-
For Part 1 (3 Points): How should researchers and developers prioritize between pushing the boundaries of technology and addressing practical user needs when developing knowledge graph-powered applications? Include the potential benefits and drawbacks of focusing primarily on technological innovation versus user-centered design in your answer.
-
For Part 2 (2 Points): After submitting your initial response, select a peerās response. Do you agree or disagree with your peerās prioritization between technological innovation and user-centered design? Why? Your answer should be respectful, constructive, and include evidence supporting your perspective.
Considering the focus on advancements by Kejriwal and Orr and the emphasis on user-centered design and commonsense reasoning by Diefenbach,
The interesting part of this question, is to first establish a definition of what is ātechnological innovtionā and what is āuser-centered designā as these do not have to be mutually exclusive.
User-centered design focuses on the needs of the user at every stage of the design process. Sometimes, technological innovation requires that users learn how to think and use new tools in novel ways that they had never thought of before.
When Adobe created Photoshop for example, they had to stage training sessions across the globe to train graphic artists in how to use the functions such as layering, masking as well as algorithmic transformations such as solarizing, rasterizing, etc. Many companies such as Adobe are constantly straddling the line between making tools that are usable, versus adding new features. They also had to invest a lot in training users in how to use their product.
Social media is an example of another web innovation that required users to learn to use the new tool including learning the new etiquette of the web.
While knowledge graphs are not specifically mentioned in this question, it would seem the implication of the question is that knowledge graph technology is a ātechnological innovationā. Where knowledge graphs intersect with users is a constant challenge for the creators of tools for people to use to build knowledge graphs as well as how to visualize, search and use graphs.
When a knowledge graph supports the architecture of a system, itās possible that a user-centered design wonāt even require a user to know they are interacting with a graph, but they can trace results given by graphs because of the nature of how graphs connect data.
User-centered interfaces such as GPT chat interfaces seem highly user-centered and are quite technologicially innovative, however they cannot always be trusted to provide useful answers that havenāt just been hallucinated by the system. This is a tricky problem as users can come to trust knowledge from such a system, but the users often have no way of checking the output. For this reason, many people are now looking at ways to integrate knowledge graphs with LLM/GPT technology so that users can benefit from the simplicity of a chat interface while also trusting and tracing the responses given by a system back to some ground truth.
I designed ImageSnippets with developers to construct an interface for users to build knowledge graphs with no code. The interface design has taken the user into consideration throughout the entire process, but it also redefines the notion of what a digital asset manager can be that has itās fundamental core, an architecture built around a knowledge graph.
Conceptually, the challenge is in not just teaching users how to think about knowledge graphs - but especially how to think about and use knowledge graphs that form a metadata centric asset management system/metadata management system.
User cent
Human-centered factors are always needed to be considered seriously in technology. Researchers , developers, and users and technology, how to balance them is a good topic. Traditional researchers and developers may fall into the misunderstanding of only pursuing technological innovation and ignoring users.
The benefit of focusing on technology is that technology is always the key point to stay competitive on the market. However, ignoring users can limit the impact of the state-of-the-art technology.
The benefit of focusing on users ensures that the application solves tangible problems. However, overemphasis on user needs might limit exploration into more advanced technology.
It is quite important to balance them and clarify the needs.
I totally agree with you that technological innovation and user-centered design do not have to be mutually exclusive and that the balance between the two is crucial. Balancing the two is less about prioritizing one over the other and more about contextually adapting the approach to meet the needs of the system and its users.
User-Centered Design (Focusing on Practical User Needs)
Focus: User-centered design prioritizes usability, accessibility, and real-world problems. The goal is to make the technology accessible to users, ensuring that the application meets the specific needs of its target audience, whether they are casual users, domain experts, or organizations.
Benefits:
- Enhanced Usability: Applications designed with the user in mind are typically more intuitive and easier to adapt, leading to higher user satisfaction and engagement. For instance, Wikibase provides an accessible platform for creating and managing knowledge graphs without requiring users to be data scientists.
- Practical Relevance: By focusing on real-world problems, user-centered design ensures that the technology is directly applicable to the usersā goals, making the tool more effective and valuable.
- Increased Adoption: User-centered applications are more likely to be adopted by a wider audience because they are built with the usersā workflows, preferences, and challenges in mind.
