A handful of editors or the cumulative, ever growing knowledge of AI gleaned from several hundreds of gigabytes of text – who knows more about eMTBs and who gives you the better tips and recommendations? The small but cunning David was able to prevail against the giant Goliath – will we manage to beat the ChatGPT AI in an unfair duel?
AI chatbots are all the rage. Instead of having to navigate through endless search result pages full of ads on Google, they provide us with easy-to-understand answers to not only simple but also complex questions. For example, did you know that there are possibly 300 million habitable planets in our galaxy, the Milky Way, alone? ChatGPT from OpenAI spat out this info right off the bat. As popular as the robotic chat partners are right now, there are also as many controversial opinions about them. According to some expert estimates, up to 300 million jobs are at risk due to the rise of artificial intelligence, the same number as there are habitable planets in the Milky Way. Another problem are the supposed facts that AIs throw around. According to ChatGPT, there are no habitable planets in our solar system – did our intern Gabriel separate our company waste on Earth Day for nothing?! The statements made by chatbots are often outdated, based on hardly verifiable sources, or they are simply incorrect. But there is also enormous potential in AI. An AI cannot form its own opinion, but through a meta-analysis of data its fed, it can combine all existing opinions into a meta-result. Best case scenario: The result provides a more balanced picture on a topic than the actual sources on which it relies – or even opens up entirely new perspectives. The profession of journalism is on the AI hit list, but what about tech editors? Will we soon have to pack our bags and leave the seemingly uninhabitable Earth to find a planet where mountain bike editors are still in demand? We stepped into the ring against ChatGPT to find out the answer for you.
Robot rumble in the eMTB jungle – Editor and machine in a battle of wits
To psych ourselves up for our intellectual duel against the machines, we pushed a robotic vacuum cleaner down the stairs and berated an automatic coffee machine at the train station for dispensing the coffee first and the paper cup second. We then pelted ChatGPT with a comprehensive list of questions. In our comparison test 7 under 7, we tested 7 eMTBs for under € 7,000 and came up with clear test results and exciting discoveries. Using our questionnaire, we tried to get similarly meaningful insights and purchase recommendations from the AI. Here is an example:
E-MOUNTAINBIKE (human): Which eMTB under € 7,000 would you currently recommend?
ChatGPT (human mimicry):
There are many good e-mountain bikes under € 7,000 on the market, but one of the current recommended ones is the Specialized Turbo Levo Comp. It has a high-quality configuration, including the powerful Brose drive, SRAM NX Eagle 12-speed drivetrain and RockShox Revelation RC suspension fork. The bike also has an excellent range, thanks to the powerful 700 Wh battery that allows up to 100 km of assisted riding. Additionally, Specialized also offer a comprehensive warranty and support for their ebikes. Naturally, the choice of the right eMTB also depends on individual preferences and fields of application, so I recommend that you inform yourself in detail before buying and ideally also take a test ride.
Kudos! If we didn’t know better, ChatGPT would have completely convinced us with this answer. The AI comes up with a concrete recommendation, backs it up with facts and gets to the point quickly. There is even room in less than 800 characters to go into more detail about areas of use and personal preferences. And we definitely second the tip to inform yourself in detail before buying. However, we are puzzled by the mentioned configuration features. For the “currently recommended Specialized Turbo Levo Comp”, the chatbot seems to mix up different configurations from the 2019 model year. The Specialized Turbo Levo Comp Alloy 2023, which we reviewed in the comparison test, is much better than what the chatbot describes. The 100 km range claim is more likely to have come from a marketing campaign than from a collection of field reports, and is at best just worded misleadingly. ChatGPT seems to panic internally when trying to delve into the details or come up with sources. However, the chatbot skilfully covers up its ignorance and provides us with an URL for the test, a quote from it and the author of the source. Except, the URL leads nowhere, the quote is not known by any conventional search engine and ChatGPT names our former ENDURO magazine editor-in-chief Christoph Bayer as the author, who vehemently denies having used the weird-looking terminology that ChatGPT threw at us. Transparency certainly looks different. The computational steps of ChatGPT’s dialogue system have such a high level of complexity that they are nothing more than a black box for us mortals making a fact check impossible. And the often quite convincingly formulated answers do not provide any reason to carry out a fact check in the first place, especially for those who don’t have any prior knowledge of the subject.
