<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Leon Ingelse on Medium]]></title>
        <description><![CDATA[Stories by Leon Ingelse on Medium]]></description>
        <link>https://medium.com/@leon_ingelse?source=rss-1879184bbde0------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*8vmSP18UP148St5KIcYK1A.png</url>
            <title>Stories by Leon Ingelse on Medium</title>
            <link>https://medium.com/@leon_ingelse?source=rss-1879184bbde0------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 01:34:29 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@leon_ingelse/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com" rel="hub"/>
        <item>
            <title><![CDATA[Why optimisation doesn’t suffer from hallucinations]]></title>
            <link>https://blog.dotsandlines.ai/why-optimisation-doesnt-suffer-from-hallucinations-9f727dd0a13b?source=rss-1879184bbde0------2</link>
            <guid isPermaLink="false">https://medium.com/p/9f727dd0a13b</guid>
            <category><![CDATA[optimization]]></category>
            <category><![CDATA[ai-hallucination]]></category>
            <category><![CDATA[ai-ethics]]></category>
            <category><![CDATA[explainable-ai]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Leon Ingelse]]></dc:creator>
            <pubDate>Thu, 10 Sep 2026 10:01:02 GMT</pubDate>
            <atom:updated>2026-09-10T10:01:02.749Z</atom:updated>
            <content:encoded><![CDATA[<p>Hallucinations — AI-generated responses with misinformation presented as fact — are one of the biggest obstacles to the adoption of <a href="https://en.wikipedia.org/wiki/Generative_AI">Generative AI</a> (GenAI). <strong>If you cannot trust your tools to generate outputs as desired, there are significant risks in their application.</strong> We have seen many examples of hallucinations causing damage, such as Deloitte using AI to <a href="https://fortune.com/2025/10/07/deloitte-ai-australia-government-report-hallucinations-technology-290000">generate</a> parts of a $290,000 report, which included multiple fake references.</p><p>Since <a href="https://dotsandlines.ai">Dots &amp; Lines</a> is a deep-tech company, we are often asked how we deal with hallucinations. The simple answer is: our optimisation models don’t use GenAI, and they don’t suffer from hallucinations. But simple answers are often unsatisfying and lack nuance. This blog post is for the curious.</p><h3>Hallucinations: where do they come from?</h3><p>Data-driven GenAI models like ChatGPT and Claude derive their capabilities from extracting relations from data and reproducing them in a probabilistic manner. In the <em>training phase</em>, they extract relations from huge datasets and quantify those relations within mathematical structures, like a vector space. This vector space is then used in the <em>generation phase</em> to probabilistically generate new elements based on a user-provided prompt.</p><p><strong>GenAI systems are black boxes</strong>, as their inner workings are obscured by the complexity of the model. In other words, the mathematical structures are so large and complex that it’s unclear which relations the models have extracted from the data. This makes it hard to analyse the root cause of the generation of a piece of text or a particular hallucination. However, hallucinations can still be attributed to characteristics of GenAI models. Here, we separate the cause into the two phases: training and generation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*A_Gz3w4y-EW66bG8yFILow.png" /><figcaption>A data representation of the real world is prone to losing details and nuance. <a href="https://betterimagesofai.org/images?artist=BeckettLeClair&amp;title=Binarisation">Binarisation</a> by <a href="https://www.linkedin.com/in/beckett-l-687b7713b/">Beckett LeClair</a>, <em>licensed by </em><a href="https://creativecommons.org/licenses/by/4.0/"><em>CC BY 4.0</em></a><em>.</em></figcaption></figure><p>During training, <strong>hallucinations can be caused by the training data failing to capture the real-world context</strong> it is supposed to represent. The data might have insufficient information or might be contaminated. It is difficult to evaluate which data items are correct, especially in large texts. With an inaccurate training set, output will also be inaccurate — this is known as the <a href="https://en.wikipedia.org/wiki/Garbage_in,_garbage_out">garbage in, garbage out</a> principle. Furthermore, as GenAI models aim to be general-purpose models, their data aims to capture all of human knowledge. This objective is arguably impossible, and the question of what’s included and what’s not, what’s prioritised and what’s omitted, becomes political.</p><p>Consider a data set containing images of doctors. Historically, white men have been overrepresented in responsible positions, and a random sample of images of doctors will thus contain more white men. This can lead to GenAI models generating more of what it has seen in the data: white men. Researchers <a href="https://www.medicaleconomics.com/view/ai-has-trouble-picturing-physicians-that-aren-t-white-men">showed</a> this in practice and noted that this “<a href="https://blog.dotsandlines.ai/optimisation-glossary-f671bbdf9028">bias</a> has the potential to reinforce stereotypes”.</p><p>During generation, <strong>the probabilistic nature of GenAI models can cause hallucinations</strong>. Consider a <a href="https://en.wikipedia.org/wiki/Large_language_model">large-language model</a> (LLM) — a text-generating GenAI model like <a href="https://en.wikipedia.org/wiki/Claude_(AI)">Claude</a>. An LLM generates words probabilistically from the relationships extracted during the training phase. Well-fitting words have a higher probability of being generated. Current models function incredibly well, and words generally fit correctly too, but problems still occur.</p><p>Possibly a more intuitive way to look at these hallucinations is to see LLMs as <a href="https://dev.to/gerus_team/llm-hallucinations-are-compression-artifacts-yes-really-and-this-changes-how-we-build-3d8j">(de)compression algorithms</a>. Just like when you watch a video stream on a low-quality setting, the larger idea of the image is conserved, but finer details do not survive. The decompression algorithm then predicts what should have been there, allowing for pixels that fit well within the image, but are different from the original.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/519/1*t7FCJ1Zx8nin8gQxtLJl6Q.jpeg" /><figcaption>A <a href="https://commons.wikimedia.org/wiki/File:Felis_silvestris_silvestris_small_gradual_decrease_of_quality_-_JPEG_compression.jpg">photo</a> of a<a href="https://en.wikipedia.org/wiki/European_wildcat"> European wildcat</a> with the compression rate and associated losses decreasing from left to right. By<a href="https://en.wikipedia.org/wiki/User:AzaToth"> Aza Toth</a>, <em>licensed by</em><a href="https://creativecommons.org/publicdomain/zero/1.0/deed.en"><em> </em></a><a href="https://creativecommons.org/licenses/by/3.0/deed.en"><em>CC BY 3.0</em></a>.</figcaption></figure><p>For example, <strong>a word can fit well within a sentence, but fit badly in the context</strong>. In May 2024, during the early days of Google’s <a href="https://en.wikipedia.org/wiki/Gemini_(language_model)">Gemini</a>, its AI overviews<a href="https://www.bbc.com/news/articles/cd11gzejgz4o"> suggested</a> using glue on pizzas to prevent cheese from sliding off. Glue fits well within the context of preventing something from sliding off, but it doesn’t fit within the context of eating pizza. Although these obvious hallucinations have generally been mitigated in more recent models (through methods such as<a href="https://arxiv.org/pdf/2204.05862"> reinforcement learning from human feedback</a> and<a href="https://proceedings.neurips.cc/paper_files/paper/2020/hash/6b493230-Abstract.html"> retrieval-augmented generation</a>), less obvious hallucinations<a href="https://www.digitalapplied.com/blog/ai-model-hallucination-rate-benchmarks-2026-study"> still occur</a> [<a href="https://axis-intelligence.com/ai-hallucination-statistics/">1</a>], [<a href="https://hai.stanford.edu/news/ai-trial-legal-models-hallucinate-1-out-6-or-more-benchmarking-queries">2</a>].</p><p>At Dots &amp; Lines, we use optimisation algorithms, not GenAI. In the following section, we explain why our algorithms don’t suffer from hallucinations.</p><h3>Optimisation and hallucinations</h3><p>Optimisation algorithms are not driven by relationships extracted from data, but by the formal representation of a real-world problem in mathematical objects. We call this representation a <em>digital twin</em>, whose development matches the training phase in LLMs. An optimisation algorithm then uses the digital twin to generate an optimised solution to the problem, which matches the generation phase in LLMs.</p><h3>Problem representation</h3><p>In optimisation, anything that is generated will fit within the “mould” of the digital twin, which is predefined by the developers. Just as the training data can misrepresent a problem in LLMs, disconnections between the digital twin and the real-world problem can occur as well. A perfect digital twin does not exist. However, <strong>there are striking differences between data misrepresentation in LLMs and disconnections with the digital twin</strong>:</p><ol><li>With LLMs, the training data generally lacks <strong>transparency</strong> as it often remains undisclosed and is too large to comprehend. In optimisation, the digital twin is developed in unison with key stakeholders, and development decisions can be documented, which is a real boon for transparency.</li><li>LLMs aim to represent all of human knowledge, potentially within a certain context. As this aim is arguably unattainable, the representation will always lack <strong>completeness</strong>. Digital twins are problem-specific and thus cover much smaller contexts. This makes completeness a more accessible aim.</li><li>The data sources of LLMs are distributed, making <strong>accountability</strong> for data misrepresentation a large problem. As we build the digital twin in partnership with stakeholders, accountability is distributed across each actor’s expertise.</li></ol><p>Due to these three differences, <strong>inaccuracies in optimisation can easily be found and addressed</strong> in partnership with the stakeholders, whereas LLM developers cannot ensure the removal of inaccuracies, except by retraining the model. Even then, it’s impossible to guarantee they won’t recur. Optimisation models, on the other hand, allow for the digital twin to be directly modified, which ensures the algorithm adheres to the newly defined rules.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*g2b3sXyc4J7wfZmo.jpg" /><figcaption>An image of a <a href="https://en.wikipedia.org/wiki/File:Photoreal-train.webm">video</a> generated using <a href="https://en.wikipedia.org/wiki/Sora_(text-to-video_model)">Sora</a> of the <a href="https://en.wikipedia.org/wiki/Glenfinnan_Viaduct">Glenfinnan Viaduct</a> in Scotland that includes various hallucinations, of which the most apparent is a non-existent second track. Video in the <a href="https://en.wikipedia.org/wiki/public_domain"><em>public domain</em></a><em>.</em></figcaption></figure><h3>Probabilistic element</h3><p>During the generation phase, hallucinations happen because of the probabilistic element inherent within LLMs. Optimisation algorithms also have a probabilistic element during output creation. This element is, however, completely different from, and less impactful than, the probabilistic nature of LLMs.