Drawbacks:
- Limited Innovation: Focusing too much on user needs may limit the scope for innovation, as developers may avoid introducing new technologies or features that could seem too complex or unnecessary for the users.
- Short-Term Focus: User-centered design may prioritize immediate user needs at the expense of long-term scalability or functionality. This could limit the applicationās ability to evolve as technologies advance.
- Over-Simplification: In the effort to create an accessible user experience, critical complexities of the underlying technology may be oversimplified, potentially reducing the overall power of the application.
Balancing Both Approaches
Kejriwal and Orr emphasize the need for technological advancements in knowledge graph applications to push the boundaries of what is possible. However, as Diefenbach notes, technological innovation must be balanced with a focus on user-centered design, ensuring that the technology does not become so advanced that it is detached from practical needs.
- A hybrid approach works best: Researchers and developers should ensure that cutting-edge technologies are designed with usability in mind, maintaining a focus on real-world user problems.
- Iterative Design and Feedback Loops can help ensure that technological advancements do not overtake the needs of users. By continuously gathering feedback, developers can refine their applications to be both innovative and usable.
In the development of knowledge graph-powered applications, researchers and developers should prioritize user needs without sacrificing the potential for technological advancement. While pushing the boundaries of technology can lead to innovative and powerful solutions, it is essential to ensure that these solutions are accessible, understandable, and relevant to the users. By striking a balance between technological progress and user-centered design, developers can create knowledge graph applications that are both groundbreaking and practical.
You raise an excellent point about the difficulty non-expert users face in distinguishing accurate responses from inaccurate ones in GPT chat interfaces.Can you expand on how to integrate KG with the LLM/GPT to solve this issue? Use LLM to parse the KG database? But there could still be a possibility of LLM not accurately identifying userās intent?
Your mention of ImageSnippets sounds very interesting. I am quite intrigued to see how this works? Could you provide any useful reference related to this subject if you have any? Thanks
Technological innovation is a luxury; user-centered design is fundamental. Unless and until the reverse is true.
My assertion sounds silly or perverse, but only in academia can innovation take the front seat before users, and to establish a business, users must come first. The exception that proves the rule is that when taking payment to perform innovation for others, the others are the users, and their needs come first.
Some situations support both, but one will override the other depending on the setting of academic vs business.
@mmw Users do not need to know they are interacting with a graph - excellent point! - and I agree. OTOH consider the following scenario where exposing the KG in UI is an essential feature:
I worked on a project at JPMC where we created an "auditorās workbenchā designed to fulfill the critical needs of bank staff (auditors) who are responsible for ensuring the bankās compliance with regulations such as KYC āknow your customer āand AML āanti-money launderingā. We built Knowledge Graphs from large volumes of data that would otherwise require visual inspection of vast numbers of reports. In the graphs, nodes have property associations and links to the regulations, internal processes, and, of course, the transactions that they were inspecting.
Here is an everyday use case of AML that also relied on KYC data:
What if a politician or executive has accounts that show no evidence of money laundering? Should the auditor stop looking? They routinely inspect the accounts of close relatives and associates. For the auditorās workbench, we made it easy in the UI to see the graph of associations among people, parties, transactions, related amounts of money, etc.
In this case, the user-centered design sprints revealed that auditors were very comfortable working directly with the graph in an āinfinite workspaceā because this gave them a viewpoint they had always wanted and an easy method of navigation they craved. The UI we provided looked like what we now see in Neo4j Bloom and AllegroGraph.
Regarding my earlier comment suggesting that technological innovation and user-centered design are fundamentally in conflict, this was a case where we attempted to find a balance by innovating rapidly and implementing user requirements. Itās worth noting that this project failed because technological innovation didnāt happen fast enoughāwe couldnāt find a performant quad storeāand the vision that we showed the users was never fulfilled.
Iām curious to learn about specific situations where a hybrid of technical innovation and user-centered design worked for the best? Perhaps this refers to behemoths like Apple or artisanal software cooperatives, but these are rare.
1 Like
Sorry if it wasnāt clear, my comment is a response to an earlier post.
@ankuku2002 Iām engaging with the authors you mention, not to critique your answer. Iām not convinced that balance is possible between such orthogonal or opposed interests.
Aside: The original question seems odd to me, so digging into it has been difficult. Thanks to you and others here for getting more out of it than I did.