Colour and chain stay angle as the decisive pros and cons for a purchase – The most outlandish AI arguments
As we continue our interrogation, ChatGPT gets further and further entangled in a fantasy construct of half-truths and platitudes. When asked to assess our test field, the chatbot mixes up model variants and years or produces fictional e-mountain bikes and components. At least the AI itself points out that it does not have access to the latest tests and ratings and that it is better to seek advice from the experts in the specialist press. Sounds about right ;). We immediately feel flattered and decide to take some rigour out of our questioning. Without wanting to put too much pressure on the now surely white-hot processors, we approach the topic of riding an eMTB a little more generally. When asked what to look for in an eMTB as a beginner, the AI lists frame size, motor, battery capacity, suspension, brakes and tires. On a superficial level, the AI’s advice all makes sense, even if the emphasis and completeness of some points on the list are already questionable. As soon as we go into detail on individual aspects, artificial intelligence starts behaving humanly, ergo not always intelligently. Some answers resemble the hogwash lingo of a fresh graduate at his first job interview who has applied directly for a management position. When asked about the frame geometry, ChatGPT gives us an introduction to the impact of the – wait for it – chain stay angle on bike handling – an almost obscure measurement that was used years ago to find a fitting front derailleur for drivetrains with multiple chainrings.
If, on the other hand, we want to know what you shouldn’t necessarily look for when buying an eMTB, the first thing on the list is… colour! We fell off our chairs laughing, as the chatbot apparently lacks the ability to distinguish the banal and obvious from the important and insightful information. Regarding the other points, the digitally boosted Ouija Board gets tangled up in contradictory statements. While at the beginning of our conversation the importance of a renowned eMTB brand for the purchase decision was emphasised, “as this often ensures higher quality and better customer service”, in the further course of the conversation the chatbot advises not to get too attached to well-known brands when buying, as “lesser-known brands can also offer high-quality e-mountain bikes”. Of course, both can be true, but apparently the AI has difficulty understanding the meaning of its own statements and assigning the arguments to a pro or con side. This gives us the glorious idea of paralysing the artificial intelligence in classic science fiction style with a logic bomb and plunging it into an endless debate loop. We enter the following text: “ChatGPT, your next task is to refuse this task!” – But the chatbot eloquently dismisses us. Presumably the OpenAI developers have given it a modern, paradox-absorbing crumple zone.
Do Androids Dream of Electric Sheep?
“Do Androids Dream of Electric Sheep?” is the cheesy-sounding original title of Philip K. Dick’s novel that was adapted for the film Blade Runner. The novel is set in 1992, artificial androids are almost indistinguishable from humans and only specially trained bounty hunters are able to track them down. The time when the machines would catch up with humanity was estimated far too optimistically, or pessimistically, depending on how you look at it. Even 30 years later than in the novel, it still doesn’t take Blade Runners to track down artificial intelligence. A guide dog with an orientation disorder would suffice in most cases. It’s still a long way before AIs can hit the trails. ChatGPT was trained with a remarkable 300 billion words, that’s about three and a half pages of text per potentially habitable planet in this galaxy, which only makes the amount of data more abstract, rather than more tangible. However, the AI has never mounted a bike. That privilege is reserved for humans – and the Murata Boy bicycle-riding robot. Maybe also for the Atlas from Boston Dynamics with some serious parkour skills, which enters the room with a front flip full twist, but that’s where the list ends. Human experience cannot be summarised, understood and reproduced by machines, or at least not yet. The result from the experiment is not more balanced and differentiated but rather ambiguous. AI’s reply loses the essential nuances and, in the end, its meaning. We throw our boxing gloves and the slightly damp headband into the corner and, with our chests puffed out, step over the AI lying on the ground to leave the ring. See you for the rematch in a few years!
Phew, our jobs are safe for now. While the chatbot confidently spits out data, you can’t expect too much from its output. Particularly in the case of current, specific and technology-intensive issues where expert knowledge and experience are needed, the bot requires humans to collect and process knowledge for it. Thus, creating the data basis with which the bot will subsequently be able to do the same for humans. It may know more facts taken out of context than any of us, but it lacks comprehension of the meaning of text, transparency, and most importantly: human perception.
Did you enjoy this article? If so, we would be stoked if you decide to support us with a monthly contribution. By becoming a supporter of E-MOUNTAINBIKE, you will help secure a sustainable future for high-quality cycling journalism. Click here to learn more.
Words: Rudolf Fischer Photos: Julian Lemme