</p><p><strong>Optimisation algorithms use randomised selection to influence intermediate output</strong>. This intermediate output is then scored by the objective function — part of the formally defined digital twin. Intermediate output that doesn’t fit the digital twin is filtered out by scoring that output poorly. This way, <strong>inaccuracies arising from the probabilistic element are removed before the final output</strong>, thereby eliminating their potential impact.</p><p>In contrast, the probabilistic aspect of LLMs directly influences the final output, so poor generation is retained, resulting in hallucinations. Current versions of LLMs tend to filter out hallucinations with <a href="https://en.wikipedia.org/wiki/AI_safety#AI_guardrails">guardrails</a>, but as they too are probabilistic in nature, the guardrails are<a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC12532640/"> not foolproof</a> [<a href="https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2025.1622292/full#s9">1</a>].</p><p>There is one final difference between LLMs and optimisation algorithms worth mentioning. Output from optimisation algorithms is easily understood by domain experts. The score gives insight into how desirable it is, and a detailed analysis can often be provided. As such, <strong>optimisation algorithms provide a level of explainability</strong>, increasing the<a href="https://digital-strategy.ec.europa.eu/en/library/ethics-guidelines-trustworthy-ai"> trustworthiness</a> of the algorithm [<a href="https://blog.dotsandlines.ai/trustworthy-ai-in-optimisation-668b3f82997b">1</a>].</p><h3>Conclusion</h3><p><strong>LLMs suffer from hallucinations</strong> and, in recent times, people have been assuming that optimisation algorithms suffer from them as well. We discussed two causes for hallucinations in LLMs: a misrepresentation of the real world and the probabilistic nature of the algorithms. Optimisation models also build upon representations of the real world and probabilistic elements. The impact of these characteristics on optimisation output, however, is of an entirely different magnitude.</p><p>In optimisation, the formal definition process differs from how data represents the real world in LLMs. <strong>Formal problem definition supports transparency, completeness of representation, and process accountability</strong>, key aspects of <a href="https://digital-strategy.ec.europa.eu/en/library/ethics-guidelines-trustworthy-ai">trustworthy AI</a>, as they increase the understanding of inaccuracies and allow for the mitigation of most risks. LLMs, on the other hand, lack those features, making inaccuracies unrecognisable and unattributable, and therefore potentially more damaging.</p><p>In optimisation, the <strong>probabilistic element only works on intermediate output</strong>. This is evaluated and filtered out by the objective function throughout the process, before it can influence the final output. In contrast, the probabilistic element in LLMs functions directly on the final output. Finally, the superior <strong>explainability</strong> <strong>of optimisation output improves the </strong><a href="https://blog.dotsandlines.ai/trustworthy-ai-in-optimisation-668b3f82997b"><strong>trustworthiness</strong></a> of optimisation algorithms.</p><p>In summary, <strong>optimisation algorithms don’t suffer from hallucinations</strong>. Inaccuracies may occur, but these issues are entirely different from hallucinations in LLM. They are easier to detect and more straightforward to address.</p><h3>Written by</h3><p><a href="https://www.linkedin.com/in/leon-ingelse/">Leon Ingelse</a> (<a href="mailto:leon@dotsandlines.ai">leon@dotsandlines.ai</a>), ethical AI researcher at <a href="https://dotsandlines.ai">Dots &amp; Lines</a></p><h3>Edited by</h3><p><a href="https://www.linkedin.com/in/goranmolnar/">Goran Molnar</a> (<a href="mailto:goran@dotsandlines.ai">goran@dotsandlines.ai</a>), CEO at <a href="https://dotsandlines.ai">Dots &amp; Lines</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9f727dd0a13b" width="1" height="1" alt=""><hr><p><a href="https://blog.dotsandlines.ai/why-optimisation-doesnt-suffer-from-hallucinations-9f727dd0a13b">Why optimisation doesn’t suffer from hallucinations</a> was originally published in <a href="https://blog.dotsandlines.ai">Dots &amp; Lines</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Trustworthy AI in optimisation for scheduling and routing]]></title>
            <link>https://blog.dotsandlines.ai/trustworthy-ai-in-optimisation-3f545f719a40?source=rss-1879184bbde0------2</link>
            <guid isPermaLink="false">https://medium.com/p/3f545f719a40</guid>
            <category><![CDATA[optimization]]></category>
            <category><![CDATA[routing]]></category>
            <category><![CDATA[scheduling]]></category>
            <category><![CDATA[responsible-ai]]></category>
            <category><![CDATA[ethical-ai]]></category>
            <dc:creator><![CDATA[Leon Ingelse]]></dc:creator>
            <pubDate>Thu, 12 Mar 2026 10:26:00 GMT</pubDate>
            <atom:updated>2026-03-12T22:32:00.130Z</atom:updated>
            <content:encoded><![CDATA[<p><em>This is the second post of a series on Trustworthy AI in optimisation. You can find the first post </em><a href="https://blog.dotsandlines.ai/trustworthy-ai-in-optimisation-668b3f82997b"><em>here</em></a><em>.</em></p><p>The risks that come with the adoption of artificial intelligence (AI) models have led to the need for Trustworthy AI. Most research, legislation and frameworks are focused on mainstream AI, which is only part of the story.</p><p>As we discussed in the <a href="https://blog.dotsandlines.ai/trustworthy-ai-in-optimisation-668b3f82997b">first part</a> of this blog series, there are foundational differences between mainstream AI, which is data-driven, and optimisation, a constraint-based sub-field that Dots &amp; Lines uses for scheduling and routing. As a result, optimisation carries its own set of specific risks that differ significantly from those of mainstream AI, and we’ve categorised these risks into the following three areas:</p><ul><li><strong>Misalignment</strong> between the digital twin and the real-world process or problem at hand,</li><li><strong>In-algorithm bias</strong>, and</li><li>A failure to comply with <strong>laws and regulations</strong></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*m9R7q1CbeeW5rJDzJw1PfA.jpeg" /><figcaption><a href="https://www.shutterstock.com/image-illustration/advanced-3d-turbine-engine-concept-digital-2577081637?trackingId=572c2ebe-f3f6-4fc5-86d0-04f7aa6e2ba2&amp;listId=searchResults"><em>Image</em></a><em> by </em><a href="https://www.shutterstock.com/g/gorodenkoff"><em>Gorodenkoff</em></a><em> from Shutterstock</em></figcaption></figure><h3>Misalignment</h3><p>Misalignment between the digital twin and the real-world process happens when developers and process stakeholders are not aligned on their understanding of the process. If the digital twin misrepresents the real-world process, the optimised solutions might look optimal on paper, but will lack usability as they fail to function in the real world. Besides losing user and client trust, there are risks when constraints or details are critical but not included in the digital twin.</p><p>For example, if you are creating a routing solution for truck drivers, ensuring they get sufficient rest on their journeys is important for safety reasons. For someone who has never driven a truck, it’s impossible to understand what “sufficient rest” means to truck drivers. We might think a 15-minute break every 2 hours is enough, but they will also need longer breaks to eat and sleep. There are even potential personal needs that apply to each truck driver. For the intricacies of a truck driver’s life, it’s best to speak to a truck driver. This also has the added benefit of helping them to adopt the software with confidence.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yorPqLVUzYmTClXeCiWcjQ.jpeg" /><figcaption><em>“For the intricacies of a truck driver’s life, it’s best to speak to a truck driver.” </em><a href="https://commons.wikimedia.org/wiki/File:Driver_beside_CF_Truck,_Feb._67.jpg"><em>Image</em></a><em> from the Creative Commons, licensed by </em><a href="https://creativecommons.org/publicdomain/zero/1.0/deed.en"><em>CC0 1.0</em></a><em>.</em></figcaption></figure><p>The above showcases the importance of including the user as a key stakeholder. In their release of GPT-5, OpenAI had focused on “PhD-level” expertise in almost any subject, but hadn’t realised their users cared more about the personal touch, leading to <a href="https://lumiera.beehiiv.com/p/gpt-5-hype-rollout-backlash-54e2">a massive product disappointment</a>.</p><p>Another problem is the optimisation objective. There is a well-known thought experiment called the <a href="https://en.wikipedia.org/wiki/Instrumental_convergence#Paperclip_maximizer">paperclip</a> maximiser, in which an AI system trained to optimise paperclip production might begin working against humans — possibly even killing them if that would benefit paperclip production. Although this thought experiment is vastly exaggerated, it aims to illustrate how optimising towards a certain goal might have unintended side effects.</p><p>In our work, we face similar challenges. The easiest way to cut costs is to fire everyone, but our clients clearly don’t mean this when they say they want to minimise costs. During his time as a researcher, our CEO <a href="https://www.linkedin.com/in/goranmolnar/">Goran Molnar</a> worked on a <a href="https://www.sciencedirect.com/science/article/abs/pii/S0191261515001265">transport optimiser</a>. An early version of it set all the prices too high, way beyond anything anyone in their right mind would accept. The profit was zero, nobody was using the system, but the costs were minimised. This optimiser effectively chose to close down the transport company. This example is amusing and easy to spot, but misalignment often happens in more subtle ways. For instance, we might implement an algorithm and then find out that a <em>hard constraint</em> is actually a <em>soft constraint</em> in exceptional cases, and this could have drastic effects on the outcomes.</p><p>Even though we can invest heavily in stakeholder involvement, continuous communication with developers, and testing, software can never capture all the nuance and messiness of the real world, especially that hard-to-define “human touch”. For example, a schedule might work well mathematically yet still trigger responses from schedulers like: “that looks great, but something’s off”. The intuition and experience of schedulers can pinpoint some relevant aspects which are impossible to code into a computer. That’s why a human in the loop is paramount for implementing optimisation. Let algorithms do the mathematical heavy lifting, and leave schedulers to apply the final human touch.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aYU4LPg2KMq18NnJlj-hWA.jpeg" /><figcaption><em>“Let algorithms do the mathematical heavy lifting, and leave schedulers to apply the final human touch.” </em><a href="https://www.shutterstock.com/image-photo/close-developer-discussion-about-app-flowchart-2466333537"><em>Image</em></a><em> by </em><a href="https://www.shutterstock.com/g/weedezign"><em>wee dezign</em></a><em> from Shutterstock.</em></figcaption></figure><h3>In-algorithm bias</h3><p>Bias is a common problem in AI. In mainstream AI, there are many instances of data-induced bias, where the data used for training the AI system contains bias towards certain predictions. Since optimisation algorithms are based on digital twins, data-induced bias is not such an issue. However, there are risks of bias deriving from the inner workings of the algorithm.