Hi @ankuku2002 ā Thanks for asking about ImageSnippets. The link is ImageSnippets is here and I have lots of material about it! Essentially the utility created for users with ImageSnippets, is that the users can parse descriptions of visual content into a graph. Each image can be viewed in a ātriple-editorā window and then all of the metadata about that image can be organized and stored in a triple store. I would be happy to provide a lot more information if you are interested!
As for your other interesting question. Expanding on how to integrate LLM/GPT with KGās is the million dollar question isnāt it?
From my experience, I have done things like: submitting turtle syntax (which is a serialization/a syntax for text representations from an RDF graph) to ChatGPT and had conversations with GPT about the contents of the data I provided to it.
I have also given ChatGPT and notebook LM various types of resources (documents, images, etc) and asked it to create knowledge graph structures from these resources and had some mixed results.
In general, the goal is to have the benefits of the natural language processing that GPT systems offer, but mixed with the reliability of the information that can be inferred from a knowledge graph, with a big benefit that results from a knowledge graph can be explained and verified.
The two systems work with very different strategies, however, and I canāt see - right now, how any GPT system can be predictably used into the future without humans examining output carefully regardless of whether KGās are being used to derive an output OR GPTās are crafting a KG from source material.
I dislike the phrase āhumans in the loopā, because I think it should be something more like machines in a human loop, but either way, I am concerned that there are not enough discerning, trained, data literate humans to examine whether or not interactions with these systems is correct or appropriate.
HI @kmc - I am fascinated by the bank use case and of course I totally agree that sometimes seeing an actual graph can be a highly useful feature that could be considered part of a good āuser-centered designā.
There is also the question of how the graph should be visualized. Many people became accustomed to design choices that were made in the āearly daysā with relational databases with forms, reports, tables, etc.
Anecdotally, I heard from a UX designer the other day that when users think about displaying their data, they think the only way to do it, is with heat maps, but then they only like the heat maps when it is their own data and donāt like them if it is someone elseās data. I guess the lesson in that would be that users have a hard time visualizing data in general unless it it data they themselves are very familiar with.
Funny - after I typed this response, @kmc - I also just went to LinkedIn with this clip discussing exactly this topic. Paradigms in user design have not significantly changed in 60 years until now: https://www.linkedin.com/posts/jakobnielsenphd_ux-ai-interactiondesign-ugcPost-7267302856028745728-s6bU?utm_source=share&utm_medium=member_desktop
I think that pushing the boundaries within data privacy safety limits is better than directly addressing user needs. I think that open source technology allows for users to directly create and reproduce work that fulfills a need. Meanwhile innovative technology creates a (albeit sometimes imperfect) path towards something new. Apple itself is on 18 major and several sub software updates for the iphone. As a point in favor of innovation, often improved and innovative versions of that work makes past iterations obsolete, which is why I would want future work to be encouraged. I think that like I mentioned, I say this within the caveat of also implementing data privacy standards like what Diefenbach talks about with the bots for wikidata It is important to have several rounds of communities checking on the work that bots are doing and setting up checkmarks to monitor what is allowable and keeping a record of them.
On the flipside, I think that a drawback of focusing on technological innovation could inspire a lack of contributions if there is insufficient training, if technology advancements do not keep pace with taxonomy and ontology work (ex. If further edits to a platform go beyond what the language translation service is). A benefit however, is that for a cutting edge and exciting piece of technology, you benefit from queryable data that could have innumerable functions and applications for research.
Apple was what I had in mind when crafting my response. I didnāt catch you had replied with this thought similarly. I think that privacy and safety standards are good practices for the āauditorās workbenchā that you reference. I agree with your point that there is a balance because if things do not move at a fast enough pace, especially in software, they do become obsolete. Going back to Apple as my frame, they have been able to maintain its standing by first putting out innovative designs and then maintaining software and a UI that encourages those products. However, especially recently they have been criticized for not being innovative enough and just say, adding better cameras to the iphone and thatās it. However, everyday people still flock to their products because they were innovative to start. This article shows that iPhone usage has grown by 76.13% since 2014. This makes me wonder what this settling period looks like for technical innovation as part of the balance, where user experience can come after, and how long it takes to provide consistent use if it does.
