accept change... but typically you adapt and become better in the process" - Benn Dunn (Watch Brothers London)

But there are two key feasibility issues... It's great to have cloud-based publicly available applications like ChatGPT but 1) not everybody would hand over their potentially sensitive information to generate an output and 2) an application trained on "the internet" will have broad appeal but fails with personal nuances. If for example Erik (Hairspring) would decide he'd like to have AI help him write for his website it should be able to at least mimic his style^[12].

What I thus see for the near future will be more tools for DIY-AI**. Local toolboxes that make custom applications for smaller businesses and us private citizens possible, easy to use and with (hopefully) relatively high data security standards. What we've seen in 2023 has been a proof-of-principle and now it'll be about customizing the experience^[5,13].

 

(E) AI Applications in Conclusion

In sum, what I hope to have... wait... why am I typing this myself? Hey ChatGPT: please summarize my 6 main conclusions:

"Certainly, here are six main conclusions from the text "Artificial Intelligence & Watchmaking - Applications, Dreams, Hopes & Fears":

1. Introduction and Experts Collaboration: The author sought insights from horological experts, including Giovanni Prigigallo, Erik Gustafson, Ben Dunn, and Faheem Kingflum, as well as a semi-serious interview with ChatGPT itself.

2. AI's Potential Impact on Watch Pricing: AI, with its data analysis capabilities, holds the potential to influence watch pricing. The article explores how AI can assist in individual pricing, particularly in the complex and stochastic vintage watch market.

3. Automated Trading in the Secondary Watch Market: Drawing parallels with algorithmic trading in financial markets, the article speculates that AI might lead to automated trading in the secondary watch market, potentially altering the pace of transactions and market dynamics.

4. AI-Supported Watch Design: The discussion extends to AI's role in watch design, emphasizing that design involves not only aesthetics but also functionality and structural considerations. AI could potentially contribute to cost-effective innovations, particularly in the development of new movements.

5. Potential Clash Between Tradition and Modernity: There's a highlighted concern about the traditional nature of the watch industry possibly conflicting with AI-driven design, leading to a clash between adherence to heritage and embracing modern technological advancements.

6. AI-Powered Content Creation: The article explores the impact of AI on content creation, pointing out the fear, devaluation, and mockery associated with AI-generated content. It discusses the potential scenarios where AI might significantly contribute to content creation, raising questions about job security and value assessment."

 

No one really knows where we're heading with new technology and we honestly don't really know where we are at right now.

"We don't know about camouflaged applications. Applications that work behind the scenes that we are not aware of." - Erik Gustafson (Hairspring)

As long as we don't know, see and experience what is possible with modern Artificial Intelligence on a personal level it will remain a mythical creature. It detriments understanding so let's discard our fears and talk directly to AI... And work together, or rather let it work for us humans. But let's give the silicon star the last words:

"In the end, it's not so much a battle of supremacy as it is a dance of collaboration. AI and humans, when working hand in circuit board, can create a symphony of [applications] that resonates with audiences in ways neither could achieve alone. The future? A dazzling duet of silicon and soul." - ChatGPT3.5

 

 

* Jumping onto another band-wagon: Transparency and traceability on such a market could be boosted through blockchain applications^[5].

** There's another interesting debate about whether training-sources for available AI models should benefit from the profits and in which magnitude. See for example "The Times Sues OpenAI And Microsoft Over A.I. Use of Copyrighted Work" [here]

 

 

References & Recommended Reading

^[1] [Recommendation] On Intelligence; Jeff Hawkins & Sandra Blakeslee, St. Martins Griffin Publishing, New York;

^[2] [Recommended SciFi] Foundation Trilogy (1942-53); Isaac Asimov, Gnome Press;

^[3] The proof and measurement of association between two things (1904); Charles Spearman, Am. J. Psychol.;

^[4] [Recommendation] The Most Human Human - What Artificial Intelligence Teaches Us About Being Alive; Brian Christian (2011), Anchor;

^[5] [Recommendation] What Will Happen In 2024; Fred Wilson, AVC; [Link]

^[6] 1/3 of Bloomberg articles are written by artificial intelligence; Jonathan Andreas, Medianism; [Link]

^[7] Algorithmic Trading Market Share, Size & Trading Statistics; Mordor Intelligence; [Link]

^[8] AI is closer than ever to passing the Turing test for 'intelligence'; Daniel Merino, The Conversation; [Link]

^[9] The Rise of the Robot Reporter; Jaclyn Peiser, New York Times (05. Feb. 2019); [Link]

^[10] 1/3 of Bloomberg articles are written by artificial intelligence; Jonathan Andreas, Medianism (20. Feb. 2019); [Link]

^[11] [For Example] Framing A Purchase Decision; KingFlum, ScrewDownCrown; [Link]

^[12] [For Example] How To Train ChatGPT On Your Own Data: Make A Custom Chatbot; Nicole Levine, WikiHow; [Link]

^[13] [Recommendation] Artificial Intelligence: the next big thing for the watch industry?; Sky Sit & David Sadigh, Europastar; [Link]

 

All rights on text and graphics reserved to the Author.