</p><p>For example, while prototyping a scheduling solution for one of our first clients, we received complaints that holiday preferences were often granted to employees with names starting with a letter at the beginning of the alphabet. We realised that our algorithm started scheduling from the top of the employee list. As there were more shifts to distribute for the first employees, it was easier for the algorithm to grant them their preferences, resulting in more advantageous schedules for the employees at the top of the list.</p><p>To minimise in-algorithm bias, and we cannot stress this enough, testing is paramount. Testing helps alignment with the digital twin and ensures that all theorised elements function accordingly in the real world. This demands regular communication between developers and other stakeholders and rigorous testing on real-world scenarios.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*89IkSw5qox1kQ16lf0n1Ww.jpeg" /><figcaption><a href="https://www.shutterstock.com/image-photo/health-care-professional-working-medical-checklist-2571961861"><em>Image</em></a><em> by </em><a href="https://www.shutterstock.com/g/donut8449"><em>khunkornStudio</em></a><em> from Shutterstock.</em></figcaption></figure><p>However, not all bias can be resolved and sometimes bias is even desired or necessary. We already discussed prioritising certain employee preferences over others. We might want our algorithms to prioritise someone who wishes to attend a child’s wedding over someone’s plan to go for a coffee with a friend. We are also experimenting with a “karma” system that monitors the number of preferences granted to each employee. In this way, an employee might receive a less preferential schedule one week, but the tracking system will ensure that they receive more favourable schedules later on, creating more equitable schedules in the long run.</p><p>However, the fact remains that most bias is undesired. As the complex problems we tackle have no single method to solve them, they require probabilistic algorithms. Understanding the inner workings of this type of algorithm is complicated by their nondeterministic nature, partly obscuring why an algorithm presents a specific schedule and any potential bias it may apply. The extent to which a model’s functioning can be interpreted is called <em>interpretability</em>. This differs from <em>explainability</em>, which covers the ability to explain a solution within the problem context. In a future blog post, we will provide a more detailed explanation of interpretability and explainability, together with a comparison of these concepts in relation to data-based AI and optimisation.</p><p>To minimise the impact of algorithmic bias, it is essential to provide <em>explainability</em> and <em>human-in-the-loop</em> features. We have supplied our clients with various explainability functions to help schedulers understand the resulting schedules more easily. One example is a colour-coded system to highlight which constraints are broken for a schedule and illustrate, to some extent, the pros and cons of a schedule. While we cannot determine whether a solution is “the best possible”, humans can understand whether a solution is “good enough”. Having a human in the loop helps greatly with compliance and bias prevention.</p><h3>A failure to comply with <strong>laws and regulations</strong></h3><p>Many areas of optimisation require compliance with applicable regulations. In the context of scheduling and routing, this often entails employment law, but sometimes specialised laws also come into play. Optimisation brings both opportunities and risks for ensuring compliance.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BKcI9uztPpxbl3JkyOA2Qw.jpeg" /><figcaption><a href="https://www.shutterstock.com/image-photo/law-theme-mallet-judge-enforcement-officers-1938588952">Image</a> by <a href="https://www.shutterstock.com/g/photobyphotoboy">photobyphotoboy</a> from Shutterstock.</figcaption></figure><p>Including legislation in the digital twin can help businesses adhere to these laws. Aspects of employment law, such as the minimum amount of free time between shifts, holiday requirements, sick leave, and shift break legislation, add extra complexity for a scheduler to juggle. Algorithms can manage many competing constraints, allowing for schedules that adhere to legislation while optimising for business and stakeholders’ needs.</p><p>On the other hand, overlooking laws or implementing them incorrectly can result in fines. This is even more difficult when businesses operate across borders, requiring them to comply with various jurisdictions. In the EU and the US, this is further complicated by local, federal and union laws.</p><p>In certain domains, high safety standards and specialised workforces require extra attention. In <a href="https://blog.dotsandlines.ai/how-to-optimise-air-traffic-control-schedules-076b25ed74cd">our work with Norway’s aviation company Avinor</a>, we were required to always comply with highly specialised air traffic control regulations, while maintaining the safety standards of one of the safest airspaces in the world.</p><h3>Conclusion</h3><p>AI has become a buzzword in recent years due to its far-reaching capabilities. However, multiple examples have shown how AI can cause scandals for governments and businesses.</p><p>In our field of optimisation, we face risks that differ significantly from the risks of mainstream AI. Most notably, and very appreciated by our users, our models do not hallucinate. This is due to the way our models are built. Mainstream AI relies heavily on data. Conversely, optimisation is based on formalising business processes and then developing a digital twin. Scheduling and routing are two areas that can particularly benefit from optimisation’s reliable solutions.</p><p>What are the risks you recognise in your area? Please feel free to comment or contact us 🙂</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IICb6a0u4y5x29H9CecAXw.png" /><figcaption><a href="https://betterimagesofai.org/images?artist=LoneThomasky&amp;title=DistortedFishSchool">Distorted Fish School</a> by <a href="https://lone-thomasky.de/">Lone Thomasky</a> &amp; <a href="https://bits-und-baeume.org/en/">Bits&amp;Bäume</a>, <em>licensed by </em><a href="https://creativecommons.org/licenses/by/4.0/"><em>CC-BY 4.0</em></a><em>.</em></figcaption></figure><h3>Written by</h3><p><a href="https://www.linkedin.com/in/leon-ingelse/">Leon Ingelse</a> (<a href="mailto:leon@dotsandlines.ai">leon@dotsandlines.ai</a>), ethical AI researcher at <a href="https://dotsandlines.ai">Dots &amp; Lines</a></p><h3>Edited by</h3><p><a href="https://www.linkedin.com/in/goranmolnar/">Goran Molnar</a> (<a href="mailto:goran@dotsandlines.ai">goran@dotsandlines.ai</a>), CEO at <a href="https://dotsandlines.ai">Dots &amp; Lines</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3f545f719a40" width="1" height="1" alt=""><hr><p><a href="https://blog.dotsandlines.ai/trustworthy-ai-in-optimisation-3f545f719a40">Trustworthy AI in optimisation for scheduling and routing</a> was originally published in <a href="https://blog.dotsandlines.ai">Dots &amp; Lines</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Trustworthy AI in Optimisation]]></title>
            <link>https://blog.dotsandlines.ai/trustworthy-ai-in-optimisation-668b3f82997b?source=rss-1879184bbde0------2</link>
            <guid isPermaLink="false">https://medium.com/p/668b3f82997b</guid>
            <category><![CDATA[responsible-ai]]></category>
            <category><![CDATA[ethical-ai]]></category>
            <category><![CDATA[digital-twin]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[optimization]]></category>
            <dc:creator><![CDATA[Leon Ingelse]]></dc:creator>
            <pubDate>Thu, 05 Mar 2026 13:20:04 GMT</pubDate>
            <atom:updated>2026-03-12T10:36:29.956Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>“<a href="https://press.un.org/en/2025/sgsm22830.doc.htm">Innovation must serve humanity — not undermine it</a>” — António Guterres, UN Secretary General, at the UN Security Council on September 24, 2025.</blockquote><p><em>This blog post is part of a series on Trustworthy AI in optimisation. Find the second part, on the implications for routing and scheduling, </em><a href="https://blog.dotsandlines.ai/trustworthy-ai-in-optimisation-3f545f719a40"><em>here</em></a><em>.</em></p><p>In recent years, “Artificial Intelligence” (AI) has become an omnipresent buzzword. Depending on who you listen to, AI could either save or destroy the world. Others think it’s merely <a href="https://blog.dotsandlines.ai/its-a-genai-bubble-0fb6a31c6f01">a bubble</a>. The facts are that AI has contributed to incredible innovation on one hand, such as <a href="https://ieeexplore.ieee.org/abstract/document/10464455">translation</a>, <a href="https://www.nature.com/articles/d41586-024-03214-7">protein-structure prediction</a>, and <a href="https://blog.dotsandlines.ai/how-to-optimise-air-traffic-control-schedules-076b25ed74cd">scheduling</a>, as well as profound negative social impacts on the other, such as <a href="https://globalvoices.org/2023/06/29/undertones-if-on-welfare-better-learn-dutch-and-be-a-man/">government collapse</a>, <a href="https://www.bbc.com/news/technology-33347866">AI systems that perpetuate racism</a>, and <a href="https://www.independent.co.uk/tech/chatgpt-sexual-harassment-law-professor-b2315160.html">disinformation</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Nk81IsMUHyR2qpwNXQIXnA.jpeg" /><figcaption><a href="https://betterimagesofai.org/images?artist=GloriaMendoza&amp;title=TheEnvironmentalImpactofDataCentersinVulnerableEcosystems"><em>The Environmental Impact of Data Centers in Vulnerable Ecosystems</em></a><em> by </em><a href="https://www.gloriamc.xyz/"><em>Gloria Mendoza</em></a><em>, licensed by </em><a href="https://creativecommons.org/licenses/by/4.0/"><em>CC-BY 4.0</em></a><em>.</em></figcaption></figure><p>AI users and developers often have the best intentions but lack the skills, knowledge, and experience to recognise risks. In worst-case scenarios, unmitigated risks can result in loss of life, legal issues, and a ruined reputation. In other cases, they might result in fines, unfair treatment of staff, and financial costs. These risks have led to the need for what’s known as “Trustworthy AI”.</p><p>At Dots &amp; Lines, we face specific risks that are characteristic of the AI subfield of optimisation, also known as <a href="https://en.wikipedia.org/wiki/Mathematical_optimization">mathematical optimisation</a>, which encompasses constraint-based models that rely on developing a digital twin based on a formal problem definition. This contrasts with mainstream AI, i.e., predictive and generative methods such as machine learning, neural networks, LLMs and chatbots like ChatGPT, which rely on vast amounts of data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*E0Zkn0HFxy4kLCC8Sx3rCg.jpeg" /><figcaption><em>The models we use at Dots &amp; Lines rely less on data and more on good old mathematics. </em><a href="https://www.shutterstock.com/image-photo/image-mathematical-equations-diagrams-against-abstract-2479199065"><em>Image</em></a><em> by </em><a href="https://www.shutterstock.com/g/vectorfusionart"><em>vectorfusionart</em></a><em> from Shutterstock.