Benefits/Drawbacks of Prioritizing Tech Innovation OVER User-Centered Design:
Benefits:
-
Advancing the Field: Pushing technological boundaries can lead to breakthroughs that redefine what is possible, opening up new opportunities for future applications. (i.e. Orr talks about the sophistication of using the BERT model (an LLM) for context to properly identify entities based on context and relationships which in turn can handle those many confusing entity cases that exist over the ātailā ā which changes the KG game)
-
Competitive Edge: Cutting-edge innovations can differentiate a product in the market, attracting early adopters and tech-savvy users. We live in an āinnovative raceā culture to be the āfirstā to innovate to get the biggest market share and thus drive revenue. If there is a new AI model that can automate KG creation based on specific data/schema and you are the first to do itā¦ you will reap the benefits sooner or before the money pot dries out
-
Long-Term Vision: By focusing on innovation, developers can anticipate future needs and create solutions that remain relevant as technology and user expectations evolve. Being able to anticipate technology use cases, biases, and societal impact becomes clearer as innovation happens; and in turn we can run experiments early and test the reliability/harmful nature of the innovation
Drawbacks:
-
Resource Inefficiency: Investing heavily in cutting-edge technology without considering user needs can waste resources on features or capabilities that users may not value or use. This is why POCs launched to a small set of beta users can identify if the innovative product is worth more resources.
-
Slower Adoption Rates: Applications that prioritize innovation over usability may face resistance from users, as they may find the tools intimidating or irrelevant to their workflows. Every company wants users (whether it is internal or external users) to buy-into the innovation so they make money or increase productivity.
-
Risk of Obsolescence: If the focus is purely on innovation, the application may become outdated quickly as user needs evolve, leaving it unable to adapt effectively. By the time āinnovation 1ā is launched āinnovation 2ā is launching tomorrow
I made a similar post about being innovative quickly is our current culture and if one doesnāt keep up they may sink. However, rushing tools to market that havenāt been FULLY tested for biases/harmful impact, has no plan to take in the user feedback to further train the model/technology, or has not considered user security/privacy SHOULD not be launched. The userās security/privacy should be considered in every tool being developed.
In my perspective, when developing KG powered applications, we should find a balance between both technological innovation and addressing practical user needs. The prioritization depends on the context, but a combined approach often yields the best results.
In my own experience, particularly during the early stages of my career, I worked as part of a small team on building a graph analytics product. We consumed CI/CD tool data to generate KPIs and metrics for analyzing product health, team productivity, and more. In this case, almost 99% of the work focused on data consumption and modeling, while the dashboards were built using the existing tool, Grafana. By leveraging a mature open-source tool like Grafana, we addressed immediate user needs while simultaneously delivering technological advancements, such as focusing on predictive analytics and implementing a recommendation engine. This balanced approach led to the product securing 15+ clients, including Fortune 500 companies. What Iām trying to convey is that finding a balance between technological innovation and user-centered design is crucial for success.
Instead of distinguishing accurate from inaccurate responses, we can enrich the LLM responses by contextualizing them through a GraphRAG pipeline integrated with the KG. There have been significant advancements in GraphRAG, text2SQL, and text2Cypher technologies. In line with @mmw 's point on NLP, this KG should be a combination of chunked graphs, Lexical (NLP entity-relationship), and Domain Graphs (involving SME inputs), which ensures that the generated knowledge is both accurate and contextually relevant.
Hereās a polished version of your message with some improvements for clarity and flow:
Thank you, Margaret. You provided some excellent insights, and I agree that integrating Knowledge Graphs (KG) with Large Language Models (LLMs) can be incredibly valuable when applied appropriately. It has the potential to be worth millions if executed well.
Iām curious about the use case for imageSnippet. Itās intriguing that you store metadata for images as RDF triples. Have you found that this approach offers greater flexibility or a more systematic framework for managing and leveraging the metadata?
@agreen Thanks for engaging with my description of a graph-centered āauditorās workbenchā. I agree with you 100% about Apple as an early innovator - usually second, usually best in breed - and their recent retreat from pure innovation. As an Apple fan, I hope they will be on the ascent again. Apple Intelligence is their foothold in AI, and if history shows anything, they will lead by following. Samsungās recent announcement of a phone with built-in knowledge graphs (using an established technology, RDFox) attempts to challenge Appleās dominance in personal devices. Apple seems to be investing in TV, home automation, and product niches where personal KGs might benefit. Iām not counting them (Apple) out in any market they choose to enter.