</em></figcaption></figure><p>The risks of mainstream AI have been widely studied, since it is the subject of Trustworthy AI research and legislation. The risks of optimisation, however, have not received the same attention. In this blog post, we discuss the types of AI risks and the difference between optimisation and mainstream AI. In a <a href="https://blog.dotsandlines.ai/trustworthy-ai-in-optimisation-3f545f719a40">follow-up post</a>, we share insights from our experience using optimisation for scheduling and routing.</p><h3>Types of risks</h3><p>AI risks can be divided into two groups. The first group consists of risks from unethical AI usage or development. Examples are the <a href="https://www.cjr.org/tow_center/qa-uncovering-the-labor-exploitation-that-powers-ai.php">exploitation of data workers</a>, <a href="https://apnews.com/article/deepfake-ai-nudes-teen-girls-legislation-b6f44be048b31fe0b430aeee1956ad38">AI-generated nudes of real people</a>, and the use of <a href="https://unric.org/en/un-addresses-ai-and-the-dangers-of-lethal-autonomous-weapons-systems/">AI in warfare</a>. Most business leaders understand the importance of responsible development for their company’s success and recognise that ethical innovation goes hand in hand with their business goals.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0BB8s09QMQLLQn6GcZT-DQ.png" /><figcaption><a href="https://betterimagesofai.org/images?artist=JanetTurra&amp;title=DataWorker"><em>Data Worker</em></a><em> by </em><a href="https://janetturra.wordpress.com/"><em>Janet Turra</em></a><em>, licensed by </em><a href="https://creativecommons.org/licenses/by/4.0/"><em>CC-BY 4.0</em></a><em>.</em></figcaption></figure><p>The second group of risks in AI adoption are the inherent risks due to potentially using and developing AI in an uninformed or incorrect fashion. Although the aims might be ethical, uninformed AI usage could cause unintentional harm.</p><p>For example, two US lawyers used ChatGPT to bolster the defence for one of their clients in a court case. Little did they know that ChatGPT is famous for generating false statements, known as <a href="https://en.wikipedia.org/wiki/Hallucination_(artificial_intelligence)">hallucinations</a>, and <a href="https://www.reuters.com/legal/new-york-lawyers-sanctioned-using-fake-chatgpt-cases-legal-brief-2023-06-22/">their defence was full of non-existent references</a>. One lawyer explained that he had thought <a href="https://nypost.com/2023/06/08/humiliated-lawyer-apologizes-over-chatgpt-court-flub/">ChatGPT was a “search engine”</a>. As well as a fine, this debacle resulted in a painfully embarrassing experience for the lawyers and their firm. Recently, Deloitte had <a href="https://apnews.com/article/australia-ai-errors-deloitte-ab54858680ffc4ae6555b31c8fb987f3">a similar scandal</a>, where they provided the Australian government with a report containing fabricated references.</p><p>Hallucinations, as in the examples above, are a risk specific to transformer models like LLMs. Optimisation algorithms, which we use at Dots &amp; Lines, don’t hallucinate but come with other, more subtle risks. To understand the risks, let’s first look at how our models work.</p><h3>Optimisation</h3><p>When people think about AI, they generally think of data-driven AI systems, which function by extracting patterns from large swathes of data. These patterns can be used to say something about the data, to make predictions about unseen data by extrapolation, and to generate new data. As these models derive their capabilities from data, many risks come from insufficient or unrepresentative data, such as data-induced bias and interpretability issues.</p><p>Optimisation works differently, and this presents a distinct set of risks. For example, optimisation doesn’t suffer from hallucinations. Instead, the risks inherent in optimisation derive from its foundation: optimising a <em>digital twin</em> of a formalised business process. Let’s unpack these concepts.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*m9R7q1CbeeW5rJDzJw1PfA.jpeg" /><figcaption>A digital twin of a turbine engine. <a href="https://www.shutterstock.com/image-illustration/advanced-3d-turbine-engine-concept-digital-2577081637?trackingId=572c2ebe-f3f6-4fc5-86d0-04f7aa6e2ba2&amp;listId=searchResults">Image</a> by <a href="https://www.shutterstock.com/g/gorodenkoff">Gorodenkoff</a> from Shutterstock</figcaption></figure><p><a href="https://en.wikipedia.org/wiki/Business_process">Business processes</a>, such as scheduling and routing, are a group of interrelated activities with a specific business objective as their goal. Normally, a process evolves with the business and is therefore not often formalised. You might remember starting a new job, asking about an aspect of the job, and receiving contradictory advice from colleagues. Person A tells you to speak to person B, while person B sends you back to person A, leaving you confused about which process to follow. While informal business processes might work well enough, organisations can benefit from an expert assessment of their processes. This is increasingly relevant for large businesses, where inefficiencies propagate across numerous divisions.</p><p>When working with our clients on scheduling optimisation, we’ve encountered a specific example of this. Schedulers must handle employee work preferences, which requires them to judge the importance of each employee’s request. Whose request should be granted when employee A has been planning a party for a year, employee B has an extra football practice because they have the chance of winning the championship next Sunday, and employee C wants to spend time with their children? A scheduler’s judgement might be influenced by their love for parties, childhood dreams of football championships or experience having kids.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*deWD4FkMF2k38_XIfi_F2w.jpeg" /><figcaption><em>Formalising a business process requires bringing stakeholders together. </em><a href="https://www.shutterstock.com/image-photo/corporate-business-team-manager-meeting-1420666514"><em>Image</em></a><em> by </em><a href="https://www.shutterstock.com/g/nd3000"><em>NDAB Creativity</em></a><em> from Shutterstock.</em></figcaption></figure><p>At Dots &amp; Lines, we formalise business processes through company discussions and alignment, which helps a business move forward by creating clarity and improving efficiency. This requires the scheduler to sit down with employees and managers to create formal rules around how and when to grant preferential requests, and to decide together which requests should have priority.</p><p>With a formalised business process in place, we can build a <a href="https://en.wikipedia.org/wiki/Digital_twin">digital twin</a> — a digital representation of the process. Imagine using a toy railway representing a country’s railway system to experiment with new tracks or railway lines before implementing them in real life. Unlike the real railway system, where a poor schedule can lead to crashes or delays, applying these experiments in a computer simulation helps us detect issues with the decisions and changes in the system.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5zyfhk3GTJzlNdph3gX-nw.jpeg" /><figcaption><a href="https://www.shutterstock.com/image-photo/electric-model-train-on-railway-stands-2257135767">Toy railway image</a> by <a href="https://www.shutterstock.com/g/Lepasik">Anelo</a> from Shutterstock.</figcaption></figure><p>Our algorithms use the digital twin to create optimised solutions for the relevant business process. For scheduling, we search for schedules that adhere to <em>hard constraints</em> — unbreakable rules like employment law, and physical impossibilities like two teachers using the same classroom — while optimising <em>soft constraints</em> — less rigid rules and objectives, such as employee preferences, excessive costs due to overtime, and emissions.</p><p>As you can see, optimisation is distinct from mainstream AI. Optimisation’s uniqueness also brings specific ethical risks. In our next blog, we will discuss the main risk areas that we have recognised in our years of working with optimisation in scheduling and routing. Stay tuned 🙂</p><h3>Written by</h3><p><a href="https://www.linkedin.com/in/leon-ingelse/">Leon Ingelse</a> (<a href="mailto:leon@dotsandlines.ai">leon@dotsandlines.ai</a>), ethical AI researcher at <a href="https://dotsandlines.ai">Dots &amp; Lines</a></p><h3>Edited by</h3><p><a href="https://www.linkedin.com/in/goranmolnar/">Goran Molnar</a> (<a href="mailto:goran@dotsandlines.ai">goran@dotsandlines.ai</a>), CEO at <a href="https://dotsandlines.ai">Dots &amp; Lines</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=668b3f82997b" width="1" height="1" alt=""><hr><p><a href="https://blog.dotsandlines.ai/trustworthy-ai-in-optimisation-668b3f82997b">Trustworthy AI in Optimisation</a> was originally published in <a href="https://blog.dotsandlines.ai">Dots &amp; Lines</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Dots & Lines at the ROAR-NET event]]></title>
            <link>https://blog.dotsandlines.ai/dots-lines-at-the-roar-net-event-9141636c89b1?source=rss-1879184bbde0------2</link>
            <guid isPermaLink="false">https://medium.com/p/9141636c89b1</guid>
            <category><![CDATA[hackathons]]></category>
            <category><![CDATA[optimization]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[academia]]></category>
            <category><![CDATA[junior-developer]]></category>
            <dc:creator><![CDATA[Leon Ingelse]]></dc:creator>
            <pubDate>Mon, 22 Sep 2025 13:19:35 GMT</pubDate>
            <atom:updated>2025-11-27T11:38:08.770Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Update: The </em><a href="https://code-fest.roar-net.eu/achievements/training"><em>results</em></a><em> are in, and </em>NullPointerException, <em>the team of Kim and Luka, came in third. Congratulations!</em></p><p>This month, two of <a href="https://dotsandlines.ai/">Dots &amp; Lines</a>’s developers participated in <a href="https://roar-net.eu/calls/code-fest-call/">ROAR-NET’s Problem Modelling Code Fest event</a>. <a href="https://www.linkedin.com/in/kimstanicic/">Kim</a> and <a href="https://www.linkedin.com/in/lukanestic/">Luka</a> formed a team with their friend <a href="https://hr.linkedin.com/in/paulo-erak-64bb1a32a">Paulo Erak</a> and participated in the 48-hour code fest around problem modelling organised across 21 hubs spanning five continents. Croatia and Portugal, where most of our staff are located, hosted three hubs. It’s great to see such interest!</p><p>Coding events allow our developers to participate in the community, create a network, learn and share knowledge and have fun! For this blog post, we spoke with Kim and Luka about their experience.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eUSS9rbCn2lg7hNDpH83jg.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RAYvBvRZ92Oh0LYYT2Ytwg.jpeg" /><figcaption>Kim and Luka during the Code Fest</figcaption></figure><p><strong>How was the event?</strong></p><p>We were relieved that the ROAR event wasn’t a standard hackathon, but a more relaxed coding event with a competition element added to it — hence the name Code Fest. The hosting professors did a great job covering basic optimisation concepts needed to complete the tasks via lectures that covered all time zones. There were two live coding sessions from which we learned how to implement <a href="https://en.wikipedia.org/wiki/Constructive_heuristic">constructive</a> and <a href="https://en.wikipedia.org/wiki/Local_search_(optimization)">local search</a> in the <a href="https://github.com/roar-net/roar-net-api-py">ROAR-NET API</a> in Python.</p><p>Apart from online participation, multiple hubs were organised across the globe where local teams could come together to solve their tasks. The hub in Zagreb, organised by<a href="https://www.linkedin.com/in/djakobovic/"> Domagoj Jakobović</a>, served both as a working and a social environment. Here, we followed ROAR lectures, discussed ideas and socialised with other local teams. Nevertheless, we missed further interaction with the rest of the teams; we only knew about them through the evaluation website, where we could see their uploaded solutions on the ranking list.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*is1Lp2bM3O2ROxvMsM_1Hg.jpeg" /><figcaption>Coding events like hackatons are great places for developers to learn and engage with fellow enthusiasts. <a href="https://www.flickr.com/photos/35034357401@N01/6424269505">Image</a> from Flickr by <a href="https://www.flickr.com/photos/breyten/">Breyten Ernsting</a>, licensed under the <a href="https://creativecommons.org/licenses/by/2.0/deed.en">Attribution 2.0 Generic licence</a>.</figcaption></figure><p><strong>What problems did you have to solve at the code fest?</strong></p><p>Every team on the Training track was tasked with modelling three combinatorial problems using the ROAR-NET API. The problems were well described with given input and output examples, so they were easily understandable. Because of the nature of the task and our cooperation, it seemed as if we were doing a coding lab exercise in uni, which made the entire experience both enjoyable and highly educational.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*j7ey9UDQrNyUSKpakne90A.jpeg" /><figcaption>Marketing Campaign Problem: We had to determine the set of people with the minimum cost to hire from a social network so that every connection in the network was covered. <a href="https://www.flickr.com/photos/86530412@N02/7975205041">Image</a> from Flickr by <a href="https://www.flickr.com/photos/86530412@N02/">Chris Potter</a>, licensed under the <a href="https://creativecommons.org/licenses/by/2.0/deed.en">Attribution 2.0 Generic licence</a>.</figcaption></figure><p><strong>How did participating help you connect academia and industry, and what did you learn?</strong></p><p>Our team participated in the training track, which focused on modelling predefined problems similar to those we witnessed during our uni days. Another track, the development track, focused on bettering the existing ROAR-NET code and proposing new ideas. The development track was more industry-related than ours.</p><p>We solved simple, straightforward problems distinct from the complex, real-world problems we tackle at Dots &amp; Lines. These more academic examples made us revisit the optimisation basics: how to model a problem. This step is actually the first we encounter with every new project in the industry: translating the problem from regular words to structured code that is cohesive, maintainable, and easily understood by both humans and the computer. A good problem model, also known as <a href="https://en.wikipedia.org/wiki/Digital_twin">a digital twin</a>, makes the optimisation part more straightforward and enjoyable.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4p6J1ayC8fwQyjN8gt_UPg.jpeg" /><figcaption>Candle Race Problem: There is a set of villages with lit candles in each, burning at different rates. We had to determine the best path for visiting all villages and blowing out their candles if the goal is for each candle to lose as little of its length as possible. <a href="https://www.flickr.com/photos/90762294@N02/54321437514">Image</a> from Flickr by <a href="https://www.flickr.com/photos/90762294@N02/">Zyuflyuc</a>, in the <a href="https://creativecommons.org/publicdomain/mark/1.0/deed.en">public domain</a>.</figcaption></figure><p><strong>Any final thoughts?</strong></p><p>Luka: The ROAR code fest was inspiring and educational. The vibes were great, and I enjoyed team brainstorming ideas and implementing the problems that piqued my interest. Praise to all hosting professors, who were supportive and open to questions and remarks. I was relieved this wasn’t a standard “hardcore” hackathon because that would mean sleepless nights and stressing about how many points we would get. Instead, the whole experience was fun, also because we got to contribute to the ROAR-NET API.</p><p>Kim: This was a very fun event with just the right amount of self-inflicted ‘stress’ and wanting to solve all six tasks. It was reminiscent of my uni days, when I tried to solve coding lab exercises with the online evaluator rejecting my submissions. Only this time, I had my teammates, who thankfully encountered similar issues, so we had fun debugging our code together. I’m satisfied with our results, as we submitted all six tasks, of which the online evaluator accepted four, and had a blast working on these tasks with Luka and Paulo. I think all of us learned something valuable, ranging from problem modelling to the power of teamwork. Can’t wait for the next one!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ibKb4ctH7Cr4MWr_MD6QdQ.jpeg" /><figcaption>Urban Waste Collection Problem: We had to find the shortest route for a waste collection vehicle to visit all containers. <a href="https://www.shutterstock.com/image-photo/selective-focus-hand-throwing-empty-plastic-682001986">Image</a> from Shutterstock by <a href="https://www.shutterstock.com/g/chartphoto3965336">CC7</a>.</figcaption></figure><blockquote><strong>What is ROAR-NET?</strong><br>ROAR-NET (Randomised Optimisation Algorithms Research Network) is a network of optimisation academics and practitioners that aims to make ROAs competitive in practice. It focuses on the needs of practitioners, from which the economic value of optimisation stems, and is funded by the <a href="https://www.linkedin.com/company/cost-office/">COST association — European Cooperation in Science and Technology</a>.</blockquote><blockquote><strong>What is ROAR-NET’s Problem Modelling Code Fest?<br></strong>At the code fest, participants are taught how to use the ROAR-NET API to solve combinatorial optimisation problems and participate in a problem-solving competition. The problems were designed by <a href="https://www.linkedin.com/in/carlos-m-fonseca-390aa8331/">Prof. Carlos M. Fonseca</a> and <a href="https://www.linkedin.com/in/adbjesus/">Alexandre Jesus</a>.</blockquote><blockquote><strong>Why?<br></strong>Besides the direct benefit for Kim and Luka in the form of technical skills, networking and fun, Dots &amp; Lines is devoted to fostering optimisation communities. There is a huge opportunity for academia and industry to learn from each other to improve optimisation’s potential and real-world benefits.</blockquote><p><strong>Written by</strong></p><p><a href="https://www.linkedin.com/in/kimstanicic/">Kim Stanicic</a> (kim@dotsandlines.ai), junior developer at <a href="https://dotsandlines.ai">Dots &amp; Lines</a></p><p><a href="https://www.linkedin.com/in/lukanestic/">Luka Nestic</a> (luka@dotsandlines.ai), junior developer at <a href="https://dotsandlines.ai">Dots &amp; Lines</a></p><p><strong>Written and edited by</strong></p><p><a href="https://www.linkedin.com/in/leon-ingelse/">Leon Ingelse</a> (<a href="mailto:leon@dotsandlines.ai">leon@dotsandlines.ai</a>), ethical AI researcher at <a href="https://dotsandlines.ai">Dots &amp; Lines</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9141636c89b1" width="1" height="1" alt=""><hr><p><a href="https://blog.dotsandlines.ai/dots-lines-at-the-roar-net-event-9141636c89b1">Dots &amp; Lines at the ROAR-NET event</a> was originally published in <a href="https://blog.dotsandlines.ai">Dots &amp; Lines</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The flight of the dandelion seed]]></title>
            <link>https://blog.dotsandlines.ai/the-flight-of-the-dandelion-seed-97a18d63ec54?source=rss-1879184bbde0------2</link>
            <guid isPermaLink="false">https://medium.com/p/97a18d63ec54</guid>
            <category><![CDATA[operations-research]]></category>
            <category><![CDATA[optimization]]></category>
            <category><![CDATA[dandelion]]></category>
            <category><![CDATA[popular-science]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Leon Ingelse]]></dc:creator>
            <pubDate>Mon, 21 Jul 2025 15:25:55 GMT</pubDate>
            <atom:updated>2026-05-31T12:50:05.474Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Researchers discovered a new type of flight in a very unexpected aviator — the dandelion.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UiaJKBqf-CSseJoYwFbE5A.jpeg" /><figcaption><em>A dandelion seed head about to take off. </em><a href="https://www.shutterstock.com/image-photo/dandelion-clock-close-macro-image-1426165466"><em>Photo</em></a><em> by </em><a href="https://www.shutterstock.com/g/Fototocam"><em>Fototocam</em></a><em> from Shutterstock.</em></figcaption></figure><p>Optimisation is ingrained in subtle ways in our daily lives. When walking to the supermarket, most people instinctively choose the best route. When we’re in a rush, we take the shortest path, but when the weather is nice and we feel more leisurely, we might include a detour through the park. We cross the street to walk in the sun on cold days or in the shade on warm days, to optimise journey times, see more sights, or improve comfort.</p><p>Nature is full of ingenious examples of optimisation. We find nature’s elegant designs in the aerodynamics of birds and fish, photosynthesis, the colours and shapes of flowers, trail paths in the forest, and even the hibernation of bears, devised to conserve energy. These designs are beautiful, but also highly efficient. Natural processes such as survival of the fittest and natural selection inspired computer scientists to create one of the most well-known optimisation algorithms: the <a href="https://en.wikipedia.org/wiki/Genetic_algorithm">genetic algorithm</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FVEI0YqYLC1NlIiR-x4sQQ.jpeg" /><figcaption><a href="https://en.wikipedia.org/wiki/Hibernation"><em>Hibernation</em></a><em> helps bears to survive the food-scarce winter months without expending too much energy. </em><a href="https://www.shutterstock.com/image-photo/brown-bear-ursus-arctos-two-cubs-1954391731"><em>Photo</em></a><em> by </em><a href="https://www.shutterstock.com/g/byrdyak"><em>Volodymyr Burdiak</em></a><em> from Shutterstock.</em></figcaption></figure><p>A gem of optimisation found in nature is the dandelion. These plants have a special place in my heart because their yellow flowers adorned many fields of my childhood. After blooming, dandelion flowers contract and return as fluffy seed heads, full of what kids from different cultures (e.g., in Spain, France and Croatia) call “parachutes”. Each parachute is made up of wispy bristles that help these seeds take flight, setting distance records of <a href="https://www.cbc.ca/news/canada/dandelion-seeds-can-fly-100-km-and-now-we-know-how-1.4868619">up to 100 km</a>. The more scientific name of these parachutes is pappus (plural pappi), derived from the Ancient Greek <a href="https://en.wiktionary.org/wiki/%CF%80%CE%AC%CF%80%CF%80%CE%BF%CF%82#Ancient_Greek"><em>πάππος</em></a> (páppos), related to beards and used affectionately for elderly men (and <a href="https://en.wiktionary.org/wiki/%CF%80%CE%AC%CF%80%CF%80%CE%B1%CF%82#Ancient_Greek">closely related to the widespread word for dad: papa</a>).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rYH8VoLLv1vDCdqOKnhblw.jpeg" /><figcaption><em>Yellow dandelion flowers look beautiful in spring. </em><a href="https://www.shutterstock.com/image-photo/latvian-landscape-white-cumulus-clouds-blue-2305130035"><em>Photo</em></a><em> by </em><a href="https://www.shutterstock.com/g/ZelmaB"><em>Zelma Brezinska</em></a><em> from Shutterstock.</em></figcaption></figure><p>As a child, I would challenge myself to blow all the dandelion’s pappi off with one breath. If I succeeded, my wishes would come true. Little did I know that these seeds are some of the world’s best-flying passive objects. I found out how much else I was unaware of when our interest in nature’s remarkable optimisation capabilities led us to explore the flight of the dandelion seed further. Join our exploration in this blog.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*UYVfKbVAf5OdAM6v" /><figcaption><em>Some cultures believe that by blowing all the seeds off a dandelion seed head, your wishes will come true. </em><a href="https://commons.wikimedia.org/wiki/File:Wish.jpg"><em>Photo</em></a><em> by </em><a href="https://www.flickr.com/people/42305326@N05"><em>Jessica Tam</em></a><em>, licensed under the Creative Commons </em><a href="https://creativecommons.org/licenses/by/2.0/"><em>Attribution 2.0 Generic licence</em></a><em>.</em></figcaption></figure><blockquote>Many stories exist around the dandelion’s pappus. In Croatia, some believe your future partner is waiting for you in the direction where the pappi fly. Other <a href="https://wonderopolis.org/wonder/why-do-people-wish-on-dandelions">cultures believe they carry messages to loved ones and lost relatives</a>, and <a href="https://ictnews.org/archive/how-far-can-a-dandelion-seed-fly-ask-a-native-american">the Ojibwe tell a story of the south wind falling in love with the dandelion</a>. What dandelion-related folklore exists in your culture?</blockquote><h3>A separated vortex ring</h3><p>Each pappus contains approximately 100 fine filaments connected to the seed by a delicate stalk. These filaments are positioned in a circle, forming a disc-like structure that partially blocks air flow.</p><p>Not long ago, scientists believed the pappi worked just like children believed them to work — as a parachute to create extra resistance and help them to be carried away, far and wide. Recently, a group of researchers from the University of Edinburgh, led by Ignazio Maria Viola and Naomi Nakayama, <a href="https://www.nature.com/articles/s41586-018-0604-2">showed</a> that the unique characteristics of the pappus enabled it to use a method of flying that was completely unknown to humanity.</p><p>More precisely, when wind blows into the pappus from the direction of the seed, most air is blocked by the filaments and pushed around them. An aerial dance between the air that passes through and above the pappus creates a low-pressure air pocket, where air flows in the form of a vortex ring. This low pressure produces a dragging force above the pappus, keeping it afloat in the air.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/400/0*JoIEpa_O71ejL6iO" /><figcaption><em>Vortex rings are beautiful natural phenomena that enable dandelion seeds to fly. </em><a href="https://commons.wikimedia.org/wiki/File:Vortex_ring.gif"><em>Animation</em></a><em> in the public domain.</em></figcaption></figure><blockquote>Vortex rings can be seen in nature as smoke rings, created by <a href="https://www.youtube.com/watch?v=nkXqP1JSa-U">humans</a> and <a href="https://www.youtube.com/watch?v=BE_bon77-ok">occasionally volcanoes</a>, and underwater ring bubbles, such as those <a href="https://thekidshouldseethis.com/post/dolphins-bubble-rings-vortices-video">blown by dolphins</a>.</blockquote><p>Although vortex rings have been observed in nature before, the one the pappus generates is unique, as it is stable and wholly separate from its source, earning its name of a <em>separated vortex ring</em> (SVR). Although the SVR had been hypothesised before, the work of Viola and Nakayama produced the first observation of the phenomenon!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XzmdgaJioK6z1EQqTowt1w.jpeg" /><figcaption>Mount Etna on Sicily is known for producing smoke rings. <a href="https://www.shutterstock.com/image-photo/vulcano-etna-sicily-most-powerful-2492756863">Photo</a> by <a href="https://www.shutterstock.com/g/Nikk_ng">Nikk_ng</a> from Shutterstock.</figcaption></figure><h3>Nature’s exceptional design</h3><p>After studying the design of the pappus, the <a href="https://pure.hw.ac.uk/ws/portalfiles/portal/25369594/A_separated_vortex_AAM_Cummins_et_al_2018_AAM.pdf">researchers</a> set out to develop artificial discs with similar behaviour. They created discs with variations of porosity — the percentage of air blocked by the disc. Each disc caused air to circulate behind it, but only the discs with similar porosity to the pappus generated the distinctive SVR. The SVR fascinated scientists, but it’s not the only design feature of the dandelion’s pappus that deserves praise. In addition to its capacity to generate an SVR, four design features maximise the pappus’s ability to disperse.</p><p><strong>Number of filaments.</strong> Vortex rings may suffer from a phenomenon called vortex shedding, which, in the case of the dandelion pappus, could destabilise it during flight. Remarkably, a pappus has precisely the number of filaments needed to avoid vortex shedding. <a href="https://journals.aps.org/prfluids/abstract/10.1103/PhysRevFluids.4.071901">A pappus with 100 filaments is perfectly tuned for the existence of an SVR</a> while avoiding vortex shedding — adding or removing only 10 filaments causes the shedding to occur.</p><p><strong>The architecture. </strong>The weight of the dandelion seed ensures the pappus is always on top, while the seed is at the bottom. This enhances stability during flight.</p><p><strong>The material.</strong> Other flying objects in nature, like maple seeds, use permeable membranes to enhance the drag force. Experiments show that using the pappus’s filament design instead of permeable membranes <a href="https://www.nature.com/articles/s41586-018-0604-2">generates a four-times-stronger drag force</a> for the same material, due to two factors: strong interaction between neighbouring filaments and the clever design of filaments, which are over 50% hollow, optimising both weight and coverage area.</p><p><strong>Taking off at the right time.</strong> Perhaps most fascinating is how the pappus “chooses” the right time to take off. As the dandelion’s pappus flies best in dry weather, the top of the pappus straightens its filaments to decrease the drag force during wet weather conditions. With drier weather, the pappus returns to its original parachute shape. This mechanism, <a href="https://www.nature.com/articles/s41467-022-30245-3">which is based on a material that swells in reaction to humidity</a>, <a href="https://biology.ed.ac.uk/plant-sciences/news/how-the-dandelion-decides-when-to-fly">helps the pappus to take off during dry periods</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Ku4P0XPxgJoocdGU" /><figcaption><em>Originally, people believed the pappus worked like a parachute, and it is in fact called a </em>parachute<em> in multiple cultures. </em><a href="https://commons.wikimedia.org/wiki/File:Flickr_-_The_U.S._Army_-_Jump_over_Mt._Rainier.jpg"><em>Photo</em></a><em> in the public domain.</em></figcaption></figure><h3>Nature-inspired human innovation</h3><p>Despite the fact that people might consider the dandelion’s pappus as something common and everyday, studies show it is a highly optimised structure, tuned to maximise flight distance, while minimising material requirements, and even increasing the likelihood of starting the flight under conditions that are <em>just right</em>. The pappus inspired scientists and engineers, as they learned from its successes to drive innovation, e.g. the development of <a href="https://www.nature.com/articles/s41586-021-04363-9">tiny drones</a> with similar flight mechanisms, and potential applications in <a href="https://arc.aiaa.org/doi/abs/10.2514/6.2021-0962">space exploration</a>, <a href="https://www.nature.com/articles/s41467-023-38792-z.pdf">environment surveillance</a> and <a href="https://www.techbriefs.com/component/content/article/45657-floating-sensors-spread-like-dandelion-seeds">agriculture</a>. One group of scientists even tested the <a href="https://www.nature.com/articles/s41586-021-04363-9">pappus-like dispersal of wireless devices</a>. Due to its simple architecture, the ability to contract the filaments has potential for <a href="https://www.nature.com/articles/s41467-022-30245-3">applications at micro-scale in robotics</a>.</p><p>Mathematics and computers helped us optimise complex problems — we developed jet planes to carry us across oceans and rockets to explore space. But centuries after our first balloon flights, people discovered a new way of flying that had been used by dandelions for millennia. Yet another reason to enjoy those golden-yellow fields and make limitless wishes while blowing dandelion pappi into the breeze.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aY3WD29m23myVhrAivAXFA.jpeg" /><figcaption><em>Next time you see a fluffy dandelion seed head, try blowing the seeds into the air while thinking about their ingenious design! </em><a href="https://www.shutterstock.com/image-photo/close-young-woman-blowing-dandelion-seeds-2498185507"><em>Photo</em></a><em> by </em><a href="https://www.shutterstock.com/g/monshtein"><em>Monshtein</em></a><em> on Shutterstock.</em></figcaption></figure><h3>Written by</h3><p><a href="https://www.linkedin.com/in/leon-ingelse/">Leon Ingelse</a> (<a href="mailto:leon@dotsandlines.ai">leon@dotsandlines.ai</a>), writer and researcher at <a href="https://dotsandlines.ai">Dots &amp; Lines</a></p><h3>Edited by</h3><p><a href="https://www.linkedin.com/in/goranmolnar/">Goran Molnar</a> (<a href="mailto:goran@dotsandlines.ai">goran@dotsandlines.ai</a>), CEO at <a href="https://dotsandlines.ai">Dots &amp; Lines</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=97a18d63ec54" width="1" height="1" alt=""><hr><p><a href="https://blog.dotsandlines.ai/the-flight-of-the-dandelion-seed-97a18d63ec54">The flight of the dandelion seed</a> was originally published in <a href="https://blog.dotsandlines.ai">Dots &amp; Lines</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A history of optimisation — from ancient Greece to AI-superpowered scheduling]]></title>
            <link>https://blog.dotsandlines.ai/a-history-of-optimisation-from-ancient-greece-to-ai-superpowered-scheduling-7bd497f1513f?source=rss-1879184bbde0------2</link>
            <guid isPermaLink="false">https://medium.com/p/7bd497f1513f</guid>
            <category><![CDATA[optimization]]></category>
            <category><![CDATA[history-of-science]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[operations-research]]></category>
            <category><![CDATA[history]]></category>
            <dc:creator><![CDATA[Leon Ingelse]]></dc:creator>
            <pubDate>Mon, 23 Jun 2025 17:38:15 GMT</pubDate>
            <atom:updated>2025-06-24T15:03:28.346Z</atom:updated>
            <content:encoded><![CDATA[<h3>A history of optimisation — from ancient Greece to AI-superpowered scheduling</h3><p>When <a href="https://www.linkedin.com/in/goranmolnar/">Goran</a> started his PhD, he felt lost and overwhelmed, like many new researchers. Writing a PhD thesis is a huge, daunting task. In an attempt to get some guidance, he decided to read about the history of the field. The idea was to get a better understanding of his area to help position his work within the broader scientific horizon. Somewhat unexpectedly, this was also highly motivating. Reading about all these brilliant results, sometimes achieved in the face of great adversity, even during wartime, was a great source of determination for him to follow through with his PhD ambitions. After realising that you’re indeed <strong>standing on the shoulders of giants</strong>, even the biggest and most daunting tasks feel much easier.</p><p>This blog explores <strong>the fascinating journey of human optimisation</strong> that has helped build the technology-rich and efficient society we have today. We take you from work carried out in ancient China and Greece to present-day research and our optimisation services at <a href="https://dotsandlines.ai/">Dots &amp; Lines</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*1dcEvxk12AfXRndqhwA1bA.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/969/1*2HY1zNcG70S65lpSKEPEjQ.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*q_Si6nPFalHzL44AoFUVzQ.jpeg" /><figcaption>Goran during and after finishing his PhD.</figcaption></figure><h3>Antiquity</h3><p><strong>Optimisation is omnipresent</strong>. Chemical systems find equilibrium in states of minimum energy. Light rays follow the shortest path between two points. Evolution can be seen as a form of optimisation which ensures that species around us are well adjusted to their environment. It led to cheetahs running at 130 km/h, plants photosynthesising, and the beautiful feather crests of birds like the royal flycatchers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BX79GEJ0ll_HrlUQGCzlPg.jpeg" /><figcaption><em>The crest of the royal flycatcher is an incredible feat of evolution. </em><a href="https://commons.wikimedia.org/wiki/File:Onychorhynchus_coronatus_340547832.jpg"><em>Photo</em></a><em> by </em><a href="https://www.inaturalist.org/users/1682329">Claude Kolwelter</a> (cropped), licensed under the Creative Commons <a href="https://creativecommons.org/licenses/by/4.0/deed.en">Attribution 4.0 International</a> licence.</figcaption></figure><p>Like nature, our ancestors had to optimise continuously. Which animal from a herd makes the best prey? Which water reservoir is closest and safest to access? These examples of optimisation were largely led by instinct and irrational processes. While formal research in the area of optimisation only began with the development of modern science, especially in the computer age, <strong>optimisation, in a nutshell, means <em>making good decisions</em></strong>, and it was present long before modern times.</p><p>Ancient Greeks were fascinated with mathematics. One of their notable optimisation challenges was an <a href="https://mathworld.wolfram.com/IsoperimetricProblem.html"><em>isoperimetric problem</em></a>, which involves determining the geometric shape that generates the largest surface given a fixed perimeter. Geometry was significant for military and land governance. A great example appeared in Virgil’s epic poem <a href="https://www.gutenberg.org/files/228/228-h/228-h.htm">The Aeneid</a>, which discusses the foundation of Carthage around 814 BCE. Banished from her lands, <a href="https://en.wikipedia.org/wiki/Dido">Queen Dido</a> convinced a local ruler near present-day Tunis to grant her as much land as she could enclose within an ox hide. To claim the largest possible area, she cut the ox hide into small strips and <strong><em>encircled</em> a large piece of land</strong>, where she built the city of Carthage.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Rc2XzJUSFXO-pFq02-75YQ.jpeg" /><figcaption>Dido used the <strong><em>circle</em></strong> to acquire as much land as possible to found Carthage. <a href="https://www.istockphoto.com/nl/vector/dido-purchases-land-for-the-foundation-of-carthage-gm471794191-35278898">Photo</a> from iStock, based on an engraving from 1630 <a href="https://en.wikipedia.org/wiki/Matth%C3%A4us_Merian_the_Elder">Matthäus Merian the Elder</a>.</figcaption></figure><p>Choosing this shape was a particularly wise choice. It seems that ancient Greek mathematicians had intuitively known that the circle surrounds the largest area, and in the second century BCE, Zenodorus even <a href="https://archive.org/details/historyofgreekma029268mbp/page/206/mode/2up">provided</a> <a href="https://old.maa.org/press/periodicals/convergence/the-sagacity-of-circles-a-history-of-the-isoperimetric-problem-the-work-of-zenodorus">a partial proof</a>. But the problem remained unsolved for more than two millennia, until 1841 when Jakob Steiner <a href="https://archive.org/details/ThePenguinDictionaryOfCuriousAndInterestingGeometry/page/n139/mode/2up">published</a> <a href="https://old.maa.org/press/periodicals/convergence/the-sagacity-of-circles-a-history-of-the-isoperimetric-problem-the-work-of-jakob-steiner">a rigorous proof</a> that the solution is indeed a circle.</p><blockquote><a href="https://blogs.cornell.edu/info2040/2018/10/17/tian-jis-horse-racing-strategy-an-ancient-chinese-story-based-on-game-theory/"><strong>The horses of Tián Jì</strong></a></blockquote><blockquote>Chinese mathematics has a long tradition. One famous story from the 4th century BCE tells of Tián Jì, who entered a horse riding competition with the king of his region. They competed in three races, using different horses for each race. Whoever won two or more races won the tournament. They both used their three best horses ranked by speed. While Tian Ji had great horses, racing with the king seemed out of his league. The king’s best horse was faster than Tian Ji’s best horse. Likewise, the king’s second and third-best horses were faster than Tián Jì’s respective horses. Tián Jì doubted his prospects. On the road, he met a stranger, Sun Bin, who said he knew how to win, and Tián Jì took him to the competition.</blockquote><blockquote>In the first match, the King entered with his best horse. Tián Jì wanted to respond by sending his own best horse, but, under the guidance of Sun Bin, he chose his third-best. It was a calculated loss of the first match. In the second match, the king put forth his second-best horse. Now, Sun Bin convinced Tián Jì to send his best horse, which resulted in a win for Tián Jì. Finally, in the last match the king’s third-best horse lost again — Tián Jì’s second-best was faster than the king’s third-best.</blockquote><blockquote>The result: Tián Jì won the competition despite his horses being respectively slower than the king’s. Impressed by his reasoning, Tián Jì took Sun Bin as his strategic advisor. This problem is now known as <a href="https://www.sciencedirect.com/book/9780124179103/elementary-linear-programming-with-applications">the assignment problem</a> in optimisation.</blockquote><p>In the 3rd century BCE, another Greek mathematician, <a href="https://en.wikipedia.org/wiki/Euclid"><strong>Euclid</strong></a><strong>, revolutionised geometry</strong>. His celebrated <a href="https://mathcs.clarku.edu/~djoyce/java/elements/elements.html"><em>Elements</em></a> include propositions that resemble optimisation problems, such as one showing that the greatest straight line within a circle is the diameter (<a href="https://mathcs.clarku.edu/~djoyce/java/elements/bookIII/propIII15.html">Book III, Prop. 15</a>) . In <a href="https://mathcs.clarku.edu/~djoyce/java/elements/bookVI/propVI27.html">Book VI, Prop. 27</a>, he discusses maximising the area in a specific arrangement of parallelograms, as well as providing the geometric solution construction method and proof that it does indeed maximise the area.</p><blockquote><strong>Pappus’s honeycomb conjecture</strong></blockquote><blockquote>The Greeks used mathematics to solve problems from daily life and analyse the natural world around them. One example of this is the <a href="https://mathworld.wolfram.com/HoneycombConjecture.html">honeycomb conjecture</a>, which occupied Pappus in the 4th century CE. He investigated honeycombs and noted that by using hexagons, bees manage to use less material while ensuring the largest space within each hole. In other words, <strong>Pappus reckoned that hexagons are the optimal shape for honeycombs.</strong> The honeycomb conjecture <a href="https://arxiv.org/pdf/math/9906042">was proven by Thomas Hales in 1999</a> (<a href="https://link.springer.com/article/10.1007/s004540010071">published in 2001</a>).</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mazBA5E8SuZ8NsjQjzX-9g.jpeg" /><figcaption><em>Honeycombs inspired Pappus in his study of geometry. </em><a href="https://commons.wikimedia.org/wiki/File:Honey_Bees_on_Comb_1.jpg"><em>Photo</em></a><em> in the public domain.</em></figcaption></figure><h3>Modern period (1500–1945)</h3><p>Although the Middle Ages saw interesting mathematical developments in the West, especially with the inclusion of Arabic advancements, it was only in the Modern period that <strong>individual optimisation problems were tackled broadly</strong>. Examples include <a href="https://en.wikipedia.org/wiki/Pierre_de_Fermat">Fermat</a>’s discovery that the derivative of a function vanishes at extrema (calculus of variations) and the ground-breaking work on the <a href="https://mathworld.wolfram.com/BrachistochroneProblem.html">brachistochrone problem</a> by <a href="https://en.wikipedia.org/wiki/Isaac_Newton">Newton</a> and the Bernoulli brothers (<a href="https://en.wikipedia.org/wiki/Jacob_Bernoulli">Jacob</a> and <a href="https://en.wikipedia.org/wiki/Johann_Bernoulli">Johann</a>), which investigated the path of the quickest descent. In 1784, Monge investigated <a href="https://homepages.cwi.nl/~lex/files/histco.pdf">a special instance of the assignment problem</a>, which was mentioned in the history of Tián Jì’s horses above. His work pioneered routing optimisation with the study of minimising transport distances.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Ffd.xuwubk.eu.org%3A443%2Fhttps%2Fwww.youtube.com%2Fembed%2F70u1t9QxOVk%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Ffd.xuwubk.eu.org%3A443%2Fhttps%2Fwww.youtube.com%2Fshorts%2F70u1t9QxOVk&amp;image=https%3A%2F%2Ffd.xuwubk.eu.org%3A443%2Fhttps%2Fi.ytimg.com%2Fvi%2F70u1t9QxOVk%2Fhq2.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="640" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/4dddc74a9f9a0f1d6464d135080bbf3b/href">https://medium.com/media/4dddc74a9f9a0f1d6464d135080bbf3b/href</a></iframe><blockquote><strong>Plateau’s problem and soap bubbles</strong></blockquote><blockquote>An interesting problem of the time was Plateau’s problem: does there exist a minimal surface enclosed by a given boundary? This problem is named after <a href="https://en.wikipedia.org/wiki/Joseph_Plateau">Joseph Plateau</a>, who worked with soap bubbles which naturally form in the minimal surface because of the forces working on the bubbles. A general (<a href="https://mathoverflow.net/questions/52075/solution-of-plateau-problem-for-a-simple-smooth-closed-curve-on-a-riemannian-ma">very complex</a>) solution was found in 1930 independently by <a href="https://www.ams.org/journals/tran/1931-033-01/S0002-9947-1931-1501590-9/S0002-9947-1931-1501590-9.pdf">Jesse Douglas</a> and <a href="https://www.jstor.org/stable/1968237?origin=crossref">Tibor Radó</a>.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/818/1*PKOR_u39tcycPPUpau6c6Q.png" /><figcaption>Soap bubbles naturally form in the minimal surface. <a href="https://commons.wikimedia.org/wiki/File:Bulle_cat%C3%A9no%C3%AFde.png">Photo</a> in the public domain.</figcaption></figure><p>At the time, optimisation as an applied tool was scattered across various fields. However, <strong>some algorithms had an impact in multiple fields</strong>. One such algorithm was Legendre’s <a href="https://web.williams.edu/Mathematics/sjmiller/public_html/BrownClasses/54/handouts/MethodLeastSquares.pdf">least-squares method</a> of 1805, which enabled more precise function fitting. <a href="https://en.wikipedia.org/wiki/Carl_Friedrich_Gauss">Gauss</a> used it to predict celestial orbits, such as where the dwarf planet <a href="https://en.wikipedia.org/wiki/Ceres_(dwarf_planet)">Ceres</a> would reappear after being hidden by the sun’s glare. The algorithm’s impact on maritime navigation was massive.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*m6JLdKaYVjukU56Z_usf-w.png" /><figcaption><em>The least-squares method can be used to find a line that best fits a dataset. </em><a href="https://commons.wikimedia.org/wiki/File:Linear_regression.svg"><em>Figure</em></a><em> in the public domain.</em></figcaption></figure><p>During the 19th century, <strong>optimisation became increasingly central to economics</strong>. Multiple economists simultaneously introduced the fundamental <a href="https://www.britannica.com/money/diminishing-returns">law of diminishing returns</a>, conjecturing an output that optimises profit. The “marginalist revolution” of the 1870s positioned optimisation as a fundamental tool for economic theory. <a href="https://en.wikipedia.org/wiki/Hermann_Heinrich_Gossen">Heinrich Gossen</a>, a pioneer in the field, stated: “Man should organise his life so that his total life pleasure becomes a maximum.”</p><blockquote>At <a href="https://dotsandlines.ai/">Dots &amp; Lines</a>, our scheduling solutions aim to optimise needs across stakeholders. This includes work-life-balance preferences of employees. You might say we optimise “life pleasure” as imagined by Gossen.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DFzsLwJQjwuhfSslKwG2UQ.jpeg" /><figcaption><em>Legendre’s least-squares method had a huge impact on astronomy and maritime navigation. </em><a href="https://www.shutterstock.com/image-photo/night-image-texel-lighthouse-serving-navigation-1982296904"><em>Photo</em></a><em> from Shutterstock.</em></figcaption></figure><p>The complexity involved in generalising optimisation problems and solution methods explains the belated establishment of a dedicated field. The cumbersome process in the extremely impactful field of <a href="https://math.mit.edu/~goemans/18310S15/lpnotes310.pdf">linear programming</a> (LP) highlights this complexity. Not considering specific instances like <a href="https://en.wikipedia.org/wiki/Gaspard_Monge">Monge</a>’s work from 1784, the first-known study concerning LP was done by <a href="https://en.wikipedia.org/wiki/Joseph_Fourier">Fourier</a> back in 1826, whose results were rediscovered in 1936 by Motzkin, 110 years later!</p><p>A turning point for optimisation came in 1947, when <a href="https://en.wikipedia.org/wiki/George_Dantzig">George Dantzig</a> presented his <a href="https://dl.acm.org/doi/pdf/10.1145/87252.88081"><strong>simplex algorithm</strong></a>. However, <a href="https://en.wikipedia.org/wiki/Leonid_Kantorovich">Leonid Kantorovich</a> and <a href="https://en.wikipedia.org/wiki/Tjalling_Koopmans">Tjalling Koopmans</a> had worked independently on the practical applications of LP during the war. In 1941, <a href="https://en.wikipedia.org/wiki/Frank_Lauren_Hitchcock">Frank Lauren Hitchcock</a> even presented a solution to an LP application in transport that was similar to Dantzig’s simplex algorithm. Dantzig had been unaware of his predecessors’ achievements, and wrote: <em>“</em><a href="https://link.springer.com/10.1007%2F978-0-387-74759-0_268"><em>What seems to characterise the pre-1947 era was lack of any interest in trying to optimise.</em></a><em>”</em></p><blockquote><strong>Ideological friction</strong></blockquote><blockquote>Leonid Kantorovich (1912–1986) was an extraordinary researcher from communist Russia. He became a full professor at the age of 22. During the Second World War, he helped with logistics on the Road of Life, which supplied Leningrad during the siege by the Nazis. There, <strong>he calculated the minimum distance between cars to safely pass on the ice, accounting for air temperature and ice thickness</strong>. In 1939 he’d already produced extensive results on LP. Even though his work showed great promise, he postponed his research because he noted “it was too dangerous to continue”: the Soviet authorities considered his optimisation in economics to be counter to orthodox Marxist principles. Nevertheless, in 1975, he was awarded the Nobel prize in economics. Justice was slow but delivered in the end!</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*Gaa4JFtgYoK37B_l" /><figcaption><em>During the Second World War, Kantorovich played a pivotal role in the Road of Life logistics during the Nazi siege of Leningrad. </em><a href="https://commons.wikimedia.org/w/index.php?curid=57278201"><em>Photo</em></a><em> in the public domain.</em></figcaption></figure><h3>Contemporary times (1945–present)</h3><p>The war significantly increased interest in optimisation. It had shown its worth when applied to vital wartime processes, such as transport, resource allocation and complex scheduling. This led to a post-war boom in optimisation. Dantzig’s <a href="https://archive.org/details/in.ernet.dli.2015.39951/page/n357/mode/2up">simplex algorithm</a> emerged as a revolutionary tool with wide-ranging applications; the new field of operations research was established, relying heavily on optimisation techniques; the first international optimisation conference was organised in the US; and the Cold War’s arms and space races, together with the ideological war, further accelerated the emerging interest in optimisation research.</p><p>In the late 20th century, <a href="https://en.wikipedia.org/wiki/Moore%27s_law">exponential increases</a> in computing power fuelled significant progress in AI research and global optimisation strategies. Recent advances are too numerous to count. The ones we like to highlight are heuristics and the field of evolutionary algorithms, as they are the basis for our work at Dots &amp; Lines. <strong>This is where our path begins</strong>.</p><p><a href="https://books.google.pt/books?hl=nl&amp;lr=&amp;id=rHQf_2Dx2ucC&amp;oi=fnd&amp;pg=PR8&amp;ots=xOBg6pxhzH&amp;sig=1P6QER5n6RZU5ZXzVTbvzez0TT8&amp;redir_esc=y#v=onepage&amp;q&amp;f=false">Evolutionary algorithms</a> (EAs) were popularised by John Holland in 1975. They are <strong>nature-inspired algorithms</strong>, simulating how nature creates well-adapted individuals through evolutionary processes. EAs produce innovative solutions to previously intractable problems like routing and scheduling problems.</p><p>The EA process can be summarised as follows: a population of potential solutions evolves through selection, crossover, and mutation — inspired by evolution in nature. Each generation builds on the previous one, with the most promising individuals selected to give birth to new individuals for the next generation. This results in <strong>an iterative process that produces increasingly refined populations</strong> and facilitates the search for optimal solutions to complex computational problems.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HfnCLUUGKME7y-K9yo6pyg.jpeg" /><figcaption>Tigers are resilient, majestic and supreme hunters, shaped by millions of years of evolution. <a href="https://www.shutterstock.com/image-photo/tiger-cold-winter-taiga-russia-snow-1767222674">Photo</a> from Shutterstock.</figcaption></figure><p>Nowadays, <strong>optimisation appears in almost all areas of society and daily life</strong>: Google Maps shows us the quickest route to our office, machines do our laundry much quicker than we can, and car engines, computer chips, and smartphones continuously evolve. Companies optimise every detail of their business processes to maintain a competitive advantage. These detailed optimisations have produced remarkable optimisation cases: <a href="https://edition.cnn.com/2017/02/16/world/ups-trucks-no-left-turns/index.html">UPS innovated vehicle routing by favouring right turns</a>, Formula 1 pit stops have been compressed <a href="https://lemonly.com/work/formula-1-pit-stops-over-time">from one minute to just two seconds</a>, and data centres now automatically adjust the temperature of their cooling systems to <a href="https://www.sciencedirect.com/science/article/pii/S2666123324000916">optimise electricity savings</a>.</p><blockquote><strong>UPS: the art of turning right</strong></blockquote><blockquote>In the 1970s, UPS, the largest US logistics company, started avoiding left turns. Because left turns go against the flow of traffic (in countries where vehicles drive on the right), they found that left turns are more dangerous, and cost more in time and fuel. Back then, they created routes as right-turning loops. Now, they have software to calculate routes, ensuring that over 90% of turns are to the right. UPS <a href="https://web.archive.org/web/20170216131327/https://www.pressroom.ups.com/pressroom/ContentDetailsViewer.page?ConceptType=PressReleases&amp;id=1426329559785-791">estimates</a> that this can save them 10 million gallons of fuel a year.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/799/0*6qEclqRclcaKErsx" /><figcaption><em>In right-driving countries, turning right is less dangerous, time-consuming</em>, and fuel-expensive than turning left. <a href="https://www.flickr.com/photos/88876166@N00/15470407660">Photo</a> by <a href="https://www.flickr.com/photos/tabor-roeder/">Phil Roeder</a>, licensed<em> under the </em><a href="https://creativecommons.org/licenses/by/2.0/deed.en"><em>Attribution 2.0 Generic</em></a><em> licence.</em></figcaption></figure><p>The old records of optimisation date back to ancient Greece and China, and describe tentative, scattered sparks of something that is now an essential area of mathematics and computer science. Throughout history, <strong>optimisation has improved the livelihood of human beings</strong>. Whether it is a PhD thesis, a faster algorithm, or a solution to a practical problem, we feel proud to contribute to an area with such a long and rich history.</p><p>At <a href="https://dotsandlines.ai/">Dots &amp; Lines</a>, we specialise in using optimisation for scheduling and logistics. The goal of optimisation, and science in general, should always be to improve human lives and the planet. Although development was originally driven by war and financial gain, optimisation solutions should always be ethically and safely implemented. <strong>Our mission is to be part of optimisation’s odyssey and influence its path towards responsibility and efficiency.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/516/1*mDgYSr-rv8mH4CSjCqL4sQ.png" /><figcaption><em>The traditional ending of a mathematical proof, Q.E.D. (quod erat demonstrandum</em>, Latin for “that which was to be demonstrated”<em>), was translated from the Greek ὅπερ ἔδει δεῖξαι (ΟΕΔ). Euclid, the renowned Greek geometer from the beginning of this article, used these words extensively in his work.</em></figcaption></figure><h3>Written by</h3><p><a href="https://www.linkedin.com/in/leon-ingelse/">Leon Ingelse</a> (<a href="mailto:leon@dotsandlines.ai">leon@dotsandlines.ai</a>), writer and researcher at <a href="https://dotsandlines.ai">Dots &amp; Lines</a> <br><a href="https://www.linkedin.com/in/goranmolnar/">Goran Molnar</a> (<a href="mailto:goran@dotsandlines.ai">goran@dotsandlines.ai</a>), CEO at <a href="https://dotsandlines.ai">Dots &amp; Lines</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7bd497f1513f" width="1" height="1" alt=""><hr><p><a href="https://blog.dotsandlines.ai/a-history-of-optimisation-from-ancient-greece-to-ai-superpowered-scheduling-7bd497f1513f">A history of optimisation — from ancient Greece to AI-superpowered scheduling</a> was originally published in <a href="https://blog.dotsandlines.ai">Dots &amp; Lines</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>